Neuropsychoanalysis

An Interdisciplinary Journal for Psychoanalysis and the Neurosciences

ISSN: (Print) (Online) Journal homepage: https://www.tandfonline.com/loi/rnpa20

New project for a scientific psychology: General

scheme

Mark Solms

To cite this article: Mark Solms (2020): New project for a scientific psychology: General scheme,
Neuropsychoanalysis, DOI: 10.1080/15294145.2020.1833361

To link to this article: https://doi.org/10.1080/15294145.2020.1833361

Published online: 24 Oct 2020.

Submit your article to this journal

View related articles

View Crossmark data

Full Terms & Conditions of access and use can be found at

https://www.tandfonline.com/action/journalInformation?journalCode=rnpa20
NEUROPSYCHOANALYSIS
https://doi.org/10.1080/15294145.2020.1833361

New project for a scientific psychology: General scheme

Mark Solms
Neuroscience Institute & Psychology Dept., University of Cape Town, Cape Town, South Africa

ABSTRACT ARTICLE HISTORY

This is a revision of Freud’s “Project for a Scientific Psychology: General Scheme.” It updates the Received 25 June 2020
original, sentence for sentence where possible, in light of contemporary neuroscientific Accepted 4 October 2020
knowledge. The principle revisions are as follows. (1) Freud’s conception of “quantity” (the
KEYWORDS
precursor of “drive energy”) is replaced by the concept of “free energy.” This is the energy Free energy principle;
within a system that is not currently performing useful work. (2) Shannon’s conception of predictive coding;
“information” is introduced, where information is equivalent to unpredictability, and is formally information; consciousness;
equivalent to “entropy” in physics. (3) In biology, the fundamental purpose of “homeostasis” is reconsolidation
to resist entropy – i.e., to increase predictability. Homeostasis turns out to be the underlying
mechanism of what Freud called the “principle of neuronal inertia.” (4) Freud’s conception of
“contact barriers” (the physical vehicles of memory) is linked with the modern concepts of
consolidation/reconsolidation, whereby more deeply consolidated predictions are less plastic
(more resistant to change) than freshly consolidated ones. (5) Freud’s notion of sensory
“excitation” is replaced with the concept of “prediction error,” where only that portion of
sensory input which is not explained by outgoing predictions is propagated inwards for
cognitive processing. (6) Freud’s conception of “bound” (inhibited) cathexis, the main vehicle of
his “secondary process” and voluntary action is equated with the buffering function of “working
memory”; and “freely mobile” cathexis (the vehicle of Freud’s “primary process”) is equated with
the automatized response modes of the nondeclarative memory systems. (7) Freud’s notion of ω
(the system “consciousness”) is replaced by the concept of “precision” modulation, also known
as “arousal” and “postsynaptic gain.”

Author’s introduction There is a risk that enthusiasm may lead to a distortion of

Freud’s use of terms and may read into his sometimes
Freud’s “Project for a Scientific Psychology” is the obscure remarks modern interpretations that they will
Rosetta Stone of neuropsychoanalysis. It was the foun- not bear. And after all we must remember that Freud
dational text for the whole of what became known as himself ultimately threw over the whole neurological
framework.
metapsychology – Freud’s basic theory of the functional
dynamics of the mind – and a sophisticated model of the Strachey concluded: “The Project must remain a torso,
neural mechanisms by which those dynamics come disavowed by its creator” (Standard Edition, 1, pp. 290,
about. In other words, the “Project” (as it came to be 293). Accordingly, an attempt by the great neuroscien-
abbreviated) was the first attempt – Freud’s own tist Karl Pribram to update the “Project” led to an embar-
attempt – to achieve what the whole interdisciplinary rassing falling out with his co-author, Merton Gill, a
endeavor called neuropsychoanalysis is trying to leading psychoanalyst of the day. In the closing sen-
achieve today. tences of their joint book, they wrote:
It is perhaps not surprising therefore that when the
Where we differ is that Gill feels that psychoanalysis must
“Project” was first published (posthumously, in 1950) it
go its own way and that means purging it of its natural
caused an uproar. It enjoyed highly ambivalent science metapsychology, while Pribram welcomes psy-
responses, not dissimilar to those that have greeted choanalysis back into the natural sciences. Pribram
our own efforts today in neuropsychoanalysis. For doubts that the differing views of the two authors are
example, James Strachey, Freud’s authorized translator, really, in the long run, incompatible, while Gill finds
while pointing out that “the Project, or rather its invisible them irreconcilable. (Pribram & Gill, 1976, p. 169)
ghost, haunts the whole series of Freud’s theoretical When I first came upon the “Project,” while still a
writings to the very end,” hastened to add: student, it was not my Neuropsychology professors

CONTACT Mark Solms [email protected]

Supplemental data for this article can be accessed here https://doi.org/10.1080/15294145.2020.1833361
© 2020 International Neuropsychoanalysis Society
2 M. SOLMS

who introduced me to it but rather a lecturer in the the revised text to facilitate this comparison (see Sup-
department of Comparative Literature, Jean-Pierre de plemental data). Please also note, the paraphrasing of
la Porte (who has kindly contributed a commentary, Freud’s “Project” requires this paper to be written in the
below). I was mesmerized; it seemed to address all the first person singular, as the original was; so it is important
questions that my training in neuropsychology neg- to acknowledge at the outset that its main ideas are based
lected, the very questions that had drawn me to the on joint theoretical work between the author and Karl
field in the first place. As Oliver Sacks famously Friston (Solms, 2019; Solms & Friston, 2018).
remarked: “Neuropsychology is admirable, but it A fresh and fully explicated account of the new model,
excludes the psyche – it excludes the experiencing, in its own terms, requires a book-length treatment. This
active, living ‘I’” (Sacks, 1984, p. 164). treatment will be published shortly, under the title The
As soon as I found the opportunity, therefore, one Hidden Spring (Solms, 2021), written in a language and
long weekend in 1984, I sat down with the “Project” style that should be comprehensible to non-specialist
on one side of my desk and my trusted neuropsychology readers. My profound hope is that the model described
textbook (Luria, 1980) on the other, and then I tried in here (and there, in expanded form) will provide our
vain to translate Freud’s opaque terms and concepts field with a metaneuropsychology which can serve the
into their presumptive contemporary equivalents. For purposes for the foreseeable future that the “Project”
three days I did nothing else. I was completely spell- served in the past. In saying this, I want to make explicit
bound, but ultimately frustrated; put simply, too much what I just alluded to, namely that this model rests
was unknown. I concluded that what was required was upon the work of giants, not only Freud, but also Pank-
not a theoretical exercise but rather a comprehensive sepp and Friston. Of course, there were many others
program of interdisciplinary, experimental and clinical besides them. In fact, it could be said that the true foun-
research. I have spent the rest of my working life trying dations for this model were laid by Hermann von Helm-
to get that research program off the ground. holtz, one of the founding fathers of the Berlin Physical
My great good fortune in the intervening decades Society (sometimes called the Helmholtz school of medi-
between then and now was to have opportunities to cine), whose work so obviously undergirded the original
work with and thereby closely learn from two outstanding “Project” and all that Freud built upon it.1
scientific pioneers of our time, first, the affective neuros- One of the editors asked me to write this Introduction
cientist Jaak Panksepp, and second, the computational to encourage daunted readers to persevere with the
neuroscientist Karl Friston. The integrative achievements challenging task of understanding my paper. I find that
of these two great minds, far more than my own clinico- I cannot do any better than she did:
anatomical research efforts, brought me to the present
In my view, the value of the piece (and the full treatment
juncture, where I believe the moment has come for us to you are elaborating elsewhere) is the articulation of an
attempt what Merton Gill declared impossible and unde- infrastructure of a truly neuropsychoanalytic model of
sirable when I was still a student: to update Freud’s the mind and brain. This model allows for theoreticians,
“Project” in such a way that it can once again perform its clinicians, and researchers to account for the complex
historic role as the Rosetta Stone of our field. interactions between cognition and emotion; impulse
and regulation; consciousness and unconscious pro-
It is important to clarify here: this is literally a revision of
cesses; genetics and experience; etc. All of this is meaning-
Freud’s classic paper; it is not an original article in the usual ful to clinicians who work with these dynamics in everyday
sense of the word. Some of Freud’s concepts are replaced work. I think if you can articulate that, it may encourage a
outright but most are found to still be viable, and in sur- few more readers to wade into the dense piece – it really
prisingly many places the wording of the original text is opens up after a few pages, and I think any readers who
followed exactly. I thought it appropriate to update it in can get into the middle and end sections will find it thril-
ling. (Maggie Zellner, personal communication)
this literal way so that scholars in psychoanalysis and
neuroscience can see precisely how the model that I am So, take a deep breath; here we go.
now proposing builds upon, and differs from, Freud’s
Ur-text. This approach unfortunately has the necessary
KEY TO ABBREVIATIONS IN THE NEW PROJECT 2
consequence that my paper is very dense and obscure Q = External states (unknowable in themselves)
in places, since the whole point of it is to track the original Qη = Generative model of external states
φ = Sensory states
(dense and obscure) “Project” as closely as possible; to M = Active states
update it, word for word, sentence for sentence and para- ψ = Predictions (based on Qη)
e = Errors (based on φ and its prediction ψ)
graph for paragraph. Readers will find it useful, therefore, ω = Precisions
to compare my revision with Freud’s original. To this end, F = Free energy (based on e and ω)
the editors have kindly provided a “marked-up” version of
 NEUROPSYCHOANALYSIS 3

New project for a scientific psychology from here, we can in the very beginning of life concep-
tualize the organism, which is but an inheritor of the
Introduction
conservative imperative of all self-organizing systems,
The intention is to attempt, once more, to furnish a psy- as cloaking these systems in a Markov blanket. Markov
chology that shall be a natural science; that is, to rep- blankets induce a partitioning of states into internal
resent mental processes as quantitatively determinate (system) and external (not-system) ones, so that the
states of specifiable physical elements, thus making system is insulated from the entropic forces that sur-
those processes perspicuous and free from contradic- round it, to resist those forces. External states (Q) can
tion. Two principle ideas are involved: (1) What dis- only influence the internal states of a self-organizing
tinguishes activity from rest is to be regarded as F,3 system vicariously as states of its blanket. The blanket
subject to the general laws of information.4 (2) is itself partitioned into “sensory” and “active” states
Neurons are to be taken as the physical elements. (these are its homeostatic receptors and effectors,
Neurons and information processing – similar models embodied in the sensory and motor systems of
are now commonplace.5 neurons: φ and M). This yields a circular form of causality:
external states can influence the internal states of a self-
organizing system via the sensory states of its blanket,
(1) First principle theorem while the internal states of the system couple back to
the external world through its active states. Crucially,
The quantitative conception
the sensory states feed back the external consequences
The theorem “what distinguishes activity from rest is to of prior active states, and thereby adjust the posterior9
be regarded as F” is derived from statistical physics but it states of the system via the homeostatic control
manifests in clinical observation, especially where exces- centers that generate predicted consequences. This is
sively intense feelings are concerned – as we shall see, learning. A complex self-organizing system is equipped
the quantitative characteristic of neural functioning with a meta-control center which controls its total “gen-
emerges more plainly with affects than with cognition. erative model” of the causes of its external states (Qη)
Processes such as stimulus, substitution, conversion and registers “prediction errors” in general. This predic-
and discharge, which were introduced into metapsy- tive model, based in instincts supplemented by learning,
chology to explain the dynamics of affect, suggested serves the singular aim and purpose of maintaining
the existence of an underlying mental energy (called the viable bounds (the ongoing existence) of the
“drive”) as a quantity in a state of flow.6 It seemed legit- system, through the minimization of free energy,
imate to attempt to generalize what was initially recog- where F is a measure of the average difference between
nized clinically. Starting from this consideration, it is now the system’s predicted and actual sensory states over a
possible to lay down a basic principle of neuronal given period of time (i.e. it is a measure of the demand
activity in relation to F, which promises to be highly for work by the system). Minimizing F represents the
enlightening, since it appears to comprise its entire func- primary function of the nervous system. Here is room
tion.7 This is the free energy principle: neuronal systems for the development of a secondary function. For
tend to minimize F. On this basis the structure and devel- among the paths for minimizing F those are preferred
opment as well as the functions of the nervous system, and retained which involve a cessation of prediction
as they relate to mental life, can be understood. errors: minimization of surprise. Here in general there is
In the first place, homeostasis underwrites a “principle a proportion between the free energy derived from sur-
of neuronal inertia.”8 This is a conservative tendency prisal and the work necessary for its minimization, so
which maintains the organism within its phenotypically that the free energy principle is not upset by this.
viable bounds, across many biological parameters. The The insulation of biological systems from surrounding
design of the nervous system clearly serves this prin- entropic forces is, however, broken through from the
ciple. The structural dichotomy of neurons into sensory first owing to another circumstance. With an increasing
(homeostatic “receptor”) and motor (homeostatic complexity of the interior of the organism, the nervous
“effector”) types is a contrivance for minimizing F. system receives entropic perturbations from the
Sensory demands for work (increasing F) generate somatic element itself – endogenous sources of F –
motor actions. Reflex movement is therefore intelligible which have equally to be minimized. They have their
as a primitive form of “work” (overseen by homeostatic origins in the cellular processes of the body and give
“control centres”) in the service of this basic tendency. rise to the major needs: hunger, respiration, sexuality
The imperative to minimize F supplies the aim and and the like. (In this formulation, the internal milieu of
purpose of reflex movement. If we go further back the organism is “external” to the blanketed nervous
4 M. SOLMS

system, no less than the world outside is. The generative and similarly constructed neurons, which have contact
model within the blanket represents and regulates the with each other through the medium of synapses,
body, but it is not the body itself.) The system cannot which terminate upon one another as they do upon por-
simply adjust its prior predictions to minimize surprisal tions of non-nervous tissue (e.g. muscle), and in which
regarding the viable states of its body, as it can with pre- certain lines of transmission are laid down in so far as
diction errors regarding sensory samples of the external the neurons receive signals through dendrites and give
milieu; it cannot employ “model updating” of autonomic them off through an axon. They have in addition modu-
reflexes since its model of the own bodily functions is latory functions which shall be described later (the
dictated by its phenotype. Vegetative error signals only importance of the distinction between transmission
cease subject to particular conditions: when prior predic- and modulation is greatly underestimated).
tions representing the major needs are met, which If we combine this account of the neurons with the
means they must be realized in the external world concept of a generative model (Qη) in service of the
when autonomic regulation reaches its threshold. (Cf. free energy principle, we arrive at the idea of an active
the need for nourishment, for instance.) In other (“cathected”) prediction which at other times may be
words, an internal sensory demand such as hunger can inactive.11 Friston’s law finds its expression in the funda-
only be met by actively taking nutrients, and this mental hypothesis of a cathectic signal flowing from a
requires motor action; no amount of updating a predic- predictive neuron to a sensory one, with the first
tive model can meet nutritional needs directly. In order neuron transmitting the expected signal to the second.
to accomplish the compulsory actions, which deserve These two classes of neuron reflect the basic function
to be named “specific,” the changes demanded by Fris- of the nervous system, since predicting the incoming
ton’s law must involve adjustment of predictions con- signal minimizes surprisal and therefore F.12 The second-
cerning the sensory states that flow from the ary function of the nervous system, however, which calls
organism’s actions in the external world, since the for learning, is made possible by the assumption of error
system is being subjected to conditions which may be signals which may override the prior predictive ones
described as the exigencies of life. In consequence, the when the distribution of sensory states does not
nervous system is obliged to abandon its original match with what was expected. In other words,
trend to reflex action. It must put up with tolerating a because predictive models are imperfect things, the
store of free energy sufficient to meet the demand for system must expect error signals and must constantly
the specific action: whenever an active state driven by update itself to accommodate them. The structure of
a major need does not have the predicted sensory con- the nervous system, which is layered somewhat like an
sequences, the generative model of the external world archaeological site,13 dictates that the predictive
must be updated so that supplementary motor skills signals will be centrifugal and the error signals centripe-
may be acquired. This is the impetus to learning from tal. This layered arrangement yields a conception of
experience, and therefore to voluntary action. The the system’s generative model unfolding over a con-
primary trend to minimize F persists in learning, but centric predictive hierarchy, which proceeds from the
trial-and-error processes take time, and uncertainty (the homeostatic core to the sensory periphery, with each
mortal enemy of self-organizing systems) is greatly layer endeavoring to predict the pattern of neuronal
increased in the process. The obligation to tolerate activity that will occur in the layer immediately beyond
error while uncertainty is resolved therefore implies it. This makes it probable that error signals are progress-
that F can never be nullified; it can only be minimized.10 ively resisted as they are propagated inward from the
All the functions of the nervous system can be com- surface to the depths, with more error being tolerated
prised either under the aspect of the primary function (expected) towards the periphery than towards the core.
or of the secondary one imposed by the exigencies of The resistances would be located at the interfaces
life. between the layers, at their points of contact, which in
this way assume the value of barriers.14 The hypothesis
of resistance is fruitful in many directions.
(2) Second principle theorem
The neuron theory (3) Resistance
The idea of combining this quantitative theory with the The first justification for this hypothesis arises from the
theory of neuronal functioning is the second pillar of this consideration that there are many more centrifugal
thesis. The main substance of the theory (no longer con- paths of transmission in the nervous system that centri-
troversial) is that the nervous system consists of distinct petal ones, not only in the motor pathways but in
 NEUROPSYCHOANALYSIS 5

sensory ones too. Far fewer neurons propagate signals brain as though there were no barriers and which,
from the sense organs to the cortex than the other accordingly, after each passage of excitation are in the
way around. For example, the ratio of afferent connec- same state as before, and (2) those whose resistances
tions to efferent ones in the lateral geniculate body make themselves felt, so that they only allow stimuli to
(which relays information from the eyes to the visual affect them with difficulty or partially.19 The latter class
cortex and vice-versa) is about 1:10. The heavy lifting is may, after each excitation, be in a different state from
done by the predictive signals that meet the sensory before and thus afford the possibility of representing
ones. The same general arrangement applies to centri- memory.
petal and centrifugal connections within the cortex Thus there are incoming neurons (offering no resist-
itself.15 This saves an enormous amount of information ance and retaining nothing), which serve for error
processing, and therefore metabolic work, as predictive signals, and outgoing ones (loaded with resistance, and
signals (ψ) “explain away” the sensory (φ) ones. Consid- holding back F ), which are the vehicles of memory (i.e.
ering that the brain consumes about 20% of our total representation) and so of cognitive processes in
energy supplies, this is a valuable efficiency.16 This general. Henceforward I shall call the former system of
gives us a hint that incoming signals are to be linked neurons (flowing from φ) the e system and the latter
with Shannon entropy. The hypothesis of resistance one (predicting φ) the ψ system.
therefore coincides with the now widely accepted It will be well now to clear our mind as to what
notion that the cortex generates, in the first instance, assumptions about the ψ neurons are necessary in
not a bottom-up assemblage of incoming sensory (φ) order to cover the most general characteristics of
data but rather a top-down “fantasy” – an unconscious memory. This is the argument. Memories are basically
predictive (ψ) “hallucination.”17 predictions; they are about the past but they are for
Furthermore, the theory of resistance can be turned the future. Predictions are permanently alterable by
to advantage as follows. A main characteristic of the passage of an e signal, but, crucially, ψ neurons
nervous tissue is memory: that is, quite generally, a can also code for expected e. If we introduce the
capacity for being permanently altered by single occur- concept of synapses: their synaptic weighting is
rences – which offers such a striking contrast to the brought into a permanently altered state. And since
behavior of a material that permits the passage of a psychological knowledge shows that there is such a
wave movement and thereafter returns to its former thing as consolidation on the basis of repeated learning,
condition. A psychological theory deserving any con- this synaptic alteration must consist in consolidation.
sideration must furnish an explanation of memory. Consolidation may therefore be equated with resistance
Now any such explanation comes up against the to change. An increasingly consolidated neural assem-
difficulty that it must assume on the one hand that blage is accordingly decreasingly plastic (cf. Hebb’s
neurons are permanently different after an excitation law).20 Combining this insight with the notion of a pre-
from what they were before, while nevertheless it dictive hierarchy, systems consolidation in the hierarchy
cannot be disputed that, in general, fresh excitations must consist in the deeper ψ layers becoming progress-
meet with the same conditions of reception as did the ively more resistant to change, and so more like core
earlier ones. It would seem, therefore, that neurons homeostatic predictions. They will behave more like
must be both influenced and also unaltered, unpreju- reflexes; they will expect (tolerate) less e and become
diced. We cannot offhand imagine an apparatus more automatized. We shall describe this state of the
capable of such complicated functioning; the situation successive layers as their degree of certainty. We can
is accordingly saved by attributing the characteristic of then say: Memory is represented by the parameters of
being permanently influenced by excitation to one certainty existing between the ψ neurons.
class of neurons, and, on the other hand, the unalterabil- What, then, does “certainty” in the ψ neurons depend
ity – the characteristic of being fresh for new excitations on? According to empiricist doctrine, the memory of an
– to another class. This coincides with the distinction in experience (that is, its continuing operative power)
computational neuroscience between “prediction units” depends on a factor which is called the magnitude of
and “error units” – a distinction, moreover, which is not the impression and on the frequency with which the
only found in machine-learning contexts but one that same impression is repeated. Translated into our
can also appeal to a wide range of neurophysiological theory: Certainty depends on the e which passes
findings (e.g. evoked cortical responses) for its support.18 through the ψ hierarchy during a learning process and
The theory of resistance, if we adopt this solution, can on the number of repetitions of the process. From this,
express it in the following terms. There are two classes of then, we see that F is the operative factor and that the
neurons: (1) those which allow stimuli to pass into the magnitude of surprisal (i.e. the depth to which an e
6 M. SOLMS

signal travels through the hierarchy)21 plus its generaliz- beyond it; their connections are more widespread
ability inversely determines predictive power. Generaliz- than are those of the e neurons. On this, indeed,
ability, in turn, is contingent upon the balance between depends the possibility of choice that is so character-
model “accuracy” and “complexity.”22 At the periphery istic of uncertainty. We shall see later how choice is
of the hierarchy, short-term complexity prevails, at the governed by Friston’s law. For now, consideration
cost of long-term generalizability which is enjoyed by need only be given to the most basic choice that is
the deeper predictive layers. Translated into functional- faced when a cathectic prediction confronts error
anatomical terms: Nondeclarative (subcortical) memory signals: If an action does not yield the predicted
traces are more certain than declarative (cortical) ones sensory consequences, then the system must either
because they are optimized for simplicity rather than (1) change its prediction to better explain the data, or
accuracy.23 This makes them more generalizable, both (2) if it remains confident about the original prediction,
spatially and temporally. But generalizability comes at it must obtain better data; that is, it must perform
a price: less complex models are less accurate when actions that will change its sensory input. These two
the context varies. The greater complexity of cortical pre- options – changing ψ or φ – are the fundamental mech-
dictions coincides with their higher plasticity and lower anisms of perception (i.e. here-and-now representation)
automaticity – that is, their tolerance of expected error. and action respectively. In Bayesian terms: the brain has
In a word, the cerebral cortex specializes in contextual two alternative ways of responding to prediction error.
memory; it restores model accuracy in complex and When faced with a hypothesis to which decreasing
therefore unpredictable situations. A compromise is “posterior” probability applies, it creates a better fit
inevitable: less automaticity means more uncertainty between the hypothesis and the data by changing
and therefore higher values of F. That means greater either (1) its “prior” prediction or (2) its input. The differ-
demand for work; so, the system codes Qη parameters ence between these alternatives comes down to the
with as much certainty as it can get away with. statistical direction of fit: error is reduced if the predic-
Here we are almost involuntarily reminded of the tion is changed to match the sensory input, and it is
endeavor of the nervous system, maintained through also reduced if the sensory input is changed to match
every modification, to avoid being burdened by F or to the prediction. Organisms alternate between these
keep the burden as small as possible. In the theory on two options all the time. (Think of a field-mouse
offer here, under the compulsion of the exigencies of darting through the scrub, stopping to look around,
life, the nervous system was obliged to tolerate a large darting again, stopping to look around again, and so
degree of uncertainty. This necessitated an increase in on.) Perception and action are more similar than they
the number of its neurons, and these had to be ψ seem. They are just alternative routes for reducing sur-
neurons which were arranged hierarchically: the prisal and therefore F. Nevertheless it is interesting to
expanded mammalian forebrain. It now avoids, partly observe that prior predictions which give rise to
at least, being overwhelmed with F (or unpredictability) errors can only be confirmed (as they always must be
by setting up resistances between the layers of the hier- in the case of endogenous needs) through renewed
archy through deepening degrees of consolidation. action; there is therefore a special relationship
Stated conversely: the system avoids its own destruction between drive and action.
by expanding the expectation of uncertainty towards the It remains to be seen in what else resistance consists.
periphery of its generative model. The predictions A further idea might be: By inverting the causal depen-
encoded by the newer ψ layers are therefore more tran- dencies that shaped the predictive hierarchy, the brain
sient (since they are less generalizable). This is short-term produces perceptual inferences. In other words, patterns
memory, the pivotal vehicle of tolerated uncertainty, of causation that were learnt in the world become pat-
which provides a buffer for mental work.24 This buffer terns of prediction that explain the world.25 Inverting
has significant capacity constraints, which further the hierarchy simply means shifting from learning to
explains the requirement for its predictions to be conso- predicting on the basis of what was learnt: reversing
lidated into the long-term systems. It will be seen, then, the system’s probabilistic model of the causal regu-
that progressive consolidation from the periphery larities that exist. The system then uses this model –
towards the core (progressive certainty and resistance) this virtual world – to generate inferences that guide
serves the primary function of the nervous system. its actions, which, in turn, should be viewed as hypoth-
The necessity for finding a place for memory calls for esis-testing to improve the model (when all goes well).
something further from the theory of resistance. Every The world of phenomenal experience is derived
ψ neuron must in general be presumed to have indirectly, from Qη, not from Q.26 Perception and
several paths of connection with neurons in the levels action both proceed from the inside outwards –
 NEUROPSYCHOANALYSIS 7

always from the viewpoint of the system. They are the endogenous origin. It was from this latter obligation,
system’s “best guesses” as to what lies beyond the indeed, that, owing to the exigencies of life, a compul-
blanket: provisional answers to the questions it is sion came about towards further biological develop-
putting to the world. This underscores the intentionality ment. We may then conjecture that it might actually
of biological systems. The information flow diagrams of be our function of variable resistance which had
cognitive science routinely overlook the fact that “infor- assumed one of these primary obligations. The flexible
mation” implies communication between an information system would be the group of neurons which the
source and an information receiver – a question asker. exogenous stimuli reach; the automatic system would
That is why information is coded in bits, 1s versus 0s: contain also the neurons which receive the endogenous
“yes”s versus “no”s. The diagram-makers fail to consider excitations. In that case we would not have invented the
this crucial issue: who is asking the questions that evoke function of resistance, we should have found it already in
these responses?27 existence – in the form of automatized reflexes and
instincts which are highly resistant to change. It still
remains to identify them with something known to us.
(4) The biological standpoint
In fact, we know from anatomy a system of neurons
The hypothesis of there being two systems of neurons, (the grey matter of the spinal cord and cranial nerves)
for prediction (ψ) and error (e) respectively, of which e which is alone in contact with the external world, and
consists in centripetal elements and ψ centrifugal ones, a superimposed system (the grey matter of the brain)
seems to provide an explanation of this one of the which has no direct peripheral connections but to
peculiarities of the nervous system – that of retaining which the development of the nervous system and the
and yet remaining capable of receiving. All mental acqui- cognitive functions are attached. The prosencephalon
sitions would in that case consist in the negentropic fits pretty well with our characterisation of the system
organization of the ψ system through partial and ψ, if we may assume that e paths lead directly, and inde-
locally determined lifting of the variable resistances pendently of φ, from the interior of the body to the brain
that distinguish e from ψ.28 With the advance of this – as indeed they do, to the diencephalic hypothalamus.
organization the nervous system’s capacity for fresh (The prosencephalon is divided into telencephalic and
reception would literally have reached a barrier. diencephalic components.) Now, the derivation and
Anyone, however, who is engaged scientifically in the original biological significance of the prosencephalon
construction of hypotheses will only begin to take their are not known to anatomists; according to our theory,
theories seriously if they can be fitted into knowledge it would, to put it plainly, be a sympathetic ganglion.
from more than one direction and if the arbitrariness Here is a first possibility of testing our theory upon
of a constructio ad hoc can be mitigated in relation to factual material. The graded resistance introduced by
them. It will be objected against our hypothesis of resist- the expanded memory systems of the telencephalon –
ance that it assumes two classes of neurons with a fun- which surrounds the diencephalon – would provide
damental difference in their conditions of functioning, the increasing possibilities of choice that are demanded
though there is at the moment little other basis for the by the exigencies of life.
differentiation. At all events, morphologically (that is, We will provisionally regard the ψ system as identified
histologically) nothing is known of the distinction.29 with the telencephalic cerebrum. It will now easily be
Where else are we to look for this division into understood from our biological remarks why it is pre-
classes? If possible, in the biological development of cisely ψ that is subjected to further development
the nervous system, which, in the eyes of natural scien- through an increase in the number of neurons, for the
tists, is, like everything else, something that has come tolerance of F. And it will now be realized how expedient
about gradually. We should like to know whether the it is that ψ should consist of neuronal assemblages with
degree of resistance can have had a different signifi- varying degrees of plasticity, since otherwise it would be
cance biologically, and if so, by what mechanism the unable to meet the requirements of the specific action.
two classes have developed this difference. What Hypothalamic reflexes are almost completely resistant
would be most satisfactory, of course, would be if the to change. But how did ψ arrive at the characteristic of
mechanism we are in search of should itself arise out graded resistance? After all, the automatic diencephalic
of the primitive biological part played by resistance; if core receives e signals from the internal milieu, which
so, we should have a single answer to both questions. is no less “external” to the Markov blanket than is the
Let us recall, then, that from the first the nervous outside world; if it shows so little plasticity, why should
system had two functions: the reception of stimuli ψ’s memory systems? To assume that there is an ultimate
from outside and the discharge of excitations of difference in synaptic functioning between the neurons
8 M. SOLMS

of the two systems has an unfortunate tinge of arbitrari- is difficult, and also why bodily needs typically trump
ness, though it would still be possible to show that tele- other errors in the hierarchy of priors. However, bodily
ncephalic and hypothalamic neurons have very different needs do not exhaust the range of endogenous
patterns of connectivity. Moreover, later we shall see demands that must be met in the outside world, with
that hypothalamic (and some other core diencephalic all its attendant uncertainties – that necessitated the
and mesencephalic) neurons involve a mode of com- development of the ψ systems, which required the
munication that is strikingly different from that of classi- system to tolerate tonic levels of F. In the taxonomy of
cal synaptic transmission.30 Panksepp, bodily needs – which he divides into
Another way out seems more fruitful, although theor- “sensory” (exteroceptive) and “homeostatic” (interocep-
etical. Let us recall that the homeostatic “beliefs” which tive) subtypes31 – are accompanied by a number of
comprise the viable states of the phenotype give rise to “emotional” needs. These, too, are endogenous and
prior predictions that must be confirmed (e.g. “I shall they are just as inexorable as the bodily ones.
remain between 36.5 and 37.5°C”). Predictions arising To the best of our knowledge, there are seven of
from such beliefs have very different biological impli- them in us mammals. (1) SEEKING demands engage-
cations from those associated with external sources of ment with the external world,32 which is where all the
F, such as, for example: “the person in front of me is bodily needs must be met. The innate prior prediction
going to turn left.” Exteroceptive beliefs can in large attaching to this need triggers “foraging” behavior,
part be updated repeatedly (hence their large tolerance which gives rise to all manner of learning from experi-
of e) whereas their endogenous counterparts are ence. The supplementing of such innate predictions
matters of life and death. That is why endogenous (called “instincts”) through learning is necessary with
beliefs are encoded from the outset at the core of the all the emotions, as with bodily reflexes, if not more
predictive hierarchy; their certainty is complete. Stated so.33 (2) LUST demands the release of erotic tension,
in terms of the theory of resistance: the e values that since sexual behavior will – in the average case – facili-
trigger them enjoy the highest possible “magnitude” tate reproductive success. Sexual reflexes and instincts
and the consequences that flow from them the greatest are very rudimentary (e.g. stroking the clitoris and
“generalizability”; the answer to the questions the penis, vaginal lubrication and erection, lordosis and
system asks of its internal milieu must ultimately mounting behavior with intromission); they therefore
always be yes. The certainty of exteroceptive hypoth- require much supplementation. (3) RAGE demands
eses, by contrast, may vary by degrees. Hence the removal of agents that come between the organism
differing biological implications of this series of percep- and the objects of its other needs. The associated
tual inferences: “the person in front of me is turning left”; innate behavior is “affective attack,” but as with all
“the person in front of me is racing towards me”; “the instincts this is frequently not the best prediction, so it
person in from of me is lunging at me with a knife.” too requires supplementation through learning. (4)
The consequences of these inferences enjoy increasing FEAR demands safety from predators and other
certainty; they therefore trigger beliefs that are dangers to life and limb. The associated instincts are
encoded at progressively deeper layers of the hierarchy. freezing and fleeing, but the animal must learn what
A difference in the essence of neurons located within the to fear and what else to do. (5) PANIC/GRIEF demands
expanding telencephalon on the one hand and the dien- the reliable presence of “attachment” objects: reliable
cephalic core on the other is thus replaced by a differ- caregivers in the broad sense; but these must be
ence in the environment to which they are destined. imprinted. “Protest” and “despair” behaviors are the rel-
Now, however, we must examine our assumption – evant instinctual responses.34 (6) CARE demands the
whether we may say that the magnitudes of e signals wellbeing of our offspring and their (learnt) equivalents.
reaching the neurons from the internal periphery are of This gives rise to innate nurturing behaviors – so-called
a higher order than those from the external periphery of maternal instincts, which again require much sup-
the body. There is in fact much that speaks in favor of this. plementation. (7) PLAY demands engagement with the
In the first place there is no question but that the social group, with a view to maximizing status and
internal milieu is the source of relentless quantities of thereby access to its territorial resources. The associated
F, since, due to the exigencies of life, the major needs instinct is rough and tumble play, which is governed by
simply must be met in the outside world when they innate rules concerning fairness and boundaries. Juven-
exceed the capabilities of vegetative reflexes (such as ile mammals must play, not least because they must
metabolizing the energy supplies deposited in adipose learn about reciprocity.
tissues, in the case of the need for nourishment); and It will be noticed that all these emotional needs (just
so we must learn how to master them. That is why life like the bodily ones) represent deviations from
 NEUROPSYCHOANALYSIS 9

homeostatic settling points – e.g. “no frustrating which means more entropy and F. Thus, in addition to
obstacles are in my way,” “no dangerous threats are in hierarchical levels of processing, categorization
sight,” “none of my dependants is distressed” – which becomes a necessity when the biological value of
are the emotional equivalents of bodily satiation. The different quantities changes over time (e.g. 8/10 for A
requirement for organisms to remain within these bio- is currently but not always worth more than 8/10 for B).
logically viable emotional bounds is no less obligatory It is conceivable that an extremely complex set of
than it is with the bodily needs. It will also be noticed model algorithms could evolve to calculate relative sur-
that emotional needs inevitably conflict with one vival probabilities in all predictable situations, to enable
another, even more so than bodily ones. This means organisms to automatically prioritize needs (and thereby
that needs must be prioritized, since the organism actions) on this basis. However, such complex models
cannot achieve everything at once. That demands com- are extremely expensive, in every sense of the word.
promise – which is again facilitated by learning. All these Statisticians call the exponential increase in compu-
facts count as exigencies of life. tational resources necessitated by a linear increase in
model complexity the “combinatorial explosion.” Here
we face another form of capacity constraint, generalizing
(5) The problem of quantity
from that of short-term memory and the related
As we have seen, complex organisms have multiple problem of over-fitting.
endogenous needs, each regulated by their own homeo- Compartmentalization is the standard statistical
static mechanism, all of which contribute e values to the method for achieving optimal balance between complex-
overall calculation of F. Biological needs are these error ity and accuracy. This takes many guises but what matters
values. When a need cannot be met autonomically, it here is that it enables the organism to rank its needs and
becomes a “drive”: a demand upon the mind to attendant predictions (the salient sources of expected F )
perform work. Having a need and feeling a need (i.e. over time, and to focus computational efforts on a prior-
prioritizing it) are not the same thing, as we shall see itized compartment. It is as if the system says:
later. Drive demands are registered subjectively as
under present conditions, this is the category of e in
affects, the fundamental feature of which is valence. which computational complexity cannot be sacrificed;
Although this valence is a qualitative feature of affect this is the one among the several categories of e signal
(pleasure is “good” and unpleasure is “bad”) it is never- converging upon me that provides the greatest oppor-
theless possible to quantify it: the more an e value devi- tunity for minimizing my free energy.
ates from the prescribed settling point, the greater the
Compartmentalization, then, appears to be the stat-
unpleasure, and vice-versa. Thus, at any given
istical-mechanical basis for the observed fact that each
moment, an organism’s hunger value might be 3/10
affect possesses not only a continuous hedonic valence
(which is worse than 1/10) and its thirst value might
(a degree of pleasure and unpleasure, which is some-
be 2/10 (which is better than 5/10). This, if it is so, is
thing common to all affects) but also a categorical
easily understandable.
quality (so that thirst feels different from separation dis-
However, all the more interesting are certain perspec-
tress, which feels different from disgust, etc.). But how
tives and conceptions which arise from these assump-
can the quality of feelings like thirst versus disgust,
tions. In the first place, need values cannot be
etc., be quantified?
summated in any simple way. The multiple needs
cannot be reduced to a common denominator; they
must be evaluated on separate, approximately equal
(6) Pain
scales, so that each of them can be given its due. One
cannot say that 3/10 of hunger plus 1/10 of thirst All contrivances of a biological nature have limits to their
equals 4/20 of total need, and then try to minimize efficiency, beyond which they fail. This failure is mani-
that sum, because each need must be satisfied in its fested in phenomena which border on the pathological
own right. It therefore makes sense for the system to dis- – which might be described as normal prototypes of the
tinguish its e values categorically, so that they may be pathological. We have found that the nervous system is
computed independently. The different categories of contrived in such a way that the major exteroceptive
quantity also have different implications for the system sources of F are kept off from the e neurons and still
in different contexts (for example, hunger trumps sleepi- more from ψ, by the Markov blanket and by resistance.
ness in some situations but not others). This contributes Is there a phenomenon which can be brought to
greatly to uncertainty – which requires more compu- coincide with the failure of these contrivances? Such, I
tational complexity, which means more information, think, is pain.
10 M. SOLMS

Everything that we know of pain fits in with this. The of the brainstem. This becomes even more interesting
nervous system has the most decided inclination to when we learn that the PAG is the final destination of lit-
withdrawal from pain. We see in this a manifestation of erally all the brain circuits transmitting endogenous e. It
the primary trend against a raising of F, and we infer is, in its turn, tightly interconnected with two adjacent
that pain consists in the irruption of large e’s into ψ. brainstem structures – the tectum and the midbrain
The two trends are in that case a single one. Pain sets locomotor region – which collectively comprise what
the e system as well as the ψ system in motion; there Merker calls the “midbrain decision triangle,” what Pank-
is minimal resistance to pain transmission, it is the sepp calls the primal “SELF.”36 Whereas the PAG registers
most imperative of sensory processes. Thus the ψ residual endogenous needs, the tectum generates a
neurons seem permeable to it; it therefore consists in compressed two-dimensional representation of the
the transmission of e signals of a very high magnitude. external world – a “saliency map” or “priority map.”37
We saw previously that the magnitude of an e signal The interface between these two structures generates
determines the depth to which it penetrates the ψ hier- a picture of “where things stand now” in terms of both
archy. This implies that pain signals penetrate it comple- subjective needs (e) and objective opportunities (φ) at
tely. They behave like endogenous e; which fits neatly the completion of each action cycle (M). Thus, the mid-
with the observation that pain dispenses with the dis- brain decision triangle performs the crucial prioritization
tinction between the e and ψ systems. It is both an exter- function discussed previously: “what to do next” – since
oceptive sensation and an affect. It therefore behaves the organism cannot do everything at once; since it
like a core homeostatic belief, which is why a highly gen- cannot suppress all sources of F simultaneously. The
eralizable withdrawal reflex is triggered by pain. Now we second question is: how is this decision registered by
realize that these characteristics apply to all the “sensory the organism? The answer is: as feeling. A prioritized
affects,” not only to pain; consider disgust, for example, need is the one that is felt (i.e. selected). This registers
and fright – but here we will use pain as our model the currently most salient source of expected F. Felt
example. Attention is always reflexively grabbed (as affects therefore convey the decision mentioned
opposed to voluntarily directed) by sensory affects. It is before: “under present conditions, this is the category
via this evolutionary bridge, we may suppose, that of e in which computational complexity cannot be
valence was originally extended from the interoceptive sacrificed.” The feeling of a need (as opposed to the
core onto external representations of objects. In fact, mere existence of one) makes a big difference to what
sensory affects are probably the origin of aesthetic the subject of that need will do next. Hence the distinc-
experience in general (e.g. music, painting and dance). tion between needs and drives. So, in a sense, the PAG is
Emotional affects, too, we now realize, entail innately both the terminus of every affect circuit and the genesis
valenced object representations (e.g. becoming separ- of every newly felt affect. This brings us back to the ques-
ated from a caregiver or being attacked by a predator) tion with which I ended the last section: how can feeling
which are the basis for learning from experience – that (the subjective aspect of drive) be quantified?
is, for associating empirical objects by analogy with
the innate ones.35 The deep connection between
(7) The problem of quality
bodily and emotional affects now comes into view. For
example, pain and PANIC/GRIEF are modulated by the Hitherto, flowing from this question, nothing whatever
same molecules – µ opioids – so that the mental pain has been said of the fact that every psychological
of separation may be considered an evolutionary exten- theory, apart from what it achieves from the viewpoint
sion of physical pain. Similarly, pain experiences encode of natural science, must fulfill yet another major require-
FEAR associations like nothing else; fear conditioning ment. It should explain to us what we experience through
occurs through single-exposure learning. It is not our consciousness; and, since this (consciousness) knows
difficult to see why that should be the case. nothing of what we have so far been assuming – quan-
Two questions flow from these considerations, and tities and neurons – it should explain this lack of knowl-
they will prove pivotal. First: where do pain signals termi- edge as well.
nate after they have breached ψ? In functional-anatom- We at once become clear about a postulate which has
ical terms, where is the core of the predictive hierarchy been guiding us up to now. Before we got to this
located; what is the φ terminus and the ultimate problem of quality, we were treating mental processes
source of M – the hierarchy’s meta-homeostatic control as something that could dispense with experience
center? This coincides with a major issue we addressed through consciousness, as something which exists inde-
before: where is the system’s question-asker, the infor- pendently of such experience. We are prepared to find
mation receiver? It is in the periaqueductal grey (PAG) that some of our quantitative inferences are not
 NEUROPSYCHOANALYSIS 11

confirmed through consciousness. If we do not let our- the original trend to minimize F must triumph once
selves be confused on that account, it follows, from more. The Markov blanket was a screen that would
the postulate of consciousness providing neither com- only allow external states (Q) to influence internal
plete nor trustworthy knowledge of our own infor- states via its sensory states (φ). The ψ system was
mation processing, that information processing (and already protecting the organism against entropic
therefore mental processing) is in the first instance to forces by meeting these φ states with outgoing predic-
be regarded to its whole extent as unconscious and is tions. This enabled it to act on the external world (via
to be inferred like other natural things. the blanket’s M states) in service of the primary function.
In that case, however, a place has to be found for the It had to deal only with updating its representational
content of consciousness in our quantitative ψ pro- model (Qη) on the basis of e processing. As a further
cesses. Consciousness gives us what are called qualia – step, it is to be suspected that the system ω, which is
qualities – phenomena which are different in a great located deeper within the brain than ψ, is moved by a
multiplicity of ways and whose difference is distin- further negentropic function. It would seem as though
guished according to its relations with the world the characteristic of quality (that is, conscious experi-
beyond the blanket. Within this difference there are ence) comes about in connection with the predictive
series, similarities and so on, but there are in fact no processing of φ via ψ. Now it dawns on us that error
manifest quantities in it. It may be asked how qualities cannot be got rid of by minimizing the difference
originate and where qualities originate. These are ques- between ψ(φ) and φ through goal-directed M and Qη
tions which call for the most careful examination and updating alone – that is, through action and perception
which can only be treated roughly here. alone.
Where do qualities originate? Not in the external At this point, however, we are met by what seems to
world. For, out there, according to the view of our be an immense difficulty. We have seen that resistance
natural science, to which psychology too must be sub- opposes e and that the ψ neurons are the vehicles of
jected, there is only information and nothing else.38 In this resistance. If the ω neurons support the ψ function,
the e system perhaps? That tallies with the fact that then they too must be resistant – perhaps even more
qualia are linked with the affective and other sensory resistant than ψ. But that is a characteristic that we
modalities, but it is contradicted by everything that cannot grant to the vehicles of consciousness. The mut-
(wrongly, as it turns out) argues in favor of the seat of ability of their content, the transitoriness of conscious-
consciousness being in the upper storeys of the ness, the easy linking of qualities simultaneously felt –
nervous system, in the telencephalon. In the ψ system all of this tallies only with complete permeability of
then. Against this, however, there is a weighty objection. the ω neurons, together with total restoration of their
During conscious perception the e and ψ systems are in former state following the passage of an e signal. The
operation together; but there is a broader class of ω neurons behave like sensory receptors or like the
mental process which is no doubt performed exclusively error units themselves, and in them we could find no
in ψ – cognition in general – and this, as we know, is for place for a memory. Permeability, then, complete transi-
the most part without quality. Most cognition brings toriness, which does not arise from resistance. From
about nothing that has the peculiar quality of conscious where else can it arise?
perception.39 Thus we summon up courage to assume I can see only one way out of the difficulty: a revision
that there is a third system of neurons – ω perhaps we of our previous hypothesis about how e (and therefore
might call it – which is excited along with felt affect F) is minimized. So far, I have regarded the only alterna-
and conscious perception but not with cognition in tives as action and perception (i.e. representation in Bren-
general, and whose contribution to information proces- tano’s sense)40 – which are fundamentally unconscious
sing gives rise to the various qualities – that is to say, to processes. But there must be still another alternative,
conscious experience. also of a predictive nature; and the mechanics of statisti-
If we keep firmly to the fact that our consciousness cal physics have allowed another characteristic of infor-
furnishes only qualities, whereas science recognizes mation to be measured. We speak of this as precision for
only quantities, a characterisation of the ω neurons short. Precision is a measure of confidence. Thus I shall
emerges, as though by rule of three. For whereas assume that an error signal (the difference between a
natural science (including information science) has set predicted distribution and the sample actually obtained)
about the task of tracing all the qualities of our experi- is assessed by the brain not only by comparing the
ence back to quantities, it is to be expected from the means of its predicted and actual sensory distributions
structure of the nervous system that it contains contri- but also the variation about those means. A large
vances for transforming quantity into quality; and here amount of variation in a φ sample (i.e. low precision)
12 M. SOLMS

makes the system less confident about the fit. Judge- precision-weighting process leads us to its anatomical
ments of difference between an outgoing prediction localization: baseline and adjusted ω values of ψ must
and an incoming error signal are easier to make when be set by the brainstem reticular activating system,
the distribution is narrow and precise. This is indeed a since what information scientists call precision is what
transient variable, but it enjoys a significant degree of neurophysiologists call “arousal.”45 We now discover
control; for error signals which are assigned low pre- that the capacity for arousal (i.e. modulation of synaptic
cision will be sequestered in the sensory epithelium – transmission, also known as “post-synaptic gain”) is what
such as occurs in sleep, for example. The ω values there- distinguishes the basic physiology of ω system neurons
fore determine the magnitude of e and the generalizability from that of ψ and e neurons. This distinctive capacity –
of ψ. In other words, precision (confidence) determines precision weighting, confidence, which is calibrated in
which e signals will be aroused, which will be selected direct relation to the feelings selected by the PAG – is
for ongoing processing, and which will be resisted. So, applied to the expanded telencephalon by primitive
quality is quantified by the system via the measurement brainstem structures. Consciousness is not, therefore,
of its confidence. after all, a function of the upper storeys of the brain.46
Here very much remains to be done in the way of If conscious experience is quantified by modulating
functional-anatomical clarification, for in neurophysiol- the system’s confidence in incoming and outgoing
ogy, too, the general laws of information must apply signals, where do the differences between its various
without contradiction. The hypothesis on offer here qualities spring from? Affects provide the simplest
assumes that the ω neurons are capable of evaluating answer. Their qualities vary in two respects. As we
e by means of precision modulation; they must prioritize have seen, the valence (pleasure-unpleasure) of an
the various e contributions to F on a categorical (i.e. affect is determined by the degree and direction of
qualitative) and contextual basis. This prioritization (or change in its homeostatic deviation, represented by
salience-measuring) function – which is performed, as the residual e values converging on the PAG following
we know, by the midbrain decision triangle – now each action cycle. Increasing confidence in a prediction
emerges as the fundamental basis of consciousness.41 as to how homeostasis may be regained (which
The ψ units, too, have their precisions; but that system implies decreasing confidence in the corresponding
is without quality whenever its ω values remain monoto- error signals) is “good” and the converse is “bad.” But
nous – when confidence doesn’t change – that is, during recall that consciousness of such valence arises only in
stereotyped, automatized behaviors. I repeat: the the prioritized (i.e. the currently most salient) category
expected precisions assigned by the system represent of affect, whose priority was determined by midbrain
the confidence of the system. Deviations from the evaluation of all the categories of residual e. The ω
expected precisions of ψ units (which necessarily vary values of whole categories of affect was assigned accord-
inversely with those of the corresponding e units)42 ingly. Thereafter expected ω values within each category
then become the quantitative basis of ψ quality – which (within the expected context that flows from it) are
in turn licenses voluntary behavior, the consequences assigned to the relevant ψ units throughout the predic-
of which are fed back to the midbrain decision triangle tive hierarchy, in accordance with the expected φ conse-
in a circular fashion. This implies that changes in the quences of each M. The expected precisions within the
expected ω values in the prioritized categories of ψ non-prioritized categories will be assigned monotonous
and e (i.e. deviations from expected precisions in the bil- ω values, yielding automatized action sequences, while
lions of little homeostats that make up the cathected those within the prioritized category (where the most
predictive hierarchy) during the unfolding of an expected salient uncertainty lies) must be palpated and adjusted
context generated in ψ (which flows from the selected during the action sequence, yielding voluntary behavior.
action cycle) must be palpated by the ω system, and Exteroceptive qualia require a more complex answer.
the precision values (the system’s confidence) must be They, too, are categorized qualitatively, with each
adjusted accordingly – in line with the actual amplitude whole modality possessing its own phenomenal “feel,”
of incoming e signals. This enables the ongoing evalu- but the modalities of perception (unlike those of
ation of “what to do next.” This palpating process (i.e. affect) typically come to consciousness simultaneously.
active modulation of prioritized ψ and e precisions) is It is true that the relative ω values of these categories,
the mechanistic basis of “bound” cathexis – that is, like the affective ones, are weighted differently depend-
reconsolidation in working memory.43 This cathectic ing upon contextual factors (e.g. vision is assigned more
process of reconsolidation is predictive work in progress, precision by day and hearing by night) but here the
which we now recognize is the most essential work of assignment of ω values (the confidence) within each cat-
the secondary process.44 The modulatory nature of this egory during the action cycle is determined by the
 NEUROPSYCHOANALYSIS 13

prevailing affect. That is, the salience of external events vary in parallel with them (that is an “easy” problem).
within all the modalities depends upon the exigencies Addressing the “hard” (causal) problem is, however,
of life; so that, for example, sights and sounds of care- now quite possible in some detail.49
givers are prioritized during PANIC/GRIEF episodes, A word on the relation of this theory of consciousness
whereas those of predators are prioritized during FEAR to others. According to an advanced mechanistic theory,
episodes. The priorities are not determined by intramo- consciousness is a mere appendage to physiologico-
dal events. Thus, exteroceptive qualia come to con- chemical processes and its omission would make no
sciousness secondarily, relative to their affective alteration in the causal passage of events. According
salience. Consciousness of perception merely contextua- to another theory, consciousness is the subjective side
lizes affect, using the common currency of felt uncer- of all neuropsychological events and is thus inseparable
tainty. For the mechanical reasons outlined previously, from the causal physiological mental process. The theory
endogenous e signals always take priority over external developed here lies between these two. Here conscious-
ones – but now we learn that the ω values assigned to ness is the subjective side of one part of the information
external e signals determines their uncertainty. This processing in the nervous system, namely of the ω pro-
applies to all the billions of little homeostats, embedded cessing; and the omission of consciousness does not
within each other over the levels of the exteroceptive leave the causal chain unaltered but involves the omis-
hierarchy. Increasing confidence in the predicted φ con- sion of the contribution from ω.
sequences of an action is “good” (more certain, less If we represent consciousness as the action of ω units,
salient) while decreasing confidence is “bad” (less several consequences follow. We have posited that
certain, more salient) relative to the prevailing affective these neurons are located in the region of the meta-
valence; so that when things turn out as expected homeostatic control center of the entire system, in the
(with “satiation”), monotonous ω (and therefore automa- mesencephalic decision triangle and reticular activating
ticity) prevails in the action cycles generated by non- system, which lie beneath the deepest layers of the tele-
prioritized needs. It will be noted that “goodness” and ncephalic hierarchy and even beneath the diencephalic
“badness” (in both affective and cognitive conscious- hypothalamus. This control center is the ultimate recei-
ness) is an essential precondition for choice. Choices ver of self-system information, the ultimate asker of
can only be made if they are tied to a value system. questions to which the world and body provide
Here I must emphasize once more that precision (or answers. This implies that precision-weighting is the
confidence) is the ultimate mechanism whereby cer- system’s deepest control mechanism: it determines
tainty/uncertainty is determined. The precision of an e which predictions and errors will be aroused, and there-
signal will determine its centripetal reach through the fore how the system will represent and act upon the
predictive hierarchy (its magnitude), and therefore its world. This in turn implies that precision optimization is
plastic influence upon the generative model (its general- the causal basis of all intentional behavior accompanied
izability). Learning is therefore governed by feeling, by by choice; that is, of any self-organizing system that is
what may be called the law of affect.47 Consciousness, capable of voluntary action. It is small wonder, then,
therefore – all of it, both affective and cognitive (includ- that Panksepp called this mechanism the primal SELF
ing perceptual consciousness) – is felt uncertainty.48 (which he conceptualized physiologically, but which I
What determines the differences between the varieties am quantifying here in terms of its statistical mechanics).
of phenomenal quality is what the uncertainty is This designation suggests what the original biological
about. That is why, for example, light waves feel value of conscious experience was.
different from sound waves. So far, we have given an incomplete account of what
consciousness is. Besides the property of intentionality
just described, self-organizing systems possess selfhood.
(8) Consciousness
This not only explains their intentionality, it also implies
It is only by means of such complicated and far from per- a point of view, arising from the formation of a Markov
spicuous hypotheses that I have hitherto succeeded in blanket which brings the system into being in contradis-
introducing the phenomena of consciousness into the tinction to the not-system. Now I must ask you to cross a
structure of quantitative science. Now an attempt must Rubicon with me. I must ask you to consider the mech-
be made to explain how it is that modulatory processes anism of precision optimization, which I have just out-
in the ω neurons bring consciousness along with them. It lined in objective terms, from the subjective viewpoint
is not only a question of establishing a correlation of the system. I must ask you to empathize with the
between the characteristics of consciousness that are system. This is a perspective that most natural scientists
known to us and processes in the ω neurons which are reluctant to take, much to the detriment of science,
14 M. SOLMS

and especially of mental science; but it is a perspective says that “free energy is (approximately) the negative
that is justified precisely by the selfhood of the system. logarithm of the probability of encountering some
Conscious experience can only be registered from the actively authored sensory states”:
subjective point of view. To rule that viewpoint out of
F ≈ –logP(w(M))
science is therefore to exclude consciousness from
science (as many people do). To be clear: free energy where P denotes probability. The second equation says
and its constituent precisions are only experienceable that “the expected free energy decreases in (approxi-
within a system when it is subjectively conceived, from mate) proportion to negative log precision”:
the viewpoint of the system; experiences cannot be
E[F] ≈ E[–logP(w)] = H[P(w)] = – 1/2 · log(|v|)
observed as experiences from without, objectively. But
the existential mechanics of biological self-organizing where E[·] denotes expectation or averaging and H
systems like you and me obliges us to ask questions denotes entropy, under Gaussian assumptions about
about our own states in relation to our ambient sur- random fluctuation.
rounds. We must chronically ask, “What will happen to Remember that the whole point of the self-organizing
my free energy if I do that?” Moreover, we must ask system’s dynamics is to minimize free energy. With these
this question in relation to multiple categorical variables; relationships in place, it becomes evident that there are
so, the answers – our vital statistics – must be both quan- three ways for the system to reduce prediction error and
tified (as continuous variables) and qualified (as categ- thereby minimize F:
orical variables). What I am describing here in technical
terms is nothing too complicated, in fact. You know it (1) It can act (i.e. change M) to alter sensations (φ) so
from your personal experience. What you experience that they match the system’s predictions. This is
all the time is fluctuating pulses of feeling in response action.
to your movements through the world, as you check (2) It can change its representation of the world (Qη) to
whether everything is as you expected to find it – and produce a better prediction (ψ). This is perception.
as you try to close the gaps, somehow, when it isn’t. (3) It can adjust precision (ω) to optimally match the
The aptitudes of intentionality and subjectivity of amplitude of the incoming prediction errors (e).
complex self-organizing systems equipped with pre- This is consciousness.
cision optimization leads to the following major con-
clusion. The “equipment-evoked responses” (in Here are the equations for these three alternatives:
Wheeler’s sense) that flow from the types of question ∂ ∂F ∂F ∂e ∂w
that systems like us are obliged to ask must include exis- M=− =− = ·v·e (1)
∂t ∂M ∂e ∂M ∂M
tential values and multiple qualities. Our confidence in
the fluctuating signals – the ω “phenomena” (in Wheel- ∂ ∂F ∂F ∂e ∂c
Qh = − =− =− ·v·e (2)
er’s sense) – that systems like us register must be ∂t ∂Qh ∂e ∂Qh ∂Qh
valenced, qualified and subjective. And that just is ∂ ∂F 1
what it is like to experience consciously. The equip- v=− = · (v−1 − e · e) (3)
∂t ∂v 2
ment-evoked responses, in the case of us vertebrates
at least, are feelings. where ∂ denotes a partial derivative and t denotes time,
and prediction error and free energy are quantified as
follows:51
(9) The functioning of the apparatus
e = w(M) − c(Qh)
It is now possible to formalize our picture of the func-
1
tioning of the apparatus constituted by the dynamics F= · (e · v · e − log(v))
between the quantities I have described.50 Q plays no 2
part in the picture, since its values are hidden from the Since the third of the numbered equations formalizes
system. That means, in what follows, we have a self-con- the precision optimization process that I associate with
tained, autonomous description of mental dynamics in the evaluation of free energy which underpins conscious
terms of the system’s own internal (Qη, ψ, ω, e) states experience, I will translate only that one into words:
and Markov blanket (φ, M ) states.
The rate of change of precision (ω) over time depends
Equipped with these terms, we can formalize a
on how much free energy (F ) changes when you
complex self-organizing system’s dynamics in relation change precision. This means that precision will look
to precision optimization. I will start with two equations as if it is trying to minimize free energy.52 The rate of
which define variational free energy. The first equation this free energy minimization process is the difference
 NEUROPSYCHOANALYSIS 15

mental effect necessitates the view that on their paths

of transmission to ψ the endogenous e signals, too,
come up against resistances which are overcome only
when these thresholds are reached. There are therefore
paths of interoceptive transmission made up of multiple
segments, having a number of reflexes (which trigger
phenotypically determined M states) interpolated
between them along the way to the ψ nucleus. The dien-
cephalic hypothalamus, discussed previously, appar-
ently represents the highest of the barriers. Above
these core homeostatic thresholds, endogenous needs
act as a drive stimulus continuously, and every increase
of e is registered as an increase of the ψ stimulus –
that is, as a demand for mental work. It follows, there-
Figure 1. fore, that there is a state in which the paths of endogen-
ous need transmission become permeable. Experience
between the variance (the inverse precision) and the shows, further, that, after a drive stimulus has been pro-
sum of squared prediction errors (e.e).53 cessed by ψ, the reflex that processes the underlying
need resumes its normal resistance to change. That is,
In more basic terms, the third equation quantifies how updating of Qη predictions through learning sup-
ongoing adjustment of precision implements Friston’s plements the organism’s reflex response but never extin-
law, alongside action and perception. In short, it quan- guishes it. That is because reflex (and instinctual)
tifies how consciousness contributes to action, perception predictions are not “memories”; they are not part of ψ
and model updating – and thereby to minimizing free and are therefore essentially impervious to learning.
energy. A process of this kind is called autonomic. The ψ mode
Figure 1 represents these dynamics visually. of processing e is triggered when the autonomic mode
reaches its limits. It is evident that there is a further, ulti-
mate line of protection – namely ω modulation at the
(10) The ψ paths of transmission
level of the midbrain decision triangle and reticular acti-
Since Qη receives e from both exogenous and endogen- vating system – which is theoretically capable of pre-
ous sources, it is preferable that our conception of the venting bodily needs from attaining consciousness
telencephalic cerebrum should be divided into its two even when they breach the thresholds, thereby prevent-
components: the outer pallium layers of ψ neurons ing them from entering ψ (consider what I said about
which comprise the cerebral cortex, and its nucleus sleep). But the consequences of this for F minimization
known as the subcortical basal ganglia.54 The nucleus are potentially catastrophic due to the phenotypic exi-
is connected with the myriad paths by which endogen- gencies described before.56 That is why strong bodily
ous quantities of e enter ψ. Without excluding the possi- needs eventually win the midbrain decision contest;
bility of these paths being connected with φ in the case they are routinely assigned (indeed, they grab) high ω
of “sensory” (and perhaps “emotional”) affects, we must values which tally with their inexorable magnitudes.
nevertheless hold to our assumption that a direct Reflex action is also known to occur in the φ paths of
pathway leads from the internal body to the nuclear ψ conduction – for instance, in the case of pain as dis-
neurons, at least in the case of “homeostatic” bodily cussed previously. However, the lesser part played by
affects.55 If this is so, however, ψ is exposed to e reflexes on the exteroceptive side of ψ speaks in favor
signals at its nuclear side without the protection of the of the view that greater degrees of uncertainty are toler-
graded predictive hierarchy, and in this fact lies the able in relation to exogenous e. Still, reflex resistances do
mainspring of the mental mechanism. operate even at the peripheral extremity of the sensory
What we know of endogenous needs is that they are receptor organs.
of a phenotypic nature, that they arise continuously, and It was a very noteworthy fact that some ψ neurons are
yet only periodically become mental stimuli. That is, able to maintain a position between the characteristics
endogenous needs only periodically become “drives” of permeability and impermeability, since some are per-
which make demands upon the mind for work. We manently altered by the passage of e and some are not.
cannot avoid the idea that there are core homeostatic This did not contradict the property we assumed the ψ
thresholds; and the intermittent character of their neurons to possess of being permanently altered by a
16 M. SOLMS

current of e. That apparent contradiction was already in their wake. That appears to be why the cortical
resolved, as follows. pallium, which specializes in uncertainty and complexity,
The restoration of ψ resistance after an e signal has evolved in the first place. The ψ nucleus is at the mercy
ceased is progressively less likely to occur through the of endogenous e, and it is thus that there arises the
predictive hierarchy from the surface to its depths. impulsion which sustains all mental activity and devel-
Towards the periphery of the hierarchy, ψ synaptic opment. Philosophers know this power as the “will” –
weightings are altered on a short-term basis. Thereafter the derivative of what we call drive.
they are consolidated into the episodic and semantic
(long-term) memory systems. Each of these systems is
(11) The experience of satisfaction
less plastic than the preceding one, which means they
are progressively more resistant to updating. We are Breaching the resistance of the nuclear ψ system – by
now in a position to localize the short-term and declara- high-magnitude e signals arising either from the
tive long-term memory systems in the ψ pallium. The endogenous core or from exogenous φ – will have as
nondeclarative systems – which are still more resistant their result an effort to discharge F, an urgency which
– are accordingly (for the most part) to be located in activates M paths in an attempt to lower the ω of the
the ψ nucleus. Combining these ideas with the notion e signal. There are two classes of “motor” (M) neuron,
of autonomic thresholds, we arrive at the following for- because the meta-homeostatic control center must
mulation. Due to the magnitude and relentless nature of contend with two sets of Q, both of which are “external”
endogenous e signals, they eventually achieve ω weight- to the Markov blanket. Experience shows that here the
ings that breach the threshold into ψ (into cognition). first path to be taken is that leading to internal change
Now, however, considering that these signals enter it (the firing of autonomic reflexes, leading for example
from the side of the homeostatic core, the possibility to changes in blood pressure). As was explained at the
arises that they will spread no further than the nonde- beginning, such action only produces satisfactory
clarative nucleus. The implication of this is that, although results below homeostatic thresholds; but still internal
a prioritized need will achieve affective consciousness, it changes continue after a signal breaches the ψ
will not necessarily gain access to the pallium layers of ψ nucleus. The reduction of e precision is also made poss-
and therefore to the predictive systems which permit ible by another intervention which (for the time being)
conscious cognition and action. The affect (the feeling manages increased F – through the activation of M
of the need) can then only be resolved through involun- states that lead to what is loosely called “expression of
tary action programs. the emotions” (crying, screaming, etc.). This intervention
It follows from this that such error signals will possess calls for an alteration in the external world (supply of
the curious status of being simultaneously conscious (as nourishment, proximity of the sexual object) which, as
affect) and unconscious (as cognition). In other words, a specific action, can only be brought about in definite
the mental work demanded by the drive will be felt, ways. At first, the human organism is incapable of bring-
but the work will not take place in working memory – ing about the specific action. It takes place by extraneous
it will not enjoy conscious representation. As it help, when the attention of an experienced person is
happens, this is a common method by which nonde- drawn to the child’s state by discharge along the path
clarative learning takes place: procedural memories of internal change. In this way the expression of
(habits and skills) are consolidated through brute rep- emotion acquires a secondary function of the highest
etition, guided by the law of affect but not by delibera- importance, that of communication; and the initial help-
tive thinking; and emotional responses (e.g. fear lessness of suckling mammals is the origin of the drive to
conditioning) readily occur non-declaratively, even CARE. Instinctual life is the primal source of all moral and
upon single exposures to painful events. In fact, this is ethical motives.
the only way that long-term consolidation takes place When the helpful person has performed the work of
during the first few years of life, before the declarative the specific action in the external world for the helpless
pallium matures.57 Much remains to be explained one, the latter is in a position by means of vegetative
about why this separation of affects from ideas might reflexes immediately to carry out in the interior of the
occur in other circumstances. We will return to this. It body the activity necessary for removing the endogen-
is already clear, however, that nondeclarative predicting ous stimulus. The total event then constitutes an experi-
is not the most effective method in complex situations ence of satisfaction, which has the most radical results for
where accuracy is required. This applies specially to the development of the individual’s functions. For three
learning how to meet emotional needs, which, both indi- things occur: (1) In the homeostatic core, a waning of the
vidually and collectively, bring considerable uncertainty endogenous e signal is effected through a gradual
 NEUROPSYCHOANALYSIS 17

reduction of its ω value until it reaches zero, so the necessarily entails the assignment of baseline levels of
urgency which had produced unpleasure is brought to confidence at each layer of the anticipated scenario. (Pre-
an end (i.e. confidence in the phenotypic prediction is dicting ω can be learnt like anything else.) From the core
restored).58 (2) In the ψ pallium, an exteroceptive predic- to the periphery, however, the baselines will be pro-
tion is encoded, namely: active state Mα (crying and gressively fainter echoes of the core homeostatic
screaming) causes sensory state φα (caregiver performs settling point. The predictions must become progress-
the specific action). (3) In the ψ nucleus, a further predic- ively less precise because there is progressively more
tion is encoded, namely: active state Mα resolves variation in self-generated signals cathected from the
endogenous need α. depths of the self outwards to the uncertain world.
Cognitive scientists all too readily forget that the I do not doubt that in the first instance this wishful
major needs, too, are “sensory” states – and therefore activation will produce the same thing as a perception
overlook the fact that endogenous sources of F are the – namely an (unconscious) “hallucination.” If reflex
mainspring of mental life. We only represent the action is thereupon introduced, however, disappoint-
outside world cognitively because we must; it is where ment cannot fail to occur, as we shall see.
our needs must be met. That is why we described the
cerebrum as a sympathetic ganglion; it functions in sym-
(12) The experience of pain
pathy with the needs of the organism. We must recall
also that complex organisms like us have multiple Normally, the ψ nucleus is exposed to e from endogen-
needs; so this is a simplified account. For example, ous paths of excitation, and abnormally, even though
with frequent repetition of the satisfaction of need α not yet pathologically, in cases where excessively
(the need for nourishment), yet another prediction will strong e signals break through the screening contri-
be consolidated in the ψ nucleus, namely: sensory vances in the pallium – such as in the case of pain.
object α (the caregiver) satisfies endogenous need β Pain gives rise in ψ (1) to a large rise in surprisal, which
(the need for attachment).59 A cascade of parallel predic- is felt as unpleasure in ω, (2) to an inclination to dis-
tions is thereby established through the hierarchy, from charge F, which can be modified in certain directions,
the homeostatic core itself through the automatized ψ and (3) to a predictive facilitation between this M incli-
nucleus to the pallium. nation and a ψ(φ) representation of the noxious object
We now become acquainted with a different aspect which caused the pain. Moreover, there is no question
of the predictive hierarchy. In consequence of a pheno- but that pain has a peculiar sensory quality, which
typic belief concerning nourishment (in our model makes itself felt as a category of exteroceptive sensation
example), a core autonomic prediction that “energy and one of endogenous unpleasure.
balance will be maintained” triggers the bodily affect If the representation of the noxious object is freshly
of hunger, which predicts the basic emotion of cathected in some way – for instance, by a fresh percep-
SEEKING, which predicts the procedural habit of distress tion – an anticipatory state (FEAR) arises which is not
vocalization, which predicts the semantic belief about pain but which nevertheless has a strong resemblance
caregivers, which predicts the episodic memory that to it. As with wishful cathexis, it activates a predictive
good object α will perform the specific action, which cascade that in this case includes, alongside the ψ rep-
predicts the here-and-now sensory state φα. Each pre- resentation of the noxious object, the unpleasure and
diction in this cascade is less generalizable – both the inclination to discharge F that corresponds to the
spatially and temporally – than the one preceding it. experience of pain. Since FEAR (like all unpleasure)
That is because the internal milieu of the body is a far signifies rising endogenous e, it must be asked where
simpler place than the external object world; as the pre- this unpleasure comes from. In the actual experience
dictive cascade is traversed outwards, the deeper predic- of pain, it is often assumed that the external stimulus
tions risk decreasing accuracy due to the increasing itself transmits the unpleasure; but the reproduction of
demand for complexity (i.e. for spatial and temporal the same experience in imagination – which is distin-
[contextual] specificity). guished from perception precisely by the lack of any
Whenever the state of urgency re-appears, this external stimulus – makes clear that the operative e
cascade will be activated; it represents the first expected signal must be endogenous in both cases. Affects are
context when that particular need (for nourishment) is always endogenous; they are states of the subject, not
prioritized as felt hunger. This state of affairs can be of the outside world. This raises a question: in what,
described as wishful activation. Each level in the predic- then, does the magnitude of an exteroceptive e signal
tive hierarchy cathects the level beyond it, by self-gener- consist, since it was this that enabled it to reach the ψ
ating the incoming signal it expects to occur there. This nucleus? The answer is: it consists in arousal (selection)
18 M. SOLMS

of the signal by assigning it a high ω value. The same learning by trial and error. Here we have the primary
applies to the endogenous e that is aroused along basis for wishful approach and defensive avoidance.
with it. Rising expected F is rising uncertainty concerning Wishful approach can easily be explained by the
a core homeostatic belief (viz., “I shall not suffer tissue assumption that the cathexis of the attractive represen-
damage”) which is felt as unpleasure. Now we see how tation (the image of a “good” object) when the subject is
a self-generated ψ signal – the ψ(φ) cathexis – can in a state of desire – that is, SEEKING – greatly exceeds in
have the same attribute as exteroceptive pain: all ω is baseline ω the cathexis which occurs when there is a
modulated from the interior of the system. random perception, so that a particularly strong facili-
It remains to add, from the physiological point of view, tation leads from the ψ nucleus to the corresponding
that modulatory arousal consists not only in the release of predictive units of the pallium.
psychoactive molecules from the reticular activating It is harder to explain defensive avoidance – which, as
system (e.g. noradrenaline in the case of pain) but also we shall see, provides a template for “defence” in
from the interior of the body. The mechanism of this general – the fact that a repulsive representation (the
release is well understood. Just as there are motor image of a “bad” object) is regularly abandoned by its
neurons which, when activated, stimulate the muscles, so cathexis as soon as possible. Nevertheless, the expla-
there are “secretory” neurons which cause the release in nation should lie in the fact that the primary experience
the interior of the body of endocrinological molecules of pain was brought to an end by reflex withdrawal. The
(e.g. hormones and peptides, of which there are a great emergence of another object in place of the noxious one
number) that operate as stimuli upon the endogenous was the signal for the fact that the experience of pain
paths of transmission to the homeostatic core and to ψ – was at an end, and the ψ system, taught biologically
motor neurons which thus influence the production of through the law of affect, seeks to reproduce the state
endogenous e and supply it in roundabout ways. Evidently in ψ which marked the cessation of pain. With the
these key hypothalamic neurons are activated when expression “taught biologically” I am referring to a
certain levels of resistance in the ψ nucleus have been basis of explanation which should have independent
breached. As a result of the experience of pain (which chal- validity, even though it does not exclude, but rather
lenges a core homeostatic belief) the representation of the calls for, a recourse to statistical-mechanical principles
noxious object has acquired an excellent facilitation to the (quantification). In the instance before us it may well
core neurons, in virtue of which unpleasure is now released be the increase of uncertainty (increasing precision in
in the expectation of pain. expected error) which occurs with the cathexis of a
Support for this formulation is to be found in our noxious prediction that forces increased confidence in
knowledge of the hypothalamic–pituitary–adrenal an M discharge and thus a decreased precision in the
(HPA) axis, which releases cortisol in the example just predictive cathexis of the noxious representation as well.
discussed, but the same principles apply with the hypo-
thalamic–pituitary–gonadal axis (HPG), hypothalamic–
(14) Introduction of the “ego”
pituitary–thyroid axis (HPT) and hypothalamic–neurohy-
pophyseal systems, in the cases of other affects. In fact, however, with the hypothesis of “wishful
approach” and the inclination to “defensive avoidance”
we have already touched on a state of ψ which has
(13) Distressing and wishful states
not yet been discussed. For these two processes indicate
The residues of these two kinds of experience (pain and that an organization has been formed in ψ whose pres-
satisfaction) which we have been discussing are distres- ence interferes with passages of e which on the first
sing and wishful predictive states. These have in common occasion occurred in a particular way (i.e. accompanied
the fact that they both involve a raising of surprisal in the by satisfaction or pain). This unconscious organization
ψ nucleus – brought about by changes in core homeo- is called the “ego.” It is important to notice that the
static ω(e) values, which increase in the case of distress ego’s ψ organization, as opposed to its selective cathe-
and decrease in the case of wish. Both states are of the xis, serves automatic action programs and therefore
greatest importance for ψ, for they leave behind them the primary function. It can easily be depicted if we con-
motives which are of a compulsive kind. The wishful sider that the regular reception of endogenous e in
state results in an attraction towards the object certain neurons of the ψ nucleus and the predictive
wished-for, or, more exactly, towards its Qη represen- effect proceeding thence will produce a repetition of
tation; the experience of pain leads to a repulsion, a dis- previously learned paths to satisfaction. A driving
inclination to keeping the noxious representation endogenous e signal which breaks into the ψ system
cathected. All of this flows from the law of affect – will automatically proceed in the direction of the
 NEUROPSYCHOANALYSIS 19

predictions with the largest certainty (greatest resist- underwrites choice and is the causal contribution of con-
ance) and will set up a freely mobile cascade in that sciousness. The guiding force by which exteroceptive
direction. A compulsion to repeat will thus be triggered precisions are adjusted is by the feelings that the unfold-
by fresh exogenous e signals of high magnitude. ing errors evoke in the PAG, throughout the action
While it must be the endeavor of the ego to give off sequence. It is on this basis that the system’s confidence
rising F by the method of satisfaction, this cannot in the consolidated predictive pathways that would
happen in any other way than by its influencing the rep- otherwise have been used is decreased. This enables
etition of experiences of pain and of distress, and by the the system to temporarily facilitate alternative pathways:
following method, which is described generally as to up-regulate their precision. This might be called “side
inhibition. cathexis”: inhibition of the automatized pathways. This
To put this more fully: an endogenous e signal (or an leads to spreading activation of predictive cathexes,
exogenous one which was sufficiently strong to have which allows qualitative palpating of a wider range of
reached the ψ nucleus, and therefore to have aroused e signals. Without inhibition of the primary processes,
affective e) will trigger a predictive cascade that will voluntary behavior – and both biologically and compu-
divide up in the direction of the pathways that provide tationally expedient learning – would be impossible.
the greatest probability of suppressing (explaining This provides a deep mechanistic account of the second-
away) the incoming e. This typically requires motor ary process: the basis of the ability to tolerate F is ω
action. The M course taken is dependent both on e optimization. Later we shall see that this mechanism
and the relative distribution of the ψ predictions – pre- underwrites thinking too.
dictions that were consolidated over a lifetime by the The ego is to be defined as the totality of the ψ cathe-
law of affect. This approach favors pathways with the xes, at the given time, in which a monotonous ω com-
greatest certainty – pathways which have proven to ponent is distinguished from a changing and
minimize F in the average situation. This yields stereo- temporary one.60 It is easy to see that changing cathexis
typed behaviors which do not serve the organism interacts with the notion of graded resistance between
equally well in all contexts. The inevitable consequence layers of ψ neurons and with their orthogonal compart-
is surprising experiences, which, if the organism survives mentalization into qualitative categories, both of which
them, may prompt further learning by trial and error. are a part of the ego’s possessions. They represent
This is biologically expensive in almost every way, cathectic possibilities, if the ego is altered, for determin-
including the fact that it comes at the cost of ever- ing its extent in the next few moments.
increasing complexity – a cost that ultimately over- Two further mechanisms now come into view, and
whelms the capacity constraints of ψ. We have, there- they, too, are pivotal. These are called repression and
fore, already come to know a third powerful factor, defence. The ψ nucleus is regularly bombarded by
which was the factor that prioritized the driving category endogenous e. These errors when prioritized by ω are
of endogenous e in the first place. This factor is ω optim- felt as affects but not as ideas – unless and until they
ization. The selection (prioritization) of an affect auto- proceed from the ψ nucleus to the pallium. On the
matically generates an expected context within which mechanism just outlined, this means that only inhibited
the currently most salient need would be resolved. ψ processes would typically become cognitively con-
This expected context, too, implicates ω, since expected scious – “declarable” – for the automatized processes
precisions must be predicted like everything else. Now proceed entirely according to the expected precisions,
we come to the enormous advantage that conscious- which must be monotonous. Therefore, when, for what-
ness bestows. Although ω greatly reduces compu- ever reason, the ego does not tolerate a particular auto-
tational complexity through need prioritization, matized prediction becoming inhibited through side
compartmentalization and precise predicting of cathexis, which means it does not tolerate it being
sensory-motor contexts, this does not explain why the held in mind (in working memory), that prediction will
ensuing action sequence must be felt. The reason it remain in a state of non-declarative certainty. This is
must be felt is because the degrees of freedom at the what happens with “repression.” Repressed predictions,
periphery of the hierarchy are so large that ψ is readily then, constitute that portion of ψ which remains or
overwhelmed by the combinational explosion. It is becomes illegitimately certain. Predictions that do not
almost impossible to predict the present in every hit the mark (that do not meet an emotional need in
detail. The system’s solution, we have seen, is to feel its reality) inevitably give rise to repeated error signals.
way through it: to palpate the incoming error signals Such predictions should be updated. If this does not
as they occur, within the prioritised category, and register occur, the ego suffers feelings and does not know why.
deviations from the expected precisions. This Its only recourse, then, in the case of repressed
20 M. SOLMS

predictions, is to explain away the affect by means of imaginary experience results in ongoing consolidation:
declarative cognitive work concerning other predictions increasing the certainty of viable synaptic connections
– which falls short of updating the actual cause of the (i.e. latent predictions) and discarding redundant or
unpleasure. This secondary process, which follows from weak ones. This “synaptic pruning” optimizes the ego’s
the inevitable failure of repressed predictions, entails balance between complexity and accuracy in an
side cathexis not of the repressed cascade itself but of ongoing fashion. This entails what we call phantasy.
other predictive assemblages which can avert or miti-
gate its consequences. This is what happens with
(15) Primary and secondary process in Ψ
“defence,” which is clearly not synonymous with
repression. We return now to the functioning of the ego in its task-
Now a word on the highly interesting “resting state” related modes. It follows from what has developed so far
of the ego.61 We have said that the ψ nucleus is routinely that the ego, which we can treat as regards its trends like
bombarded from the diencephalic and mesencephalic the nervous system as a whole, will, when its task-related
core. When these e signals break through (become prior- processes are uninfluenced by inhibition, be made help-
itized) they activate core predictions which run their less and suffer injury under two conditions.
course through the hierarchy. In all cases where there This will happen in the first place if, while it is in a
is lack of certainty, the predictive cascade should be pro- wishful state, it newly cathects the memory of a
gressively inhibited towards the periphery, enabling side desired object and then acts upon it. In that case satis-
cathexis (working memory) and voluntary behavior, and, faction must fail to occur, because the object is not
thereby, expedient learning from experience. However, real but is present only as a memory. Ψ is not in a pos-
this does not occur with non-prioritized needs – which ition, to begin with, to make this distinction if it works
include the repressed ones by definition. These must on the basis of the predictive cascade described
be relegated to automatized predictive cascades (with before. Thus it requires a criterion from elsewhere in
monotonous ω), where inexpedient learning takes order to decide between perception and memory.
place – through high-magnitude exogenous errors Likewise, ψ is in need of an indication that will draw
which “grab” precision. The safer lessons learnt its attention to the re-cathexis of the memory of a
through working memory and voluntary behavior, by noxious object – as opposed to the presence of a real
contrast, are fed back to ψ through graded consolida- one – and enable it to obviate, by means of side cathexis,
tion. We have seen that only the most generalizable of the consequent release of unpleasure. If ψ is able to put
these newly acquired predictions are consolidated as this inhibition into operation soon enough, the release
far as the nucleus, whereupon they become resistant of unpleasure, and at the same time defence, will be
to change. slight; otherwise there will be immense unpleasure
Consolidation entails generalizability. We know and excessive defence, as if the noxious thought were
already how this is ordinarily achieved, but we have real.
seen also that the process can be very expensive. Sim- Both wishful cathexis and unnecessary release of
plification now emerges as the essential contribution of unpleasure, where a memory is cathected anew, can
the resting state, which characterises the ego’s mode be biologically detrimental. This is true of wishful cathe-
of functioning when it is unoccupied by major exogen- xis whenever it exceeds a certain amount and so acts as
ous sources of e. In this state, in the absence of pressing an enticement to action; and it is true of a release of
demands, the ω decision triangle prioritizes default- unpleasure whenever a noxious cathexis results not
mode SEEKING – even, and perhaps especially (as we from external events but only from ψ itself. Here, once
shall see), during sleep. This engages the “default again, it is a question of an indication to distinguish
mode network,” which includes the ψ nucleus and between a perception and a memory.
extends into relatively deep layers of the pallium – It is surely the ω neurons which furnish this indication:
mainly in the cortical midline – where declarative cogni- the indication of reality. We have seen that it is a miscon-
tion can occur. In the default mode, SEEKING does what ception to assume that every perception is accompanied
it always does: it engages uncertainty (e.g. novelty) with by conscious experience; quality is generated only on
the aim of reducing it. This mode of cognition is also the basis of prioritized needs. The affect generated by
called “mind wandering” – it is a sort of internal foraging need-prioritization renders normally unconscious per-
– whereby wide-ranging imaginary scenarios are pal- ceptions (those with monotonous ω) salient. Now they
pated under safe conditions; that is, virtually. (These must be palpated, and their ω values must be adjusted
safe conditions are ensured through φ and M deactiva- accordingly. That is made possible by attention, which
tion, as will be discussed further below.) Learning from now emerges as the exogenous equivalent of endogenous
 NEUROPSYCHOANALYSIS 21

need prioritization. The system’s confidence in a predic- to assemblages a + c. Since this will be the more
tion (its ω value) is determined initially by the driving common case, more common than that of identity, it
affect, which carries the twin risks we have been discuss- calls for more exact consideration. Biological experience
ing. But perception is not the same as hallucination; it is will teach here that it is unsafe to initiate action if the
controlled hallucination. Putative hallucinations are con- indications of reality (the ω assigned to the error
trolled by incoming error signals, leading to reconsolida- signals generated by sample pattern c) do not confirm
tion of the cathected predictions, as modulated by the whole expected assemblage but only a part of it. A
precision. This modulatory control – which happens way is now found, however, of completing the similarity
throughout the layers of the hierarchy – is supplied by into an identity. The sample pattern, if it is compared
the ω values that are assigned to each incoming error with other sampled patterns, can be dissected into a
signal (and its corresponding outgoing prediction). So, component portion, pattern a, which on the whole
hallucination is prevented through withdrawal of confi- remains the same, and a second component portion,
dence from predictions that attract strong disconfirma- pattern b, which for the most part varies. Language
tory error signals. This arouses consciousness of the will later apply the term judgement to this dissection
error signals (which then loop back to update the and will discover the resemblance which in fact exists
priors, to whatever depth is necessary). What comes to between the nucleus of the ego and the constant
consciousness, then, is not a prediction but rather its sampled component on the one hand and between
sensory contradiction – i.e. modulation of the mismatch. the changing pattern in the pallium and the inconstant
The indication of reality may therefore be conceptual- component on the other; language will call pattern a
ized as a mechanism for disambiguation. We see once the thing and pattern b its activity or attribute – in
more that consciousness, all of it, including perceptual short, its predicate.
consciousness, is felt uncertainty. Perceptions that carry Thus judging is a ψ process which is only made poss-
full confidence fade from consciousness (cf. habitu- ible by inhibition by the ego and which is evoked by the
ation).62 We see now that, here too, it is the tolerance dissimilarity between the wishful cathexis of a prediction
of error supplied by secondary process inhibition that and a sensory pattern which is similar to it. It can be
makes possible the criterion for distinguishing inferred from this that matching between the two
between perception and memory. becomes a biological signal for ending the act of cogni-
tion and for allowing action to begin. Their mismatch
gives the impetus for the activity of conscious (and
(16) Cognition and declarative thought
effortful) cognition, which is terminated once more
We have brought forward the hypothesis that, during with their match.
the process of wishing, inhibition by the ego brings The process can be analyzed further. If sensory
about a moderated cathexis of the object wished for, pattern a matches the predicted one but pattern c is
which allows it to be cognized as not-yet real; and we sampled instead of pattern b, then the activity of the
may now proceed with the analysis of this process. ego follows the precise error signals generated by this
Several possibilities may occur. In the first case: simul- pattern c and by means of palpating the confidence
taneously with the wishful ψ cathexis of a representation levels along its collateral connections, through side
of the object, a φ signal of it is present. If so, the two pat- cathexis, causes new sensory patterns to emerge until
terns of neural activity match – which cannot be made perception b is generated. As a rule, the collateral con-
use of biologically – and the wishful cascade ends in nections activate the memory of an action which is
automatized action, which need not attract attention. interpolated between assemblages c and b; and when
This case is easily dealt with but it almost never this inference is freshly activated through a movement
happens in reality that predictions and sensory carried out really, the perception of pattern b, and at
samples match perfectly. This leads to the second case: the same time the identity that is being sought, are
the wishful cathexis is present and along with it a established. The alternative to this movement – as we
sample which does not match it wholly but only in saw in Section 3 – would be for the ego to adjust its pre-
part. For the time has come to remember that percep- diction from patterns b to c; but this solution is typically
tual patterns never involve single neurons but always excluded in the case of endogenous needs. Let us
distributions of sensory signals in neuronal assemblages. suppose, accordingly, that the object wished for (i.e. pre-
So far we have neglected this feature; it is time to take it dicted) by the infant is the neuronal assemblage corre-
into account. Let us suppose that, quite generally, the sponding to the image of the mother’s breast and
wishful cathexis relates to neuronal assemblage a + front view of the nipple, and that the first pattern
neuronal assemblage b, and the sampled distribution sensed is a side view of the same object, without the
22 M. SOLMS

nipple. This generates salient error (and therefore con- predictions in question. Unpleasurable paths, however,
sciousness). In the child’s memory there is the represen- retain their great value for obvious biological reasons
tation of an experience, made by chance in the course of in directing declarative cognition.
sucking but consolidated through the law of affect, that
with a particular head movement the front image turns
(17) Remembering and judging
into the side image. The side image which is now seen
leads to high-precision error signals which continue Thus declarative thought has a practical aim and a bio-
until the passage of the movement brings the expected logically established end – namely, to lead ω from the
front image into view. superfluous (unwanted) sample back to the cathexis of
There is not much judgement about this as yet; but it the unfulfilled prediction. With this, identity and the
is an example of the possibility of arriving, by precision right to action are achieved, if in addition the indication
optimization, at an action which is already one of the of reality appears from assembly b. The process can,
trial-and-error offshoots of the specific action, without however, make itself independent of the latter aim and
enduring the dangers associated with fresh learning strive only for identity. If so, we have before us a pure
from experience. act of thought, though this can in any case be put to
There is no doubt that it is ω optimization that under- practical use later. Here, moreover, the cathected ego
writes this cognitive palpating of temporary action/per- is behaving in exactly the same manner.
ception possibilities which gives rise to voluntary We now come to a third possibility that can arise in a
movement. The choice is dominated not by ψ consolida- wishful state: when, that is, there is a wishful cathexis
tion but by an aim. What is this aim and how is it and a sensory sample emerges which does not coincide
achieved? in any way with the wished-for pattern (mem+). There-
The aim (i.e. SEEKING) is to find the missing sensory upon there arises an interest in cognizing this sensory
pattern b and to decrease precision in the error signal image, so that it may perhaps after all be possible to
caused by pattern c and thereby improve confidence find a pathway from it to mem+. It is to be assumed
in the corresponding prediction of pattern b. It is that, with this aim in view, the sensory pattern (or
reached by an experimental displacement of ω along rather, the error signal it gives rise to) is assigned a
every salient pathway, and it is clear that for this high precision, as happened in the previous case with
purpose sometimes a larger and sometimes a smaller only a component of it, pattern c. If the sensory image
tolerance of side cathexis is necessary, according to is not absolutely new, it will now recall and revive a
whether one can make use of the certainties that are memory image with which it coincides at least partly.
present or whether one has to work against them. The If the sensed pattern is completely new, the same
struggle between the consolidated predictions and the process occurs by analogy with something known. The
changing and uncertain ω cathexes which underwrite previous process of thought is now repeated in connec-
choice and voluntary action is characteristic of the sec- tion with this memory image, though to some extent
ondary process of declarative cognition, in contrast to without the aim which was afforded previously by the
the automatized primary sequence with its false cathected wishful idea.
certainties. In so far as the patterns match, they give no occasion
What is it that directs this cognitive palpating? The for activity of thought. On the other hand, the non-
fact that the representation of the wished-for object matching portions “arouse interest” and can give
can be held in mind (in working memory) along with – occasion for thought activity in two ways. The attention
or due to – all the associated uncertainty. As we now is either directed onto the aroused memories and sets an
know, indications of reality (perceptual consciousness) aimless activity of memory at work, which is thus moved
plays a pivotal role in the process, but this too is ulti- by differences and not similarities, or it remains attached
mately directed by the potential pleasures and unplea- to the components of the sensory sample which have
sures that are prioritized in the midbrain decision newly emerged and in that case exhibits an equally
triangle. aimless activity of judging.
In the course of this palpating it may happen that the Let us suppose that the object which furnishes the
changing ω weighting activates a memory which is con- sensory pattern resembles the subject: a fellow human
nected with an experience of pain, and it thus gives being. If so, the theoretical interest taken in it is also
occasion for the release of unpleasure in the interior of explained by the fact that an object like this was simul-
the system. Since this is a sure sign that sensory taneously the subject’s first satisfying object and
pattern b is not to be reached along that pathway, the further its first noxious object, as well as its sole care-
cognitive cathexis is at once diverted from the giver. For this reason it is in relation to a fellow human
 NEUROPSYCHOANALYSIS 23

being that a person learns to cognize. Then the sensory an experience of one’s own. Judging thought operates in
patterns proceeding from the encounter with this fellow advance of declarative thought by furnishing it with
human being will in part be new and not-matchable – its ready-made synaptic connections for further associative
facial features, for instance, in the visual sphere; but traveling. If after the conclusion of the act of thought the
other visual impressions – e.g. those of the movements indication of reality reaches the perception, then a judge-
of its hands – will coincide in the subject with memories ment of reality, confidence, has been achieved and the aim
of quite similar visual impressions of its own, of its own of the whole activity is attained.
body, memories which are associated with memories of As regards judging, there is further to be remarked
movements experienced by itself. Other impressions of that its basis is obviously the presence of bodily experi-
the object too – if for instance, it screams – will ences, sensations and motor images of one’s own. So
awaken the memory of the subject’s own screaming long as these are absent, the variable portion of the
and at the same time its own experiences of pain. sensory pattern remains ununderstood – that is, it can
These memories are encoded in assemblages of what be reproduced but does not point a direction for
are called “mirror neurons”63 (although the fact that further paths of thought. Thus, for instance (and this
mirror neurons encode memories – predictions – is fre- will become important for what is to follow in Part II),
quently overlooked). Thus the pattern of the fellow no sexual experiences produce any effect upon predic-
human being falls apart into two components, of tive pathways for satisfying LUST so long as the
which one makes an impression by its constant structure subject is ignorant of mature sexual desire – in
and stays together as a thing, while the other can be general, that is, till the beginning of puberty. To be
understood by the activity of mirror neurons – that is, clear: while early sexual experiences have immediate
can be traced back to information from the subject’s consequences for the management of drives that are
own body. This dissection of the sensory pattern is not in a state of latency – such as SEEKING, FEAR,
described as cognizing it; it involves a judgement and PANIC/GRIEF and PLAY – and therefore can elicit second-
when this last aim has been attained it comes to an end. ary emotions like shame and guilt, the predictive effects
Judgement, as will be seen, is not a primary function, upon LUST (and only LUST) are largely deferred.
but presupposes ω cathexis from the ego of non-match- Primary judging seems to presuppose a lesser
ing portions of the sensory sample;64 in the first instance influence by the cathected ego than do declarative
it has no practical purpose and it seems that during the acts of thought. In primary judging it is a matter of pur-
process of judging the palpating of the non-matching suing an association which is due to partial matching
component is ended, for this would explain why activi- between the wishful (predicted) and actual sensory pat-
ties, “predicates,” are separated from the subject- terns – an association to which no modification is
pattern by a comparatively loose pathway. applied. And indeed cases also occur in which the associ-
It would be possible from this point to enter deeply ative process of judging is carried out with a full amount
into the analysis of the act of judgement; but this of quantity. The sensory sample may correspond to an
would divert us from our topic. Let us content ourselves object nucleus plus a motor image. While one is
with bearing firmly in mind that it is the original interest sensing the φ pattern, one copies the movement
in establishing the situation of satisfaction that has led, in oneself – that is, one innervates so strongly the motor
the one case, to declarative consideration and in the other pattern of one’s own which is aroused towards matching
to judging, as a method of proceeding from the sensory the φ pattern, that the movement is carried out. Hence
situation that is registered in φ to the situation that is one can speak of a perception having an imitation
wished for in ψ. The necessary precondition for this value. Or the sensory pattern may arouse the memory
remains that the ψ processes should not pursue their of a sensation of pain of one’s own, so that one feels
passage uninhibited but in conjunction with an active the corresponding unpleasure and repeats the appropri-
ego. The eminently practical sense of all thought activity ate defensive movement. Here we have the sympathy
would in this way seem to be demonstrated. value of perception.
However these two cases are not yet empathy, which
is a further developmental achievement. In these two
(18) Thought and reality
cases we must no doubt see the primary process in
The aim and end of all thought processes is thus to bring respect of judging, and we may assume that all second-
about a state of identity, the conveying of e signals ema- ary judging has come about through a mitigation of
nating from φ onto neuronal assemblages cathected these purely reflex processes. Thus judging, which is
from the ego. Cognitive or judging thought seeks an iden- later a means for the cognizing of an object that may
tity with a bodily cathexis, declarative thought seeks it with possibly be of practical importance, is originally an
24 M. SOLMS

associative process between signals coming from the There is no doubt, however, that the process of
outside and from one’s own body – and identification thought does leave lasting traces behind it, since a
of information from φ and from within. It is perhaps not second thinking, a re-thinking, calls for so much less
wrong to suspect that judging at the same time rep- energy than a first. In order that reality shall not be
resents a method by which e’s coming from φ can be falsified, therefore, special traces are needed, signs of
transmitted and dealt with. What we call things are resi- the processes of thought, constituting a thought
dues which evade being judged. memory which it is not yet possible to shape. We shall
The example of judgment gives us for the first time a hear later (in Part III) by what means the traces of
hint of the difference in their quantitative characteristic thought processes are distinguished from those of
which is to be discovered between thought and the reality.
primary process. It is justifiable to suppose that during
thought a slight magnitude of motor innervation passes
from ψ – only, of course, if during the process a motor (19) Primary processes – sleep and dreams
or key neuron has been innervated. Nevertheless, it
would be wrong to take this active state for the process The question now arises as to what, then, the quantitat-
of thought itself, of which it is only an unintended subsidi- ive means may be by which the ψ primary process is sus-
ary effect. The process of thought consists in the cathexis tained. In the case of an experience of pain it is evidently
of ψ neurons, accompanied by a change, brought about the irrupting e from φ; in the case of self-generated dis-
by side cathexis from the ego, in what is imposed by tressing states it is the endogenous e released by remem-
the synaptic weightings. It is intelligible from the statisti- bering. In the case of the secondary process of
cal-mechanical point of view that here only a part of the e declarative thought a greater or lesser ω can obviously
is able to follow the certainties and that the magnitude of be assigned to neuronal assemblage c from the ego,
this part is constantly regulated by the ω cathexes. But it is and this may be described as thought interest, and be
also clear that at the same time enough e is economized proportionate to the affective interest where that may
by this to make the declarative process profitable as a have developed. The question is only whether there
whole. Otherwise, all the e, which is finally needed for are ψ processes of a primary nature for which the e sup-
action, would be given off at the points of M action plied from φ is sufficient or whether the cathexis of
during the course of its passage. Thus the secondary sensory e is automatically supplemented by a ω contri-
process is a repetition of the original ψ passage of e, at a bution (attention) which alone makes a ψ secondary
lower level, with smaller quantities. process possible (see below). This question must
“With e’s even smaller,” it will be objected, remain an open one, though it may perhaps be
decided if it is specially applied to some psychological
than those that ordinarily pass through in the ψ system? facts.
How can it be arranged that such small e’s shall have It is an important fact that ψ primary processes, such as
open to them pathways which, after all, are only traver-
have been biologically suppressed in the course of ψ
sable by larger ones that ψ as a rule receives?
development, are daily presented to us during sleep. A
The only possible reply is that this must be a mechanical second fact of the same importance is that the patho-
result of the side cathexes. We must conclude that logical mechanisms which are revealed in some mental
matters stand in such a way that when there is a side disorders by the most careful analysis have the greatest
cathexis small e’s flow through the resistances which similarity to dream processes. The most important con-
would ordinarily be traversed only by high ones. The clusions follow from this comparison, which will be
side cathexis as it were binds a quota of the free e trans- enlarged on later.
mitted through the system. First, the fact of sleep must be brought into our
There is a further condition that thought must satisfy. theory. It is easy to overlook the fact that sleep, no less
It must make no essential change in the resistances than energy balance and hydration, is a bodily need
made by the primary processes; otherwise, indeed, it which periodically intrudes on consciousness in the
would falsify the traces of real experiences. Of this con- form of a drive: the need for sleep. Here, once more,
dition it is enough to remark that consolidation is prob- we see the critical role that need prioritization plays.
ably the result of a single quantity of e and that cathexis, Babies sleep so long as they are not tormented by any
though very powerful at the moment, nevertheless does other physical need or external stimulus, including
not leave any comparable lasting effect behind it. The sensory affects (hunger and cold from wetting). They
small e’s that pass during thought cannot in general go to sleep only when these other needs are satisfied
prevail against consolidation. – although, of course, eventually the e from the need
 NEUROPSYCHOANALYSIS 25

for sleep attracts such high ω that it must be prioritized. predictive pathways giving expression to the “expected
Adults, too, fall asleep more easily after dining and copu- context” of the drive to sleep, which implies that the tol-
lating. Accordingly, the precondition of sleep is a lower- erance of exteroceptive e will be greatly increased.66 This
ing of the other endogenous loads pressing for access to implies also that indications of reality (and therefore per-
the ψ nucleus, which makes the secondary function ceptual consciousness) are terminated during sleep. As
superfluous. It becomes superfluous because the need we have conjectured, along with this the innervation
for sleep demands, in the first instance, withdrawal of of attention comes to a stop as well. It is from here,
ψ cathexis from M and φ (as occurred with the resting too, that the enigma of hypnosis may be approached.
state). In sleep an individual is in the ideal state of com- The apparent unexcitability of perceptual consciousness
plete (or nearly complete) automaticity, rid of the need in deep hypnosis must rest on this withdrawal of the
for voluntary ψ activity – including voluntary thought cathexis of attention.
activity. Thus, by a withdrawal process, the counterpart of
In adults, this voluntary cathexis is collected in the directing attention, the ψ pallium excludes φ
“ego”;65 we may therefore assume that it is the unload- impressions so long as it itself is uncathected.
ing of the ego which determines and characterises But what is strangest of all is that during sleep ψ pro-
sleep onset. And here, as is immediately clear, we have cesses occur – dreams – which have many characteristics
the precondition of mental primary processes. that seem to contradict the drive to sleep.
It is not certain whether in adults the ego is comple-
tely relieved of its burden in sleep. In any case it with-
(20) An analysis of dreams
draws an enormous number of its cathexes, which,
however, are restored on awakening, immediately and Dreams exhibit every transition to the waking state and
without trouble. This contradicts none of our presuppo- to a mixture with normal ψ processes; yet it is easy to sift
sitions; but it draws attention to the fact that we must out what is genuinely in the nature of a dream.
assume that the synaptic weightings (the resistances) (a) Dreams are devoid of motor discharge. We are
between neurons which are properly linked remain unal- paralyzed in dreams.
tered during sleep. Or do they? The easiest explanation of this characteristic is the
Sleep is characterised above all by motor paralysis absence of spinal M precathexis, which renders φ
(paralysis of externally directed M). The “will” is aban- signals largely irrelevant. As I explained before, action
doned. This is due to the fact that in sleep spinal tonus and perception are two sides of the same coin; they
is relaxed; and this applies all the more during REM may accordingly both be terminated in essentially the
sleep. However internally directed M persists during same way, through modulation of expected ω. In some
sleep together with the endogenous sources of ψ dream states (e.g. somnambulism) movement is not
excitation. excluded. This cannot be the most essential character-
It is a highly interesting fact that the state of sleep istic of dreams.
begins and is evoked by a closure of those sense (b) The ideational connections in dreams are partly
organs that are capable of being closed. Perceptions nonsensical, partly feeble-minded, or even meaningless
should not be made during sleep, and nothing disturbs or strangely crazy.
sleep more than the emergence of sense impressions: This latter characteristic is explained by the fact that
signals entering from φ and cathected by ψ. This is con- in dreams a compulsion to associate prevails, as it does
sistent with our hypothesis that during day-time a tonic, primarily in mental life generally. Two neuronal assem-
even though displaceable, cathexis (attention) is sent blages that are present simultaneously must, so it
into the pallium neurons, which palpates expected pat- seems, be brought into connection. The relationship
terns of φ activation, so that the carrying out of the ψ between this compulsion and the default SEEKING
secondary processes is made possible with the adjust- drive may now be determined. It raises an important
ment of expected ω. The salient φ neurons are, as it question: how can SEEKING be prioritized simul-
were, precathected. Pallium cathexis is, however, ulti- taneously with sleep? The answer is that the drive to
mately contingent on the deeper predictions emanating sleep takes priority only at sleep onset. This prioritization
from the ψ nucleus, which are in turn activated from the does not mean that SEEKING activity ceases; it means
ω decision triangle as prioritized affect. If ψ withdraws only that it becomes automatized. If, then, SEEKING is
these pallium cathexes (i.e. when the need for sleep is subsequently re-prioritized while ω and M are dec-
prioritized) the φ signals fall upon largely uncathected athected, sleep need not come to an end; the manage-
neurons and the possibility of e transmission becomes ment of the need for sleep then becomes
slight. That is, ω values are increased in the centrifugal automatized. (This is of course facilitated by the fact
26 M. SOLMS

that the need for sleep decreases as sleep itself it, is an endogenously generated property. Endogenous
proceeds.) sources of e persist in sleep, and so they arouse affects –
The two other characteristics, which are in fact iden- which renders salient their attendant instinctual and
tical, show that a part of the dreamer’s mental experi- learnt predictions: the expected context. Confusion on
ences have been forgotten. Actually, indeed, all the this score is the perennial price that cognitive scientists
biological experiences which ordinarily inhibit the pay for believing that consciousness flows in through
primary process are forgotten, and this is owing to the the senses. The biological purpose of dreaming is now
lack of ego cathexis. The senselessness and illogicality clear: hallucinatory resolution of affective demands is
of dreams are probably to be attributed to this very required for the preservation of the state of sleep. This
same characteristic. It seems, however, that activity in explanation is further supported by the circumstance
the ψ nucleus does not cease with sleep onset, and that in dreams the vividness of the hallucination is
that it continues to wander through the deeper directly proportionate to the salience – that is the ω
pallium layers. If the ego were completely unloaded, cathexis – of the idea concerned. This indicates that it
sleep would necessarily be dreamless (and indeed, is ω modulation that determines hallucination. If a per-
thoughtless). At this point, to avoid unnecessary rep- ception comes from φ in waking life, it is no doubt
etition, I must ask you to re-read the final paragraph of made clearer by ψ cathexis (interest), but not more
Section 14, concerning the memory consolidation func- raw; it does not alter its affective characteristic. As we
tion of the resting state. know, this always comes from within. But still we face
(c) Dream ideas are of an hallucinatory kind; they the question: what is the source of the changing e
awaken consciousness and meet with belief. values in the outer layers of the ψ pallium that
This is the most important mental characteristic of demand ω modulation during sleep?
sleep. It appears at once when there are alternating (d) The aim and sense of dreams (of normal ones, at
spells of sleeping and waking. One shuts one’s eyes all events) can be established with reasonable certainty.
and hallucinates; one opens them and thinks in words. Dreams are wish-fulfillments – that is, primary process
There are several explanations of the hallucinatory SEEKING following upon experiences of satisfaction of
nature of dream cathexes. In the first place, it might be the various drive demands (i.e. “wanting”); and they
supposed that when the prioritization of action as a are only not recognized as such because the release of
means for reducing endogenous e during waking life pleasure (the reproduction of pleasurable discharges of
ceases with sleep onset, the alternative pathway (per- F) in them is slight, because in general they run their
ception)67 must be given priority. The only argument course almost without effect (without motor action).
against this is the consideration that the φ neurons, by That this is their nature is, however, easily shown; the
the fact of being uncathected from ψ, should be pro- evidence for the hypothesis that dreaming is driven by
tected from generating e signals. But this overlooks dopaminergic SEEKING is now overwhelming.68 Combin-
two points. A withdrawal of ψ cathexis implies only a ing this fact with our insights concerning default-mode
cessation of its secondary process. And, moreover, “mind wandering” in the service of consolidation (sim-
primary processes in the ψ pallium are unconscious plification of the generative model), which characterises
unless and until they generate indications of reality – the ego’s resting state when it is unoccupied by external
by way of the ω mechanisms described in Section 16. tasks, we arrive at the following hypothesis.69 Default-
This implies that the primary task of the nervous mode pruning of recently acquired synaptic connections
system to suppress surprisal, which must continue (i.e. recent Qη updates) will come upon traces of new
throughout sleep, gives rise in the first instance only to learning in the ψ pallium which contradict repressed ψ
uncontrolled “hallucinations” (i.e. unconscious updating nucleus predictions. Recall that repressed predictions
of the predictive model). Secondly, therefore, we revert cannot be updated. These mismatches must be resolved
to the nature of the primary process and point out through disambiguation: hence the mental work of
that the primary activation of a perceptual memory dreaming; which, it follows from the nature of repres-
trace is always an “hallucination,” and that only inhi- sion, must aim to explain away the mismatching,
bition of the ego has taught us to palpate the expected exogenously acquired knowledge. That is how primary-
ω assigned to the incoming e signals. It is this secondary process wishful cathexes during sleep give rise to halluci-
process that yields the indications of reality which natory consciousness. It is now easy to see why dreams
bestow “belief” (i.e. confidence) in the resultant percep- so frequently are unpleasurable: repressed (illegitimately
tion. How, then, does it happen that hallucinations automatized) predictions can fail to resolve any one of
during sleep awaken consciousness and meet with the emotional drives enumerated in Section 4; and situ-
belief? The answer is to recall that consciousness, all of ations evoking FEAR, RAGE, PANIC/GRIEF, etc., are just as
 NEUROPSYCHOANALYSIS 27

likely to be encountered during virtual “foraging” as they Between this thought and the propyl (A) there is an associ-
are in waking life. Failure to manage these affects would ation from sexual chemistry (B), which Freud had discussed
mean a failure of the biological function of dreaming with his friend Fliess, in the course of which he had
which is in service of the drive to sleep. brought Freud’s special attention to trimethylamine (C).
(e) It is noteworthy how poorly dreams are remem- This now becomes conscious owing to it drawing e
bered and how little harm they do as compared with signals from both sides (from B and D).
other primary processes. But this is easily explained It is very puzzling that neither the intermediate link (B
from the fact that, for the most part, they follow old – sexual chemistry) nor the diversionary idea (D – the
synaptic connections and thus make no change in sexual nature of the illness) becomes conscious as well,
them and, moreover, from the fact that the synaptic and an explanation of this is called for. One would
pruning that characterises default mode activity suppose that the ω cathexis of predictions B or D
(especially regarding the repressed) serves forgetting alone is not ambiguated enough to make its way
more than learning. In addition, owing largely to the through to a conscious hallucination (i.e. to disambigua-
paralysis of motility, dreams do not leave traces of tion of the ψ and e signals), but that C, drawing error
fresh φ experiences. signals from both sides, would bring it about. In the
(f) It remains interesting that consciousness in dreams example chosen, however, D (the sexual nature of the
furnishes quality with as little trouble as in waking life. illness) was certainly as intense as A (the propyl injec-
This shows once more that consciousness is primarily tion) and the derivative of these two, the chemical
affective; it does not cling to the ego but can become formula (C), was immensely vivid. The puzzle about
an addition to any ω (precision optimization) process. unconscious intermediate links applies equally to
This warns us, in short, against possibly identifying id waking thought, where similar events are of daily occur-
processes with unconscious ones! rence. But what remains characteristic of dreams is the
If, when the memory of a dream is retained, we ease with which e is displaced in them and accordingly
inquire into its content, we find that the meaning of the replacement of B by C which becomes superior to
dreams as wish-fulfillments is concealed by a number it quantitatively. This can only be due to a defensive
of ψ processes (defences): all of which are met with use of precision modulation, to reduce unpleasure (on
once more in mental disorders and characterise the the model of distressing memories generally, discussed
latter’s pathological nature. in Section 13) and therefore to preserve sleep.
Similarly with the fulfillment of wishes in dreams gen-
erally. What happens is not, for instance, that the wish
(21) Dream consciousness
becomes conscious and its fulfillment is then halluci-
Consciousness of dream ideas is above all discontinuous. nated, but only the latter: the intermediate link is left
What becomes conscious is not a whole succession of to be inferred. It has quite certainly been passed
associations, but only separate stopping points in it. through, but without being able to develop declara-
Between these there lie unconscious intermediate links tively. It is evident, however, that the predictive cathexis
which we can easily discover when we are awake. If of the wishful idea cannot possibly be stronger than the
we investigate the reasons for this skipping, here is motive impelling it. Thus the passage of ideas in dreams
what we find. Let A be a dream idea which has takes place in accordance with e; but it is the ω weight-
become conscious and which leads to idea B. (I will set ing rather than the base value which decides the ques-
the symbols for conscious ideas in bold type and the tion of it becoming conscious.
unconscious ones not.) Instead of B, C is found in con- It should be inferred from dream processes, as from
sciousness, and this is because it lies on the pathway perception, that consciousness arises during the modu-
between B and a D cathexis which is simultaneously lation of ω; that is, it is not awakened by a monotonous
present. Thus there is a diversion brought about by a cathexis.71 It should further be inferred, as we know, that
simultaneous cathexis, of a different kind, which, is not consciousness does not attach to deeply consolidated
itself conscious. For that reason, then, C has taken the predictions, since they are routinely assigned high pre-
place of B, though B fits in better with the connection cision values. It is to their uncertain outcome that con-
of thought, with the wish-fulfillment. sciousness attaches – and this occurs progressively
For instance, in the famous “specimen dream,”70 Otto towards the periphery of the predictive cascade. Thus,
has given an injection of propyl to Irma. The dreamer we can only conclude as we did before that repression
(Freud) then sees trimethylamine before him very vividly, and defence are two separate processes. Repression
hallucinated as a formula. Explanation: The thought simul- arises from illegitimate consolidation (long-term resist-
taneously present is the sexual nature of Irma’s illness (D). ance to change) while defence arises from short-term
28 M. SOLMS

precision modulation, diverting the predictive cascade 5. This “orphan” sentence is a good example of the chal-
from its automatic outcome – which if left to its own lenges posed by Freud’s dense text. In the original
version, the sentence reads: “N and Qη – Similar exper-
devices can only bring intense surprisal in its wake (it
iments are now frequent.” There, N denoted “neurons”
being the outcome of an illegitimate prediction). We and Qη denoted “quantity of an intercellular order of
must in addition take account of the fact that surprisal magnitude.” What Freud meant to convey was that he
arising from repression is felt affectively (near the core was not the only scientist using these two basic ideas
of the hierarchy) whereas that arising from defence is to conceptualize the dynamics of the nervous system;
of a more cognitive nature. Defence causes errors of and I am trying to convey the same, in relation to
neurons and information theory. – Incidentally, the
reasoning and logic rather than the shattering of core
neuron had only just been discovered at the time of
beliefs. Freud’s writing the “Project.”
The peculiarities of dream consciousness may, in con- 6. Freud (1915, pp. 121–122) defined drive as “the psychi-
clusion, be added to the evidence for our view that the cal representative of the stimuli originating from
id should not be conflated with the repressed. The within the organism and reaching the mind, as a
measure of the demand made upon the mind for work
repressed consists in life-long beliefs which, when
in consequence of its connection with the body.” Here,
enacted illegitimately (virtually or really) can only yield I am equating “drive” with F, bearing in mind that the
errors which will be felt in the brainstem core. Id con- “free energy” (F ) within a system is the energy that is
sciousness, feeling, then punishes the unconscious ego not currently performing useful work. Helmholtz con-
for its errors, by making renewed demands for work. trasted it with “bound energy.”
Reconsolidation (cognitive consciousness, which must 7. The basic principle in question is the free energy prin-
ciple, which is formulated in Friston’s law: “All the quan-
be tolerated by the ego) enables it to mend its ways.
tities that can change; i.e. that are part of the system, will
Since this solution is excluded in the case of repressed change to minimise free energy” (Friston & Stephan,
beliefs, defence – the suppression of affect through mis- 2007). This law (like the law of affect, described below)
leading cognitions – is its only recourse. is probabilistic; it applies only in the average case, i.e.
April 17th, 2020 over sufficient time periods. See Friston (2009) for an
introduction to this unifying principle of brain
functioning.
8. The “principle of neuronal inertia” is the theoretical pre-
Notes cursor of Freud’s “death drive.“ This is ironic because
homeostasis is the grounding mechanism of all life.
1. Friston’s work builds directly upon Helmholtz’s
9. The (Bayesian) terms “prior” and “posterior” here refer to
insights, not least concerning the conservation of
the fact that the actions and perceptions of a self-organiz-
energy. The first law of thermodynamics states that
ing system entail experiments which test hypotheses gen-
energy is never lost or created, it is only transformed.
erated by its predictive model. Prior hypotheses are
The second law states that energy always dissipates
supplanted by posterior ones which take account of the
during natural processes. The driving mechanism of
outcome of each experiment. See Hohwy (2013) and
self-organizing (e.g. living) systems is to resist this dis-
Clark (2015) for accessible accounts of the “Bayesian brain.”
sipative tendency, by minimizing free energy. That is
10. Freud conceptualized this compromise, his “constancy
the free energy principle.
principle,” as a special case of Fechner’s “tendency
2. Please note: the Greek letters used here are purely con-
towards stability.” – The simplest neural contrivance
ventional; they have no literal meaning. Also, although
that gives expression to this imperative is the default-
the symbols denote concepts that are equivalent to
mode SEEKING system, described below, which
those that Freud used, they are not identical with
engages proactively with sources of uncertainty (see
them; this is because the concepts have been substan-
Panksepp, 1998).
tively updated.
11. Cathexis will be further defined below.
3. “Free energy.” There are various types of free energy. F
12. F can also be defined roughly as the sum of squared pre-
denotes “variational” free energy, or Friston free
diction error.
energy for short. This is analogous to Helmholtz free
13. See Mesulam (2000).
energy, where there is an information exchange as
14. Cf. Freud’s concept of “contact barriers,” which is here
opposed to a thermodynamic exchange between a
replaced by a broader conception of resistance.
system and its environment. (See Friston, 2009.)
15. See Felleman and Van Essen (1991).
4. In Shannon’s (1948) mathematical treatment of “infor-
16. See Tozzi et al. (2016): “Maximizing mutual information
mation,” the less predictable an event is, the more infor-
[i.e. minimizing demand for information flow] and mini-
mation it carries. The average information of a system is
mizing metabolic costs are two sides of the same coin.”
its entropy (a concept formally related to but not identi-
17. These terms are used by Hohwy (2013) and Clark (2015)
cal with free energy). Therefore, the less information
respectively, but by no means only by them.
required by a self-organizing system (i.e., the lower its
18. See Friston (2005) for a review.
uncertainty, the fewer yes/no questions it needs to
19. I am disregarding the interneurons here.
ask), the better for the system. Uncertainty is, for
20. “Neurons that fire together, wire together” (Hebb, 1949).
obvious reasons, dangerous for biological systems.
 NEUROPSYCHOANALYSIS 29

21. Freud linked this magnitude with Fechner’s law: “Sen- its meaning, its very existence entirely – even if in
sation varies with the logarithm of the intensity of the some contexts indirectly – from the apparatus-elicited
stimulus.” answers to yes or no questions, binary choices, bits. It
22. Technically, free energy may be decomposed into accu- from Bit symbolizes the idea that every item of the phys-
racy and complexity. Model evidence is the difference ical world has at bottom – at a very deep bottom, in
between accuracy and complexity, since models with most instances – an immaterial source and explanation.”
minimum F provide accurate explanations of data 39. See Kihlstrom (1996) and Bargh and Chartrand (1999) for
under complexity costs, which in turn means that redu- reviews of the literature on unconscious cognition.
cing model complexity improves model generalizability 40. Brentano (1874).
but at the cost of accuracy. The balance between accu- 41. Technically, as we know, the most salient signal is the
racy and complexity determines a model’s efficiency. In one which provides the greatest opportunity for mini-
biology, efficiency is everything. (In Bayesian terms: mizing F.
“likelihood” must be assessed in relation to “probability,” 42. And which, please recall, can be both grabbed and
to prevent over-fitting.) assigned.
23. The nondeclarative and declarative systems are equival- 43. “Freely mobile” cathexis is automatized.
ent to Freud’s unconscious and preconscious systems 44. Cf. Freud (1920, p. 25): “Consciousness arises instead of a
respectively. (The term “declarative” implies “capable memory trace.” Conversely, a memory trace arises
of becoming conscious.”) The pivotal topic of conscious- instead of consciousness; when consciousness ceases,
ness is discussed below, as is its relationship to plasticity. certainty is restored.
24. Cf. the concept of “working memory,” discussed further 45. See Pfaff (2005).
below. Mental work here refers to ψ cathexis in the face 46. See Merker (2007), Solms (2013).
of e. In physiological terms, a “cathected” ψ trace is 47. The law of affect states: “If a behaviour is consistently
subject to reconsolidation, which is predictive work in accompanied by pleasure it will increase, and if it is con-
progress. This mechanism will become clearer below, sistently accompanied by unpleasure it will decrease.”
when the concept of “precision” is introduced. This law is attributable to Panksepp (1998) who
25. Despite being so obvious, this simple mechanism has derived it from Thorndike’s (1911) law of effect.
enormous clinical ramifications. 48. The first scientist to arrive at this important insight was
26. Q in itself is “unknowable.” Fotopoulou (2013).
27. Cf. Wheeler (1990): “That which we call reality arises in 49. The terms “easy” and “hard” here refer to Chalmers
the last analysis from the posing of yes/no questions (1995). The hard problem is addressed in far greater
and the registering of equipment-evoked responses; in detail in Solms (2021).
short […] all things physical are information-theoretic 50. See the “Key to Abbreviations.” Please note that these
in origin.” quantities are vectors, apart from ω and F which are
28. I.e. through reconsolidation. scalars. The dot notation in the equations below
29. Error and prediction units are both assumed to be pyra- implies a dot product (i.e. matrix or vector
midal neurons, although they have different patterns of multiplication).
connectivity. In cortex, the former originate in supragra- 51. As stated above, these simplified equations effectively
nular layers and terminate (centripetally) in layer 4 spiny reduce free energy to the likelihood of a Gaussian distri-
stellate cells; the latter originate in infragranular layers bution. In fuller treatments, one would also need to con-
and target (centrifugally) infra and supragranular cells. sider hierarchical generative models (with precisions at
However, it is important to say that prediction units each level) and accommodate conditional uncertainty
are very far indeed from being limited to the cortex. about external states. Furthermore, the equations
30. This is post-synaptic modulation; see below. lump all sensory prediction errors together – including
31. Panksepp (1998). “Sensory affects” are discussed below. I the endogenous and exogenous modalities.
do not use Panksepp’s term for the interoceptive bodily 52. Technically, this is called a “gradient descent,” where the
affects because it can cause confusion: as we shall see, gradient is the rate of change of free energy with
all affects are homeostatic. precision.
32. As we shall see later, it demands engagement with our 53. Under our simplifying assumptions about the encoding
representations of the external world. of Bayesian beliefs.
33. Emotional needs, unlike bodily ones, pertain in large 54. Unlike Freud, I am using the term “nucleus” here
measure to other agents in the world, the behaviour of to denote what is conventionally called the “subpal-
which is far less predictable than that of inanimate lium.” I have already divided the prosencephalon
objects. – Please note: learning does not update into its telencephalic and diencephalic components.
reflexes and instincts; it supplements them. However, the diencephalic structure I discussed pre-
34. See Bowlby (1969). viously was the hypothalamus only, the role of
35. Consider the common phobias. which (as a source of endogenous e) is clarified
36. Panksepp (1998), Merker (2007). further in this section. The functional-anatomical situ-
37. White et al. (2017). ation is complicated: the diencephalon includes
38. The earlier quotation above from Wheeler (1990) is pre- nuclei (e.g. the lateral geniculate, mentioned before)
ceded by the following passage: “It from Bit. Otherwise which clearly belong to the φ system and others
put, every it – every particle, every field of force, even (e.g. the subthalamic nucleus) which are functionally
the spacetime continuum itself – derives its function, inextricable from the basal ganglia – i.e. the “nucleus”
30 M. SOLMS

of ψ which I am discussing here – and yet others (e.g. Bowlby, J. (1969). Attachment. Hogarth Press.
the intralaminar nuclei) which form part of the ω Brentano, F. (1874). Psychologie vom empirischen Standpunkte.
system. These complexities were not recognized in Duncker & Humbolt.
Freud’s times. Carhart-Harris, R., & Friston, K. (2010). The default-mode, ego-
55. I.e. affects arising from vegetative needs. See Note 31. functions and free-energy: A neurobiological account of
56. This apparently occurs in some psychosomatic diseases. Freudian ideas. Brain, 133(4), 1265–1283. https://doi.org/
The potential consequences of reducing precision on 10.1093/brain/awq010
strong emotional error signals are less catastrophic, in Chalmers, D. (1995). Facing up to the problem of conscious-
the short term at least, as we have long known from ness. Journal of Consciousness Studies, 2, 200–219.
“hysteria.” Clark, A. (2015). Surfing uncertainty: Prediction, action, and the
57. Hence “infantile amnesia,” which applies only to episo- embodied mind. Oxford University Press.
dic and semantic (declarative) memories. Ditchburn, R., & Ginsborg, B. (1952). Vision with a stabilized
58. Confidence in an e value is inversely proportional to that in retinal image. Nature, 170(4314), 36–37. https://doi.org/10.
its prior prediction, so if the error = 0, the prediction = 1. 1038/170036a0
59. Cf. the PANIC/GRIEF category described above. This does Ebisch, S., Salone, A., Ferri, F., De Berardis, D., Romani, G. L.,
not imply, as Freud thought, that the need for attach- Ferro, F. M., & Gallese, V. (2012). Out of touch with reality?
ment “leans upon” the need for nourishment (cf. “anacli- Social perception in first episode schizophrenia. Social
sis”); these are independent needs and both must be Cognitive and Affective Neuroscience, 8. https://doi.org/10.
met. 1093/scan/nss012
60. This corresponds to the distinction between “bound” Felleman, D., & Van Essen, D. (1991). Distributed hierarchical
and “freely mobile” cathexis, mentioned above. processing in the primate cerebral cortex. Cerebral Cortex,
61. Cf. the important paper on this by Carhart-Harris and 1(1), 1–47. https://doi.org/10.1093/cercor/1.1.1
Friston (2010). Fotopoulou, A. (2013). Beyond the reward principle:
62. See the classical descriptions of this by Riggs and Ratliff Consciousness as precision seeking. Neuropsychoanalysis,
(1951) and Ditchburn and Ginsborg (1952). 15(1), 33–38. https://doi.org/10.1080/15294145.2013.
63. See Rizzolatti and Craighero (2004) for a review. 10773715
64. Vittorio Gallese’s group describe what happens when Freud, S. (1900). The interpretation of dreams. In Standard
mirror-neuron activity is not inhibited in this way, as edition of the complete psychological works of Sigmund
occurs in schizophrenics who fail to distinguish the Freud, 4 & 5. Hogarth.
perceived object from themselves (Ebisch et al., Freud, S. (1915). Instincts and their vicissitudes. In Standard
2012). edition of the complete psychological works of Sigmund
65. In babies, ω modulation is entirely in the hands of the Freud, 14 (pp. 117–140). Hogarth.
midbrain decision triangle: in the “id.” Freud, S. (1917). A metapsychological supplement to the
66. Please note, this is not an homogenous cathexis. When theory of dreams. In Standard edition of the complete psycho-
sleep is prioritized it remains possible for ψ to “listen logical works of Sigmund Freud, 14 (pp. 219–235). Hogarth.
out” for salient e signals such as, for example, the Freud, S. (1920). Beyond the pleasure principle. In Standard
sound of a baby crying. Here the prediction (the wish) edition of the complete psychological works of Sigmund
that “the baby will not cry” is assigned a low ω value Freud, 18 (pp. 7–64). Hogarth.
so that the relevant e signal can attain a sufficiently Friston, K. (2005). A theory of cortical responses. Philosophical
high magnitude to cause re-prioritization of one’s Transactions of the Royal Society B: Biological Sciences, 360
needs from sleep to CARE. Likewise, a sufficiently high (1456), 815–836. https://doi.org/10.1098/rstb.2005.1622
e signal emanating from anywhere in φ will always be Friston, K. (2009). The free-energy principle: A rough guide to
capable of grabbing attention in the manner described the brain? Trends in Cognitive Sciences, 13(7), 293–301.
before. https://doi.org/10.1016/j.tics.2009.04.005
67. See the end of Section 3. Friston, K., & Stephan, K. (2007). Free-energy and the brain.
68. See Solms (2011) for a review. Synthese, 159(3), 417–458. https://doi.org/10.1007/s11229-
69. See Hobson and Friston (2012). 007-9237-y
70. Freud (1900). Hebb, D. (1949). The organization of behavior: A neuropsycholo-
71. Here we have a deep mechanical account of the prin- gical theory. Wiley.
ciple of constancy. Hobson, J. A., & Friston, K. J. (2012). Waking and dreaming con-
72. Although Freud (1920) later forgot this. sciousness: Neurobiological and functional considerations.
Progress in Neurobiology, 98(1), 82–98. https://doi.org/10.
1016/j.pneurobio.2012.05.003
Disclosure statement Hohwy, J. (2013). The predictive mind. Oxford University Press.
No potential conflict of interest was reported by the author(s). Kihlstrom, J. (1996). Perception without awareness of what is
perceived, learning without awareness of what is learned.
In M. Velmans (Ed.), The science of consciousness:
Psychological, neuropsychological and clinical reviews (pp.
References
23–46). Routledge.
Bargh, J., & Chartrand, T. (1999). The unbearable automaticity Luria, A. R. (1980). Higher cortical functions in man. Basic Books.
of being. American Psychologist, 54(7), 462–479. https://doi. Merker, B. (2007). Consciousness without a cerebral cortex: A
org/10.1037/0003-066X.54.7.462 challenge for neuroscience and medicine. Behavioral and
 NEUROPSYCHOANALYSIS 31

Brain Sciences, 30(1), 63–68. https://doi.org/10.1017/ when it was published in 1950 (in German, the English trans-
S0140525X07000891 lation appeared in 1953). In this work, Freud attempted to
Mesulam, M. M. (2000). Behavioral neuroanatomy: Largescale achieve what has since become the goal of an entire interdis-
networks, association cortex, frontal syndromes, the limbic ciplinary field called neuropsychoanalysis: namely, to integrate
system and hemispheric lateralization. In M. M. Mesulam our psychoanalytic and neuroscientific perspectives upon the
(Ed.), Principles of behavioral and cognitive neurology (2nd mental instrument. For this reason, I have found it worthwhile
ed., pp. 1–120). Oxford University Press. to try to update the “Project” in light of subsequent develop-
Panksepp, J. (1998). Affective neuroscience: The foundations of ments, particularly in affective and computational neuro-
human and animal emotions. Oxford University Press. science. I now hope that you will help me to further develop
Pfaff, D. (2005). Brain arousal and information theory. Harvard this “New Project.”
University Press. Freud’s original text was drafted in the context of an
Pribram, K., & Gill, M. (1976). Freud’s project re-assessed: Preface ongoing correspondence with his friend and colleague
to contemporary cognitive theory and neuropsychology. Basic Wilhelm Fliess. Two weeks after completing it, he submitted
Books. the draft to Fliess for comment, together with Part II, while con-
Riggs, L., & Ratliff, F. (1951). Visual acuity and the normal tremor tinuing with the drafting of Part III. Fliess’s side of the corre-
of the eyes. Science, 114(2949), 17–18. https://doi.org/10. spondence has not survived but, in response to his
1126/science.114.2949.17 comments, Freud revised his “General Scheme” fundamentally.
Rizzolatti, G., & Craighero, L. (2004). The mirror-neuron system. He even went so far, in a letter dated January 1st, 1896, to tell
Annual Review of Neuroscience, 27(1), 169–192. https://doi. Fliess that he had moved the hypothetical ω system of neurons
org/10.1146/annurev.neuro.27.070203.144230 from its deep position in the processing stream to a slot
Sacks, O. (1984). A leg to stand on. Simon & Schuster. between the φ and ψ systems, thereby paving the way for
Shannon, C. (1948). A mathematical theory of communication. his subsequent decision to combine the metapsychological
Bell System Technical Journal, 27(3), 379–423. https://doi.org/ systems Pcpt. (φ) and Cs. (ω) into a single system Pcpt.-Cs.
10.1002/j.1538-7305.1948.tb01338.x (Freud, 1917).
Solms, M. (2011). Neurobiology and the neurological basis of In my view, that was a mistake. By conflating perception
dreaming. In P. Montagna & S. Chokroverty (Eds.), with consciousness, Freud lost the opportunity for recognizing
Handbook of clinical neurology, 98 (3rd series) Sleep disorders, that the final arbiter of feeling was located not in the cortex but
Part 1 (pp. 519–544). Elsevier. much deeper within the system, even deeper than the “ego”
Solms, M. (2013). The conscious id. Neuropsychoanalysis, 15(1), nucleus, in what he would later call the “id” (see Solms,
5–19. https://doi.org/10.1080/15294145.2013.10773711 2013). Notwithstanding this unfortunate decision, the 1896
Solms, M. (2019). The hard problem of consciousness and the revision had the enormous advantage of allowing Freud to
free energy principle. Frontiers in Psychology, 9, 2714. recognize that all “mental energy” is generated endogenously;
https://doi.org/10.3389/fpsyg.2018.02714 none of it flows in through the senses.72
Solms, M. (2021). The hidden spring: A journey to the source of I am submitting my “New Project” to you in the same spirit
consciousness. Norton. that Freud submitted his “Project” to Fliess. It is a rough
Solms, M., & Friston, K. (2018). How and why consciousness working draft, with many loose threads and some gaping
arises: Some considerations from physics and physiology. holes. I hope that you will help me to tie up the threads and
Journal of Consciousness Studies, 25, 202–238. begin filling the holes. I am inviting you, in particular, to do
Thorndike, E. (1911). Animal intelligence. Macmillan. so because of your considerable expertise in one (or more)
Tozzi, A., Zare, M., & Benasich, A. (2016). New perspectives on of the specialist topics trenched upon in my text, sometimes
spontaneous brain activity: Dynamic networks and energy in amateurish fashion. Feel free to comment on only those
matter. Frontiers in Human Neuroscience, 10. https://doi. topics that concern or interest you. There is every reason to
org/10.3389/fnhum.2016.00247 believe that you can assist where I have gone astray; so that
Wheeler, A. (1990). A journey into gravity and spacetime. W.H. we may collectively move closer to completing the task of fur-
Freeman. nishing, as Freud put it, “a psychology that shall be a natural
White, B., Berg, D., Kan, J., Marino, R. A., Itti, L., & Munoz, D. P. science.”
(2017). Superior colliculus neurons encode a visual saliency In order to follow the sometimes strange turns of my paper,
map during free viewing of natural dynamic video. Nature I invite you to consider it in tandem with Freud’s original
Communications, 8(1), 14263. https://doi.org/10.1038/ “Project,” as I have keyed my text to his, sentence for sentence
ncomms14263 where possible, and paragraph for paragraph where close
paraphrasing was impossible. The editors are sending you a
document with “tracked changes” to the original version,
showing all the revisions and additions I have made. You will
Appendix. Letter of June 25th 2020 see there that Freud’s “Project” was written in a highly con-
densed style. He expanded this draft in his later metapsycholo-
Dear Peer Commentators, gical papers. My forthcoming (2021) book provides an
equivalent expansion of this “New Project.”
Sigmund Freud sketched Part I of his momentous “Project for a I very much look forward to your comments.
Scientific Psychology” over a period of two weeks, between With sincere thanks for accepting the invitation, and with all
September 10th and 25th, 1895. Although it remained unpub- good wishes,
lished during his lifetime, the “Project” caused a huge splash Mark Solms