Epistemology, Algorithms, and Echo chambers

TECHNOLOGICAL SEDUCTION Epistemology, Ethics, & Algorithms DR. CHRIS RANALLI, VRIJE UNIVERSITEIT AMSTERDAM ‣ On Jan 6th 2019, Proud Boys member and QAnon conspiracy theorist Buckey Wolfe murdered his brother for being a “lizard” (someone part of the ‘Illuminati’) ‣ What’s worrying is the role of algorithm-driven tech in facilitating so-called extremist beliefs and actions AGENDA How does algorithm-driven tech facilitate extremist and other beliefs? - Epistemology Primer Q&A Web Personalization & Technological seduction Epistemic bubbles, Echo chambers Q&A EPISTEMOLOGY Normative evaluation of cognition - When is a belief rational vs. irrational? What should you believe? Which assertions are permitted? What does it take to turn a belief into knowledge? When is inquiry responsible? Which habits are virtuous intellectual habits? EPISTEMOLOGY Normative evaluation of cognition - When is a belief rational vs. irrational? What should you believe? Which assertions are permitted? What does it take to turn a belief into knowledge? When is inquiry responsible? Which habits are virtuous intellectual habits? EPISTEMOLOGY What are the good-making features of beliefs? Good-making features Belief Accuracy EPISTEMOLOGY What are the good-making features of beliefs? Good-making features Belief Support EPISTEMOLOGY What are the good-making features of actions? Good-making features Action Good EPISTEMOLOGY What are the good-making features of actions? Good-making features Action The right reasons EPISTEMOLOGY Ethical evaluation Action Epistemic evaluation Good-making features Good-making features Good and performed the right reasons Accurate and believed for the right reasons (supported) Belief EPISTEMOLOGY Rational? ‘There’s a fire down the fall’. Evidence EPISTEMOLOGY Rational ‘There’s a fire down the fall’. Evidence EPISTEMOLOGY Rational Evidence ‘There’s a fire down the fall’. “This is a test” Defeater EPISTEMOLOGY Rational ‘There’s a fire down the fall’. Evidence EPISTEMOLOGY Evidentialism: a belief that P is rational when and only when it is supported by the total available evidence. You should believe P just when P is supported by the total available evidence. EPISTEMOLOGY Reason-Action Principle: I should act as though P is the case just when it is rational for me to believe that P. Otherwise the action is unjustified. EPISTEMOLOGY Evidence-Action Principle: I should act as though P is the case just when my total available evidence supports believing that P. Otherwise the action is unjustified. EPISTEMOLOGY Jail case: 20 people in the jail courtyard, and 19/20 participated in an attack. The guard pulls one person from the group. “This person is guilty” EPISTEMOLOGY Evidence-Action Principle: I should act as though ‘X is Guilty’ just when my total available evidence supports believing that ‘X is Guilty’. Would the guard be justified in punishing the person? EPISTEMOLOGY The statistical evidence makes it highly likely that they’re guilty. What would be the responsible thing to do? Is the possibility of error itself a defeater? EPISTEMOLOGY Infallibilism: your belief that P is rational only when your support for P couldn’t lead you to error. Much harder to be rational! EPISTEMOLOGY Fallibilism: your belief that P can be rational even when your support for P could lead you to error. ‣ Prob(P|E) > .5 EPISTEMOLOGY Pragmatism: non-epistemic factors (factors not related to truth or accuracy) can affect whether one’s belief is rational. The stakes are high! This causes the standard for ‘rationality’ to go up for belief and thus action. EPISTEMOLOGY Case A: “He’s guilty”; statistical evidence Justified? X gets a warning Case B: “He’s guilty”; statistical evidence X gets punished Justified? EPISTEMOLOGY Theoretical case: Detective believes that A is the criminal because their DNA was found on the knife used to harm B, and B’s DNA was found under their shoe. ‣ Detective doesn’t understand why E supports A’s being the criminal EPISTEMOLOGY Evidentialism #2: you are rational in believing that P when and only when your total available evidence E supports P at that time and you understand why E supports P. EPISTEMOLOGY Evidentialism #3: you are rational in believing that P when and only when your total available evidence E seems to you to support P at that time. Subjective evidentialism EPISTEMOLOGY Problems: Subjective evidentialism justifies counterintuitive verdicts: ‣ myside-biased beliefs can come out as rational. ‣ Racist and ‘extreme’ beliefs can come out as a rational. Q&A WEB PERSONALIZATION Social media personalizes ads based on your searches and content based on user history 2.7 Billion Facebook users A lot of people use Facebook for news, but don’t know how or why it works the way it does. The Daily You: Personalised results web searches Follower/Friend suggestions/Hashags Personalised Newsfeeds Recommended consumer products/services Recommended sources/sites/ideas Filter Bubble: “Google doesn’t tell you who it thinks you are or why it’s showing you the results you’re seeing. You don’t know if its assumptions about you are right or wrong — and you might not even know it’s making assumptions about you in the first place . . . From within the bubble, it’s nearly impossible to see how biased it is” (2011, 10). Epistemic bubble: excludes sources of evidence. Echo chamber: discredits sources of evidence and re-affi rms sources of evidence. Objection: What’s so special about social media and the internet? ‣ Social connection: we tend to seek out people like ourselves. This reinforces our values and social identities. (Similarity-attraction effect’). ‣ Don’t our social groups filter information? ‣ Group ignorance Response: No. Filter bubbles are different. ‣ Hyper-personalized: traditional media isn’t tailored to an individual person. ‣ “Stealthy ignorance”: many users don’t know that they are in a bubble or how to get out of it. ‣ Lack of transparency: the information that is made available to the user depends upon past online behavior, and this behavior needn’t be transparent to the user. Technological seduction 70% of watch-time is driven by the recommender algorithm Technological seduction 70% of watch-time is driven by the recommender algorithm Technological seduction Top-Down Tech Seduction: •Design/UX/Presentation •Design: Narrative framing •Nudges the user Bottom-Up Tech Seduction: •Aggregated user data •Digital footprint EPISTEMIC BUBBLES Epistemic bubble: a socialepistemic environment that fails to be adequately coverage-reliable because it excludes sources of information or evidence. Epistemic bubble: a socialepistemic environment that fails to be adequately coverage-reliable because it excludes sources of information or evidence. Coverage reliability: broad and more complete set of information that is relevant to the propositions you’re inquiring about. - - Broadness: larger epistemic community. Completeness: revealing all the evidence; including the higher-order evidence about whether it’s widely endorsed, controversial, etc. Relevance: adequate number of inquirers consider it seriously. HOW-TO: BUBBLES • Selective exposure: inadvertently excluding information by omitting it from one’s social-epistemic environment • Agential filtering: when a powerful agent or institution blocks or mitigates your access to information • Algorithmic filtering: when algorithms fi lter your information; profi ling. Bootstrapped corroboration Argument P1. When you evaluate evidence within an epistemic bubble, you’ll encounter more agreement in how you evaluate the evidence (bootstrapped corroboration). P2. If you’ll encounter more agreement in how you the evidence, then you’re likely to be much more confident about your beliefs than rationality permits. Hence, C. When you evaluate evidence within an epistemic bubble, you’re likely to be much more confident about your beliefs than rationality permits. Believe: “The headline is correct” Did I get more evidence for my belief? Truth-tracking Argument P1. The criteria for content filtering algorithms is whatever optimizes user attention and retention. P2. If the criteria for content filtering algorithms is whatever optimizes user attention and retention, then the criteria isn’t a truth-tracking Therefore, C. The criteria isn’t a truth-tracking process— instead, it tracks what is interesting, likeable, or comforting. - Defeater for belief? An objection “What if I usually get content from reliable sources?” “…restricting our information sources is not objectionable by itself, and it can even have benefits. ….if I consult one highly reliable media outlet on a regular basis, I’ll not only block out a lot of noise, I’ll also end up forming lots of true beliefs about critical issues. If I add other sources simply to avoid worries about the insulation of beliefs — without any regard for their reliability — I’ll end up out of an echo chamber, but far worse off as a knower” (Lackey 2018) ECHO CHAMBERS Echo chamber: an epistemic community in which there is a disparity in trust between ingroup and out-group members. - Agree on a core set of beliefs. - In-group trust is amplified; out-group distrust is amplified. Evidential Preemption Argument P1. Echo-chambers preemptively discredit out-group information. P2. Out-group information can be reliable. Therefore, C. Echo-chambers preemptively discredit information which can be reliable. An Objection Suppose you're in the good case. - Preemptively discrediting unreliable information. - Protecting your accurate beliefs! Couldn’t Echo chambers still be harmful? Echo chambers might indoctrinate their members Indoctrination: when an insulated source causes you close-mindedly believe what you do. What is close-minded belief? You believe P close-mindedly if and only if you believe P but are unwilling to consider seriously the relevant alternatives to P. Couldn’t Echo chambers still be harmful? Echo chambers indoctrinate their members Indoctrination: when an insulated source causes you close-mindedly believe what you do. What is close-minded belief? You believe P close-mindedly if and only if you believe P but are unwilling to consider seriously the relevant alternatives to P. ?? Q&A BIBLIOGRAPHY Alfano et al. (2018). Technological-seduction and Self-radicalization. Journal of the American Philosophical Association. Battaly, Heather (2018). Close-mindedness and Dogmatism. Episteme. Lackey, Jennifer (2018). True-story: Echo chambers are not the problem, op-ed in Morning consult. Lynch, Michael (2016). Our Digital Forms of Life, in The Internet of Us. Norton. Nguyen, C. Thi (2018). Echo-chambers and Epistemic Bubbles. Episteme. Ranalli, Chris (forthcoming) Close-minded Belief and Indoctrination. American Philosophical Quarterly.