2018 The New Digital Future for Welfare 1

THE NEW DIGITAL FUTURE FOR WELFARE: DEBTS WITHOUT LEGAL

PROOFS OR MORAL AUTHORITY?

TERRY CARNEY AO*

This article reviews Centrelink’s online compliance initiative (‘OCI’) to

determine whether the Senate Community Affairs References Committee was
right to recommend that Centrelink resume responsibility for obtaining all
information necessary for calculating working age payment debts based on
verifiable actual fortnightly earnings rather than on the basis of assumed
averages, or whether responsibility has always remained with Centrelink
when the person is unable to easily provide records. It argues that legal
responsibility ultimately has always rested with Centrelink in such cases and
outlines distributional justice and best practice reasons why the OCI system
should be brought into compliance with the law.

I INTRODUCTION

A new digital future for administration and administrative review is much discussed, with
Britain touted as a leader.1 Automation of decision-making through application of
machine learning algorithms is one way efficiency and accuracy is pursued,2 including
Australia’s online compliance intervention (‘OCI’) debt recovery system – colloquially
known as ‘robo-debt’ – which is one part of the government’s Better Management of the
Welfare System initiative3 projected to recover $2.1 billion of social security
‘overpayments’ over four years.4

* Emeritus Professor of Law, The University of Sydney (Eastern Avenue, University of Sydney, NSW
2006, Australia; fax: +61 2 9351 0200; email: [email protected]); Visiting Research
Professor, University of Technology Sydney.
1
Joe Tomlinson, ‘A Primer on the Digitisation of Administrative Tribunals’ (Primer, University of
Sheffield, 12 September 2017) <https://ssrn.com/abstract=3038090>.
2
Deven R Desai and Joshua A Kroll, ‘Trust but Verify: A Guide to Algorithms and the Law’ (2017)
Harvard Journal of Law & Technology forthcoming <https://ssrn.com/abstract=2959472>; Cary
Coglianese and David Lehr, ‘Regulating by Robot: Administrative Decision Making in the
Machine-Learning Era’ (2017) 105 Georgetown Law Journal 1147.
3
See Senate Community Affairs References Committee, Parliament of Australia, Design, Scope,
Cost-Benefit Analysis, Contracts Awarded and Implementation Associated with the Better
Management of the Social Welfare System Initiative (2017)
<http://www.aph.gov.au/Parliamentary_Business/Committees/Senate/Community_Affairs/SocialWe
lfareSystem/Report>.
4
Scott Morrison and Mathias Cormann, ‘Mid-year Economic and Fiscal Outlook 2016–17’
(Statement, December 2016) 5, 44, 189 <http://www.budget.gov.au/2016-
17/content/myefo/download/2016-17-MYEFO-combined.pdf>.
2 UNSW Law Journal Forum 2018

It is trite maths that statistical averages (whether means or medians) tell nothing about the
variability or otherwise of the underlying numbers from which averages are calculated.
Only if those underlying numbers do not vary at all is it possible to extrapolate from the
average a figure for any one of the component periods to which the average relates.
Otherwise the true underlying pattern may be as diverse as the experience of Australia’s
highly variable drought/flood pattern in the face of knowledge of ‘average’ yearly rainfall
figures. Yet precisely such a mathematical fault lies at the heart of the introduction from
July 2016 of the OCI machine-learning method for raising and recovering social security
overpayment debts. This extrapolates Australian Taxation Office (‘ATO’) data matching
information5 about the total amount and period over which employment income was
earned, and applies that average to each and every separate fortnightly rate calculation
period for working-age payments.6

ATO data-matching previously was very properly used to trigger further enquiries about
a portion of the approximately 300 000 discrepancies (and possible debts) identified
annually by the Department of Human Services (better known to the public as
Centrelink). Based on risk management profiling, Centrelink formerly selected around
seven per cent of discrepancies for manual review and enquiry, to obtain firm information
about actual earnings in each payment fortnight (whether provided by the person or from
invoking its compulsory powers to require employers to provide pay slip records, or
banks to disclose statements). From July 2016 the OCI scheme targets and raises debts in
every case where the person cannot disprove the possible overpayment (or its quantum),
such as by producing or obtaining copies of pay slips. The Ombudsman identified the
dramatic scale of the change, writing that ‘DHS estimates it will undertake approximately
783 000 interventions in 2016–17 compared to approximately 20 000 compliance
interventions per year under the previous manual process’.7 While the full pipeline effect
of the new system had yet to be fully felt due to the lag in review processing, debt cases
lodged with the Administrative Appeals Tribunal (‘AAT’) increased by 28.5 per cent in
the first full year of the OCI scheme in 2016–17, compared to the previous year (rising
from 3387 to 4354).8

Why is this of legal, policy and moral interest? It is of interest to the law because, as
argued below, the so-called ‘practical onus’ to establish a debt and its size continues to
remain with Centrelink; the failure of a person to ‘disprove’ the possibility of a debt is

5
Collected and exchanged between the ATO and Centrelink pursuant to the Data-Matching Program
(Assistance and Tax) Act 1990 (Cth).
6
For an overview, see Kate Galloway, ‘Big Data: A Case Study of Disruption and Government
Power’ (2017) 42 Alternative Law Journal 89, 93–4. The scheme was not applied to pensions such
as the Age Pension, where pension rates are calculated on the basis of any changes in the rate of
annual income of the pensioner (to be reported within 14 days of any change). It is speculated that
the initial focus on working age payments was based on political considerations.
7
Richard Glenn, Acting Commonwealth Ombudsman, ‘Centrelink’s Automated Debt Raising and
Recovery System’ (Commonwealth Ombudsman Report No 2, April 2017) 5 [2.4]
<http://www.ombudsman.gov.au/__data/assets/pdf_file/0022/43528/Report-Centrelinks-automated-
debt-raising-and-recovery-system-April-2017.pdf>.
8
Senate Community Affairs References Committee, above n 3, 75.
2018 The New Digital Future for Welfare 3

not a legal foundation for a debt.9 This is not new. It was recognised in Centrelink’s pre-
OCI guideline which, while (somewhat dubiously) accepting averaging as a ‘last resort’,
correctly added ‘[t]he raising and recovery of debts must satisfy legislative requirements.
Evidence is required to support the claim that a legally recoverable debt exists’.10 And it
is also of legal interest because the nature of the issue (the monetary, moral and practical
implications of contending that a debt is owed) raises the bar for Centrelink in terms of its
discharge of that practical onus. This is what I loosely term the ‘rule of law’ challenge
(Part II below).

It is of wider policy interest because, in practice, when confronted with suggestions of

having an overpayment, often from up to seven years ago, the least literate, least
powerful, and most vulnerable alleged debtors will simply throw up their hands, assume
Centrelink knows that there really is a debt, and seek to pay it off as quickly as possible.
Alleged debtors do so even though the Ombudsman’s report demonstrated that most
debts calculated this way were greatly inflated, and that some were false (zero debts), and
they continue doing so because the otherwise worthy recommendations of the
Ombudsman and the Parliamentary Community Affairs Committee fail to correct the
fundamental legal error. It is of moral or ethical interest because Centrelink did not advise
recipients of the need to keep pay records for longer than six months, and because it is
difficult to see how the current system meets requirements of model-decision-making at
primary level or ‘model litigant’ obligations for internal and AAT review. Finally, it is of
interest because it is a test-bed for assessing the fitness for purpose of the administrative
review system (especially its normative impact on good primary decision-making) and as
a window into the digital future (Part III below).

The article briefly concludes in Part IV by arguing that the OCI system urgently be
rendered compliant with the law, lest it undermine public confidence in the positive
contribution machine learning can bring to better administration.

II THE RULE OF LAW CHALLENGE

Centrelink’s OCI radically changed the way overpayment debts are raised11 by purporting
to absolve Centrelink from its legal obligation to obtain sufficient information to found a
debt in the event that its ‘first instance’ contact with the recipient is unable to unearth
information about actual fortnightly earnings. As noted by the Ombudsman, the major
change was that Centrelink would ‘no longer’ exercise its statutory powers to obtain
wage records and that the ‘responsibility’ to obtain such information now lies with
applicants seeking to challenge a debt.12 Writing a little later, the Senate Community
Affairs References Committee challenged this, contending that

9
See also Peter Hanks, ‘Administrative Law and Welfare Rights: A 40-Year Story from Green v
Daniels to “Robot Debt Recovery”’ (2017) 89 AIAL Forum 1.
10
Department of Human Services, ‘Acceptable Documents for Verifying Income when Investigating
Debts’ (Operational Blueprint 107-02040020), quoted in Glenn, above n 7, 42, n 103.
11
Ibid 5–6, 31–8.
12
Ibid 5 [2.5].
4 UNSW Law Journal Forum 2018

6.13 It is a basic legal principle that in order to claim a debt, a debt must be proven to be
owed. The onus of proving a debt must remain with the department. This would include
verifying income data in order to calculate a debt. Where appropriate, verification can be
done with the assistance of income support payment recipients, but the final responsibility
must lie with the department. This would also preclude the practice of averaging income
data to manufacture a fortnightly income for the purposes of retrospectively calculating a
debt. …

6.16 The committee recommends the department resume full responsibility for
calculating verifiable debts (including manual checking) relating to income support
overpayments, which are based on actual fortnightly earnings and not an assumed
average.13

As will be shown, this as yet unimplemented recommendation is already the law.

For reasons now explained, while there may well be overpayments of undetermined
magnitude in some instances, Centrelink fails to reach the required state of satisfaction
about the existence or precise size of any such overpayments. Because these debts often
date from 2010 onwards (when adequate data was first retained), and because many
people will have held several casual jobs of varying durations across the relevant
financial years covered by the ATO records relied on by the OCI, few alleged debtors
will have retained pay slips for the relevant periods of employment, and many employers
will no longer be in business or able to supply them. Moreover, Centrelink’s then-website
advice (until late 2016) only advised keeping pay slip records for six months.14 So, as
argued in more detail below, legally it remains Centrelink’s obligation to obtain the
fortnightly pay information.15

There are several strands to explaining the legal error in OCI’s claim that responsibility
for proving the existence of a debt and its size can be shifted from Centrelink onto clients.
The starting point is that working age payments (such as Newstart allowance for the
unemployed) are calculated on gross (pre-tax) income in each fortnight,16 adjusting for
any fluctuations due to changes in casual or part-time earnings from one fortnight to the

13
Senate Community Affairs References Committee, above n 3, 109, [6.13]–[6.16] (emphasis in
original).
14
Glenn, above n 7, 13 n 22.
15
The reference in internal OCI training and policy statements to providing ‘assistance’ to debtors in
gathering the information and ‘navigating the system’ is sufficient in my opinion only in those few
situations where the person is experiencing difficulties in loading documents, or where information
is very readily to hand. See Glenn, above n 7, 16–19, especially the internal documents summarised
at n 34. See also ‘Centrelink Tells Frontline Staff to Only Help the Vulnerable and Send Everyone
Else Online’, news.com.au (online), 12 January 2017 <http://www.news.com.au/national/centrelink-
tells-frontline-staff-to-only-help-the-vulnerable-and-send-everyone-else-online/news-
story/322c992468ce5c91adc63c4b5797198f>. For the latest guidance, see Department of Human
Services, ‘Employment Income Confirmation’ (17 January 2018)
<https://www.humanservices.gov.au/individuals/subjects/employment-income-confirmation>.
16
This differs from family payments (annual taxable income) or pensions (gross annual amount of
income): see further Terry Carney, Social Security Law and Policy (Federation Press, 2006) ch 6.
2018 The New Digital Future for Welfare 5

next.17 There are threshold levels below which income does not affect the rate, and rules
about the amount of the reduction of the payment for each dollar of income above the
relevant threshold, as well as rules about accrual and depletion of ‘work credits’ which
smooth income across periods of episodic or broken employment.18 But there is no
statutory scope for a notional ‘average’ fortnightly income to substitute for the actual
fortnightly income, or for ignoring the effect of working credits in offsetting that income
in a given fortnight.

As explained by the Full Federal Court in McDonald v Director-General of Social

Security,19 there is no evidentiary ‘onus’ of proof borne by applicants or by Centrelink in
social security, where the test for decision-makers is reaching a state of ‘satisfaction’
about relevant matters.20 Or as Woodward J wrote:

It is true that facts may be peculiarly within the knowledge of a party to an issue, and a
failure by that party to produce evidence as to those facts may lead to an unfavourable
inference being drawn – but it is not helpful to categorise this common-sense approach to
evidence as an example of an evidential onus of proof. The same may be said of a case
where a good deal of evidence pointing in one direction is before the Tribunal, and any
intelligent observer could see that unless contrary material comes to light that is the way
the decision is likely to go. Putting such cases to one side there can be no evidential onus
of proof in proceedings before the AAT unless the relevant legislation provides for it …21

However, this is not the end of the matter, since there will often be what Robin Creyke
has described as a ‘practical’ onus22 to be distilled from the particular features of the
legislation. Woodward J spoke of it in terms that:

If the AAT finds itself in a state of uncertainty after considering all the available material,
unable to decide a question of fact either way on the balance of probabilities, it will be
necessary for it to analyse carefully the decision it is reviewing. If, for example, it is a
decision whether or not to cancel a pension in the light of changed circumstances, then it
has failed to achieve the statutory requirement of reaching a state of mind that the pension
should be cancelled. If, on the other hand, it is a decision, to be made in the light of fresh

17
The rate is set out in Rate Calculators such as Benefit Rate Calculator B: Social Security Act 1991
(Cth) s 1068, as applied by s 643.
18
Social Security Act 1991 (Cth) ss 1073D–1073J.
19
(1984) 1 FCR 354.
20
See ibid 356–8 (Woodward J); 366 (Northrop J); 368–9 (Jenkinson J). The principle was reaffirmed
by the Full Court in Re Australian Telecommunications Commission v Shirley Else Barker [1990]
FCA 489, [18]. As Jenkinson J expressed it, unlike courts who may ‘determine a matter against the
party on whom lies the onus of proof, and who fails to offer any proof … without further enquiry’,
decision-makers such as Centrelink or the AAT must still determine the question on the merits, even
where it ‘may find itself unpersuaded either that a circumstance exists or that it does not exist. (The
same may be said of a past or a future circumstance.)’: McDonald v Director-General of Social
Security (1984) 1 FCR 354, 369.
21
McDonald v Director-General of Social Security (1984) 1 FCR 354, 358 (Woodward J) (emphasis
added).
22
Re Russell and Secretary, Department of Families, Housing, Community Services and Indigenous
Affairs [2011] AATA 52, [35] (Senior Member Professor Creyke).
6 UNSW Law Journal Forum 2018

evidence, whether or not the pension should ever have been granted in the first place, then
it has failed to be satisfied that the person ever was permanently incapacitated for work.23

Or in the words of Jenkinson J, it means that:

The … administrative authority [ie, now Centrelink] will determine, by reference to the
substantive law, whether it is the existence or the non-existence of the circumstance which
is determinative of the question for decision.24

While care should always be taken not to misrepresent the meaning when reducing
passages like the above to simple propositions, this means that unless a decision-maker is
satisfied about pertinent facts or criteria, then the status quo prevails.25 Or to put it
another way, if a decision-maker is unable to be satisfied about key matters, then a
‘default’ outcome may result.

In the case of a decision to raise and recover a working age overpayment debt, the default
is that there is no debt unless Centrelink establishes its existence and size consistent with
the requirements of the fortnightly rate calculation.26 While the data-matching legislation
does authorise Centrelink to use data from matching to ‘take action’ to ‘recover an
overpayment of personal assistance’,27 subject to notifying the person in writing and
giving 28 days for response,28 this does not displace the normal rules about raising debts.
The only minor exception is for any overpayment arising during the 28-day notice
period.29 So is it sufficient for Centrelink to discharge that practical onus on the basis of
the ‘doubts’ or suspicions of there being an overpayment based on averaged fortnightly

23
McDonald v Director-General of Social Security (1984) 1 FCR 354, 358 (Woodward J) (emphasis
added).
24
Ibid 369 (Jenkinson J) (emphasis added).
25
Re ACT Department of Health and Nikolovski (1996) 42 ALD 599, 601, quoted in Re Waller and
Secretary, Department of Family, Community Services and Indigenous Affairs [2007] AATA 1902,
[23] (Member Frost), cited with approval by Senior Member Walsh in Re Parker and Secretary,
Department of Education, Employment and Workplace Relations [2011] AATA 98, [34]; Re Russell
and Secretary, Department of Families, Housing, Community Services and Indigenous Affairs
[2011] AATA 52, [35] (Senior Member Creyke).
26
This is because an overpayment debt only arises where specifically stipulated by legislation (see
Social Security Act 1991 (Cth) s 1222A(a); also A New Tax System (Family Assistance)
(Administration) Act 1999 (Cth) s 70) and thus a new decision has to be made by Centrelink to ‘raise
and recover’ any such overpayment debt, based on provisions such as Social Security Act 1991 (Cth)
s 1223 (covering the difference between the amount a person is entitled to receive and that paid); see
to the same effect Hanks, above n 9. While s 1224C catches any debt created by the Data-Matching
Program (Assistance and Tax) Act 1990 (Cth), the only fresh specific debt created is one arising
during a 28-day notice period, as discussed subsequently.
27
Data-Matching Program (Assistance and Tax) Act 1990 (Cth) s 10(1)(a)(iv).
28
Data-Matching Program (Assistance and Tax) Act 1990 (Cth) s 11(1)(d), (e).
29
Subsection 11(6) of the Data-Matching Program (Assistance and Tax) Act 1990 (Cth) provides only
for the overpayment during this period to be a debt where the notice foreshadowed ‘proposed action
to cancel or suspend, or reduce the rate or amount’ and the person ‘does not show cause why the
action should not be taken’: at s 11(6)(a)–(b). The subsection is very explicit in referring to the
amount arising ‘during the period specified in sub-para (1)(e)(ii) [being the 28-day notice period]
[as] a debt due to the Commonwealth’: at s 11(6).
2018 The New Digital Future for Welfare 7

income, coupled with the inability of the alleged debtor to upload or otherwise provide
evidence of actual earnings to contradict that material? As now explained there are two
reasons why I argue this is inadequate.

The first reason is that the ATO-based evidence and associated averaging calculation is
not necessarily indicative of any debt at all.30 Simple mathematics dictates that wherever
episodic and/or variable weekly earnings are in play and feed into the legislative context
of a requirement to determine a fortnightly earnings attributable to that fortnight, as
further moderated by any reduction of raw gross earnings figures by reference to the
‘earnings bank’ provisions,31 then almost any speculated figure between no debt and the
alleged debt amount is capable of being calculated. This is no hypothetical assertion; the
worked case examples cited by the Ombudsman revealed very substantial changes indeed
between the actual and any true overpayment, while some alleged debts disappeared
altogether.32 This likelihood is heightened not only because the ATO data from which
averages are calculated may cover a whole year or other extended number of weeks even
though the person did not work throughout the period, but also because a portion of ATO
recorded start and end dates of employment are unreliable. Of course, even if this were
not the case, information about average fortnightly incomes would never address the
question of the precise amount of income in a particular fortnight unless the income
remained constant and the period of its receipt was unbroken.

The second very weighty reason why Centrelink fails to avoid the ‘no debt default’ is that
its averaging evidence falls well short of the required standard under the High Court’s
Briginshaw principle.33 This principle maintains the same test of satisfaction (civil
balance of probabilities) but states that the strength of the evidence needed to reach that
level of satisfaction varies according to the ‘nature’ or the ‘effect’ of what it is that is to
be established. As Dixon J expressed himself in Briginshaw:

reasonable satisfaction is not a state of mind that is attained or established independently of

the nature and consequence of the fact or facts to be proved. The seriousness of an
allegation made, the inherent unlikelihood of an occurrence of a given description, or the
gravity of the consequences flowing from a particular finding are considerations which
must affect the answer to the question whether the issue has been proved to the reasonable

30
In its 2004 report on machine learning systems the Administrative Review Council referred to the
procedural fairness obligation that such decisions meet the usual test of being based on ‘some
rationally probative evidence’: Administrative Review Council, ‘Automated Assistance in
Administrative Decision-Making’ (Report, Administrative Review Council, 2004) 25, citing Deane
J in Minister for Immigration and Ethnic Affairs and Pochi (1980) 44 FLR 41, 62 and Australian
Broadcasting Tribunal v Bond (1990) 170 CLR 321, 366–9.
31
Social Security Act 1991 (Cth) ss 1067G-J1–1067G-J11, see especially ss 1067G-J1 [diagram],
1067G-J3 [method statement].
32
Glenn, above n 7, 8 n 15: ‘For example: Ms D’s debt was reduced from $2203.24 to $332.21, Mr S’s
debt from $3777.43 to zero, Ms H’s debt from $5874.53 to zero, Ms G’s debt from $2914.20 to
$610.07 and Ms B’s debt from $1441.64 to $267.51’.
33
Briginshaw v Briginshaw (1938) 60 CLR 336. See also the joint decision of the High Court in Neat
Holdings Pty Ltd v Karajan Holdings Pty Ltd (1992) 110 ALR 449, 449–50, and of the Full Federal
Court in Rana v University of South Australia [2007] FCAFC 188 [31].
8 UNSW Law Journal Forum 2018

satisfaction of the tribunal. In such matters ‘reasonable satisfaction’ should not be

produced by inexact proofs, indefinite testimony, or indirect inferences.34

As a matter of common understanding, an allegation of a debt has moral and practical

consequences for credit worthiness standing and ratings advice.35 The gravity of the
allegation should not be underestimated, including as a disclosable event for professional
admission purposes.36 It is well accepted therefore that establishment of a debt and its
size is a matter which leads to an ‘upwards variation’ in the strength of material
required.37 I conclude that it is inconceivable that OCI’s ‘indirect inferences’ of the
existence of some debt or its ‘inexact proofs’ of its quantum could possibly pass scrutiny
under the Briginshaw test.

III HOW SHOULD OCI DEBTS BE REVIEWED AND REFORMED?

There are really two main aspects of this: first, how to better deal with debts already
raised and sought to be recovered solely on the basis of an uncontradicted average; and
secondly, what could be done to avoid or at least reduce the legal, policy and ethical
deficiencies of continuing with such a debt recovery mode. Each will be considered in
turn below.

A What is the Role of the AAT in Dealing with Robo-debts?

The principal weakness in current handling of OCI debts reaching the review stage
(mandatory authorised review officer (‘ARO’) reconsideration and then up to two levels
of external AAT merits review38) is that Centrelink at best has failed to initiate its own
enquiries (as it once always did) by invoking its statutory powers to require employers or
banks to supply relevant records and information. At worst, primary and ARO decisions
misleadingly give alleged debtors the impression that the information for each fortnight
has been checked and is accurate (an impression compounded by most people having nil

34
Briginshaw v Briginshaw (1938) 60 CLR 336, 362 (Dixon J) (emphasis added).
35
This includes that held by agencies such as Dun & Bradstreet. See also ASX, Credit Ratings
<http://www.asx.com.au/products/bonds/credit-ratings.htm>. Centrelink argues that privacy
protections prevent secondary use of Centrelink debt information provided to collection agencies
(Senate Community Affairs References Committee, above n 3, 104), but it is the raising of the debt
and its personal stigma of debt that remains powerful.
36
See Mark Thomas, ‘Disclosable, but Not Necessarily Fatal? Welfare Overpayments in the Uncertain
Landscape of Fitness for Practice’ (2017) 40 University of New South Wales Law Journal 1072.
37
See for example the analysis by Deputy President Jarvis in Re Secretary, Department of Education,
Employment and Workplace Relations and Kambouris [2008] AATA 221, [30]–[32], and also
Senior Member Bayne in Re Johnson and Secretary, Department of Family and Community Services
[2000] AATA 424, [39].
38
Social Security (Administration) Act 1999 (Cth) ss 129, 135 (ARO), 142 (AAT1), 179 (AAT2). See
further, Terry Carney and Chris Bigby, ‘Social Security and Welfare Rights – What Role for Social
Work?’ in Simon Rice, Andrew Day and Linda Briskman (eds), In the Shadow of Law (Federation
Press, 5th ed, 2018) forthcoming.
2018 The New Digital Future for Welfare 9

or faint understanding of the correct basis of calculation39), or imply that debtors have
now ‘accepted’ calculation on an ATO average (which would not obviate the decision-
makers legal responsibility to themselves be satisfied40). This is both morally dubious
and, in Kate Galloway’s words, it ‘represents a breakdown in standards of governance’.41

Given that Centrelink has not taken the opportunity to challenge the argument outlined
earlier, there is also an issue about whether maintaining averaged debts on internal (ie
ARO) review and at first tier AAT (‘AAT1’) is consistent with the Commonwealth’s
‘model litigant’ policy.42 This policy codifies a fundamental principle of the 17th century
Restoration Settlement that government be assiduous to avoid conducting litigation in
ways oppressive to citizens, or other than consistent with principles of ‘fair play’ – as
Griffith CJ long ago expressed it in a 1912 High Court case, describing this as an
‘elementary’ principle that he would be ‘glad to think I am mistaken’ in believing may
not be known or be thought to be ‘out of date’.43 As recently suggested by Logan J of the
Federal Court, failure of the Crown to act as a model litigant not only is a breach of this
principle (and of the code) but potentially is also open to professional misconduct,
contempt or criminal sanction.44 Current Centrelink review and appeal practice arguably
breaches that model litigant standard.45

There is of course no reason to doubt the validity of Centrelink initiating enquiries and
investigations in any cases of discrepancies between ATO data-match averages and the
basis on which the person’s fortnightly rate was initially paid. This is because the
threshold for exercising its powers to require provision of information is low to
negligible, such as that it be ‘consider[ed] that [the information sought] may be relevant’

39
It is very common for applicants to confuse annual taxable income for family payments with the
fortnightly gross income basis of working age payments or the changes in the annual amount of
gross income used for pensions.
40
It is an error of law to accept a concession or other alternative to making findings of fact about all
relevant aspects of the decision, as explained in Staunton-Smith v Secretary, Department of Social
Security (1991) 32 FCR 164, 171, 177 (O’Loughlin J).
41
Galloway, above n 6, 94.
42
For an outline see Eugene Wheelahan, ‘Model Litigant Obligations: What Are They and How Are
They Enforced?’ (Paper delivered at Tax Ethics Seminar: Model Litigant Obligations What Are
They, What Do They Add and How Are They Enforced?, Melbourne, 15 March 2016)
<http://www.fedcourt.gov.au/digital-law-library/seminars/ethics-seminar-series/20160315-eugene-
wheelahan>.
43
Melbourne Steamship Co Ltd v Moorehead (1912) 15 CLR 333, 342. The first reported AAT2
review of a robo-debt followed Centrelink acceptance of an AAT1 set-aside and remission for
recalculation based on employer pay records, adding weight to the argument that it is a breach of
model litigant obligations to permit such debts to be pursued on an unsound legal basis until
overturned: see Re Gosse and Secretary, Department of Social Services (Social services second
review) [2018] AATA 55 (Senior Member Tavoularis).
44
Shord v Commissioner of Taxation [2017] FCAFC 167, [167]–[174], see especially [174] (Logan J).
45
In the experience of the author as a member of AAT1, in practice Centrelink fails to refer at all to
the legal foundation for debts routinely defended before AAT1, is elliptical at best when directed to
make written submissions to AAT1, and appears to have elected not to challenge adverse AAT1
robo-debt rulings by applying to AAT2.
10 UNSW Law Journal Forum 2018

to a social security issue.46 Equally there is everything to support the practice of first
asking the person for any records or other information which would clarify the debt – it is
both quicker, cheaper, more efficient, and simple good common sense to do so. Nor
might it be problematic should Centrelink decide, in the absence of any fuller information
than bank deposit records of earnings, to calculate what may be termed a ‘bedrock debt
amount’ based solely on the ‘after-tax’ bank deposits (rather than what in common terms
translates as the ‘gross earnings’ figures strictly required47). It again may be cheaper,
fairer and more economical to do so (though a true algorithm surely could fairly readily
be devised to correctly extrapolate gross earnings, leaving only the tricky precise
attribution to each payment fortnight). All of this makes sense.

From a pure cost-efficiency of administration perspective it is also understandable that

the OCI system has sought to maximise the burden of investigation for citizens and
minimise administrative cost-overheads for Centrelink. That cost-benefit calculus is
already legislated in another context (waiver of very small debts48) and it is well accepted
that a major reason that approximately 93 per cent of data-matches were not selected for
further manual investigations prior to the introduction of the OCI system was due to
recovery costs outweighing the sum potentially recoverable. In addition to fiscal
considerations, the moral imperative of recovery of monies overpaid without lawful
authority (the common law Auckland Harbour Board principle that made all
overpayments recoverable once proven) also explains the drivers of the new system.49

However as shown already, absent sufficient evidence of an actual debt based on the
proper fortnightly data, there can be no legally sustainable decision to raise and recover
the debt as speculated from averaging. So the first question is how should the AAT deal

46
See, eg, Social Security (Administration) Act 1999 (Cth) s 192. Regarding the purpose and breadth
of this power see Hendy v The Manager Centrelink (Ipswich) [2004] FMCA 579, and (in respect of
its predecessor) Sheil v Secretary, Department of Social Security (1999) 56 ALD 465, where Katz J
explained ‘Parliament’s purpose in including the provision was to confer a power capable of being
used in aid of the prevention or recovery of unjustified payments of social security benefits’: at 472
[39]. The power displaces privacy objections: Rahman v Ashpole [2007] FCA 1067, [18]–[19]
(Graham J); and includes the obligation to provide any relevant estimates: Re Adkins and Secretary,
Department of Family and Community Services [2005] AATA 714, [18] (Senior Member
Friedman).
The exercise of the power was assessed on a test of reasonableness by Senior Member Fice in Re
Almosawi and Secretary, Department of Social Services [2015] AATA 968, [17]. See also Deputy
President Hotop in Re Budalich and Secretary, Department of Employment and Workplace
Relations [2007] AATA 1258, [30]–[31] on the issue of ‘relevance’ and of specification of what is
required to be supplied.
47
Social Security Act 1991 (Cth) s 8(1) (definition of ‘income’); ‘Parliament chose to define ‘income’
… in terms of considerable width to ensure that it brought within its net as wide a range of
categories and sources of income as possible’: Rose v Secretary, Department of Social Security
(1990) 92 ALR 521, 523–4 (The Court); Read v Commonwealth (1988) 167 CLR 57.
48
Social Security Act 1991 (Cth) s 1237AAA.
49
Auckland Harbour Board v The King [1924] AC 318. Social security debt recovery provisions
modify this both by restricting recovery to statutorily defined debts (see n 26 above) and by
provision for waiver or deferral of debts such as on establishing sole departmental error or special
circumstances: Social Security Act 1991 (Cth) ss 1237A, 1237AAD respectively.
2018 The New Digital Future for Welfare 11

with such matters on review? In the Federal Court case of Harris,50 after indicating that
parties might be expected to provide material such as a ‘properly prepared application’,
Gyles J observed (omitting references), that:

The AAT stands in the shoes of the Department and is in precisely the same situation as the
decision maker … The provisions of s 33 of the AAT Act give ample scope for the AAT to
arrange investigation of a claim. The decision-maker is bound to use his or her best
endeavours to assist the AAT to make its decision … The AAT has inquisitorial powers
and may exercise them where appropriate. … It is not, of course, every case that will
require such measures. In general, an applicant for a benefit must satisfy the decision-
maker of the necessary criteria. However, cases such as this may demand such an approach
[ie active steps by Centrelink].51

Gyles J was dealing with an application for disability support pension, but the point about
being mindful of the availability of activist or ‘inquisitorial’ powers and procedures
available to the AAT, and of course to Centrelink, is a pertinent one. However, with
respect his Honour assumes a level of resourcing and plentitude of time for full blown
inquisitorial enquiry quite at odds with the contemporary reality of resourcing available
for AAT review, at least at AAT1.

The main powers available to the AAT in its Social Services and Child Support Division
(‘SSCSD’) are: (i) to set the decision aside and substitute no debt;52 (ii) to set the decision
aside and send it back to be re-determined in accord with directions;53 (iii) to adjourn the
proceedings and exercise Tribunal powers to seek information directly;54 or (iv) to
adjourn the proceedings and exercise powers to require Centrelink to supply the Tribunal
with such employment records it can obtain.55 (Unlike other AAT divisions, the special
power of ‘remittal’ for reconsideration ‘at any stage of a proceeding for review’ is not
available,56 leaving the last mentioned power to require Centrelink to use its powers).

At least at AAT1,57 the complexity of earnings investigations, and the lack of information
about the contact details of the employers, renders inappropriate the third option ((iii)
above) of resort to the AAT’s direct inquisitorial powers, as contemplated in Harris.58
The first option, of setting the debt aside and substituting a decision that there is no debt,

50
Harris v Secretary, Department of Employment and Workplace Relations (2007) 158 FCR 252.
51
Ibid 257 [19] (Gyles J).
52
Administrative Appeals Tribunal Act 1975 (Cth) s 43(1)(c)(i).
53
Administrative Appeals Tribunal Act 1975 (Cth) s 43(1)(c)(ii).
54
Social Security (Administration) Act 1999 (Cth) s 165A(1) (issuing a notice to any person with
relevant information, requiring it to be supplied).
55
Social Security (Administration) Act 1999 (Cth) s 166 (directing the exercise of the information
gathering powers of section 192).
56
This is because the power is withdrawn by Administrative Appeals Tribunal Act 1975 (Cth) s
42D(1).
57
If the question of legal validity of the debt is overlooked or wrongly decided at AAT1, the case for
exercise of options (iii) or (iv) on an AAT2 appeal would however be stronger, given the ultimate
responsibility of the Tribunal to determine the merits of the matter, however time-consuming that
may be: Hanks, above n 9, 21–2.
58
See n 49 and accompanying text above.
12 UNSW Law Journal Forum 2018

might seriously be entertained for exercise. This is because no rules actually prevent
Centrelink from making a fresh, securely grounded future decision about a new debt
(broadly speaking there are no estoppel or res judicata barriers to prevent this59).
However, it must be doubted that such a decision would be in the public interest or
consistent with AAT objectives. This is because it would appear ‘odd’ to ordinary people
that their AAT ‘no debt’ ruling could later be raised afresh, and because lack of proof of a
debt based on averaging does not mean that there is no other debt amount to be
investigated and properly calculated on a fortnightly earnings basis.

The last option ((iv) above) of adjourning to await the results of an AAT direction to
Centrelink to exercise its powers to compel production of information60 could also be
entertained, but it too runs contrary to AAT objectives to be expeditious;61 and in any
event it is able to be incorporated in the exercise of the remaining option (setting aside
with directions). Because there may or may not be a recoverable overpayment (usually in
a different and lower amount), I therefore argue that the most appropriate power at AAT1
is to set aside the decision based on averaging and send it back to be reconsidered in
accordance with a direction that any overpayment as may or may not be raised be
calculated on precise information of earnings in relevant fortnights.

So where does this leave the future, and could handling of these cases be improved to
avoid or reduce the need for AAT review?

B The Future?

Machine learning algorithms and other digital applications to improve the accuracy and
efficiency of decision-making are unquestionably the way of the future, as the rapid
expansion of such systems across a wide range of administrative settings in the USA
testifies.62 Apart from the rule of law challenge in designing such systems, there is the
challenge of rendering it consistent with principles of sound administration, such as the
27 principles laid down by the Administrative Review Council in its 2004 report, 63 or
Jerry Mashaw’s accuracy, efficiency and ‘dignity’ objectives. 64 As demonstrated,
machine learning initiatives contravene dignity and fairness principles if citizens are
disadvantaged by presumed digital literacy (access to or ability to use computers), lack of
understanding of the true nature of the issue (as in not knowing that fortnightly income
outcomes are very different to application of an average), are overcome by (possibly

59
For a detailed review of such principles see Deputy President Forgie in Re Rana and Military
Rehabilitation and Compensation Commission (2008) 48 AAR 385.
60
See n 54 and accompanying text above.
61
The AAT objectives are to be ‘accessible’, ‘fair, just, economical, informal and quick’ and
‘proportionate to the importance and complexity of the matter’: Administrative Appeals Tribunal Act
1975 (Cth) s 2A(a)–(c).
62
See Coglianese and Lehr, above n 2, 1160–7; Desai and Kroll, above n 2, 1–2.
63
Administrative Review Council, ‘Automated Assistance in Administrative Decision-Making’, above
n 30.
64
Jerry L Mashaw, Bureaucratic Justice: Managing Social Security Disability Claims (Yale
University Press, 1983) 26, 95–6, discussed in Desai and Kroll, above n 2, 5–6, 55.
2018 The New Digital Future for Welfare 13

misplaced) feelings of fear and guilt about a suggested moral wrong such as incurring a
debt, or otherwise ‘cause a disproportionate impact on members of certain classes or
groups’.65

Addressing such concerns calls for creativity and ongoing debate on alternatives. Just as
modes of achieving accountability alter when, say, delivery of welfare is shifted from
government to private sector auspices (as with job placement services in Australia) or to
charitable agencies (as in some instances in the United States of America), such changes
are not intrinsically better or worse, but call for careful weighing up of attributes of these
radically different ‘regimes’.66 Arguably so too when moving from more traditional
human agency decision-making to greater (or complete) reliance on machine learning
systems. For example reasonable minds still differ over whether legal paradigms of
greater ‘transparency’ of system design and operations is the answer, with Desai and
Kroll persuasively arguing instead for a ‘technological’ remedy of incorporating into
regulatory and accountability frameworks the ‘trust but verify’ approach adopted by the
sector when building and testing systems.67 By contrast, Coglianese and Lehr assess
machine learning against traditional legal standards of non-delegation, due process
(procedural fairness), non-discrimination and transparency;68 worthy standards of course,
but ones which the history of robo-debt demonstrates proved to be inadequate to redress
its systemic deficiencies, at least within the current system of review and appeal. Once
Australia’s OCI scheme is reformed to be compatible with the rule of law and more
compatible with best practice principles of administration, attention should turn to these
wider considerations for other machine learning digital initiatives.

However, there are also potential lessons for refinement of AAT practice. Lorne Sossin
argues that selection of the best model of a tribunal, and its finer aspects of design,
ideally should involve:

a holistic enterprise, involving the expertise of policy-makers and lawyers, administrators

and IT professionals, organizational and behavioural specialists together with
communication experts. All aspects of the tribunal experience should be considered
together – that is, the statutory authority of the tribunal together with its physical and
virtual presence, the budget and staffing of the tribunal together with its approach to
proportionality or streaming of caseloads, the rule-making together with the strategies for
accessibility, inclusion and accommodations.69

65
Coglianese and Lehr, above n 2, 1217.
66
Jerry Mashaw, ‘Accountability and Institutional Design: Some Thoughts on the Grammar of
Governance’ in Michael W Dowdle (ed), Public Accountability: Designs, Dilemmas and
Experiences (Cambridge University Press, 2006) 115, 129.
67
Desai and Kroll, above n 2, 78–9.
68
Coglianese and Lehr, above n 2, 1176–213.
69
Lorne Sossin, ‘Designing Administrative Justice’ (Osgoode Legal Studies Research Paper No
26/2017, 27 January 2017) <https://ssrn.com/abstract=2906784> 18.
14 UNSW Law Journal Forum 2018

For instance, the conversational, symbolic and other atmospherics of hearings can be
critical to real engagement and accessibility.70

Refining administrative review to fit contemporary circumstances is not new, as Lorne

Sossin observes.71 For its part, the SSCSD of the AAT, as successor to the Social
Security Appeals Tribunal (‘SSAT’), experienced significant changes over the last
decade72 as it balanced competing pressures of justice and efficiency. For instance the
legislative requirement for quick decision-making is said to stifle the degree to which the
AAT itself now actively seeks additional information from agencies or others.73 The raw
numbers demonstrate the dramatic pressures on the SSCSD, with a 44 per cent increase in
appeal numbers in the last five years coinciding with a roughly 33 per cent decrease in
membership; and perhaps not entirely coincidentally, a seven percentage point decline in
set aside decisions (from 27 per cent to 20 per cent) over the decade 2007–17.74 So
procedure matters.75

The main implication I suggest is that the SSCSD should focus on ways in which its
decision-making can boost the normative or educative impact of review in improving
primary decision-making. Given that the illegality of OCI debt raising suggested here
continued unchecked for 18 months as at the date of writing, despite AAT1 decisions
invalidating it, and that those legal doubts remained unbroached publicly, it is clear that
neither the normative nor the educative power of current review is optimal. Selective
publication of AAT1 decisions, especially where, for whatever reason, the agency elects
not to seek review AAT2 of an adverse AAT1 decision,76 is one remedy. Another is being
alert to unintended consequences – such as any premature but notionally ‘voluntary’
withdrawal of applications by under-informed applicants during any pre-hearing

70
Vicki Lens and Susan Elizabeth Vorsanger, ‘Complaining after Claiming: Fair Hearings after
Welfare Reform’ (2005) 79 Social Service Review 430.
71
Sossin, above n 69, 5–6.
72
Further, Carney and Bigby, above n 38.
73
Juliet Lucy, ‘Merits Review and the 21st Century Tribunal’ (2017) 24 Australian Journal of
Administrative Law 121, 128–9, 137. The power to require provision of information or documents
held by Centrelink is found in Social Security (Administration) Act 1999 (Cth) s 165(1), while s
165A(1) empowers seeking information from others.
74
Carney and Bigby, above n 38. Tightening of eligibility for Disability Support Pension, particularly
the imposition of a requirement to spend 18 months working with a specialist disability employment
provider (unless ‘severely’ disabled as defined), no doubt accounts for a significant portion of the
decline, given that DSP cases accounted for 41 per cent of the AAT1 caseload: Abigail Rice, Senior
Reporting Officer, ‘Data Provision for Rice and Day Social Work Chapter’ (AAT Data, 2017) (copy
available from the author).
75
By way of historical comparison from a 1978 experimental trial of routine invitations to attend in
place of the then practice of initially reviewing all cases ‘on the papers’ before determining what
further inquiries or invitations would be pursued, set aside rates rose from 10.3 per cent to 33.3 per
cent (and from late 1978 onwards hearings became routine based on this research): Terry Carney
and Elizabeth Marshall, ‘Social Security Appeals Tribunals: Report on Tribunal Procedures in
Victoria’ (Report, Social Security Appeals Tribunal, 1978) 3 (copy available from the author).
76
For a possible ‘nonacquiesence’ explanation, see Carolyn A Kubitschek, ‘Social Security
Administration Nonacquiescence: The Need for Legislative Curbs on Agency Discretion’ (1989) 50
University of Pittsburgh Law Review 399; Michael J Froehlich, ‘Administrative Nonacquiescence in
Judicial Decisions’ (1984) 53 George Washington Law Review 147.
2018 The New Digital Future for Welfare 15

screening;77 or the absence of feedback to the agency when delivering oral decisions; or
any undue exercise of powers to endorse (unpublished) settlement of such matters at
AAT2 – all of which weaken educative feedback loops to decision-makers.

Given that many people in receipt of working age payments are vulnerable and thus may
not participate in their hearing,78 care is also needed in exercising dismissal and
reinstatement powers. For instance it is wrong to apply the same test when dealing with a
reinstatement application79 as when dismissing for failure to attend a scheduled hearing:80
the latter is mainly a question of whether the person was adequately notified and any
excuse they may have for non-attendance, while reinstatement crucially also requires
application of a presumption of reinstatement and some consideration of the merits of the
matter.81 Channelling Juliet Lucy,82 there may also be other creative (and highly cost-
effective) possibilities. Examples include greater use of pre-hearing powers to require
Centrelink provision of additional documents or information before AAT1 hearings, or
clarification of analysis or reasoning, effectively ‘front-ending’ considerations otherwise
only incorporated as part of AAT directions when deciding the application.

Ultimately, however, the aim must be to ensure that primary decision-making is of the
highest quality, integrity and legality, minimising the need for what Juliet Lucy terms the
AAT’s function of ‘facilitat[ing] “administrative second thoughts”’.83

IV CONCLUSION

If the undoubted benefits of machine learning digital initiatives are to be realised, and
public confidence retained, they must be squared with the rule of law, with best practice
principles of administration, and with ethical considerations of fairness and distributional
equity. Writing about such technological changes, former High Court Justice Kenneth
Hayne recently insisted that ‘[t]here must be a system of general rules … [and] [t]hose

77
The subject of an as yet unpublished SSCSD small trial.
78
For example, the unemployed disproportionately experience insecure housing, demoralisation and
low self-esteem due to inability to find work, and are over-represented by mental illnesses such as
depression: see Frances M McKee-Ryan et al, ‘Psychological and Physical Well-Being during
Unemployment: A Meta-analytic Study’ (2005) 90 Journal of Applied Psychology 53. For a
discussion of the strengths and limitations of concepts of vulnerability, see Kate Brown, Kathryn
Ecclestone and Nick Emmel, ‘The Many Faces of Vulnerability’ (2017) 16 Social Policy and
Society 497.
79
Administrative Appeals Tribunal Act 1975 (Cth) s 42A(8A), (9).
80
Administrative Appeals Tribunal Act 1975 (Cth) s 42A(2)(a).
81
Three principles were stated by Deputy President Forgie in Re White and Secretary, Department of
Families, Community Services and Indigenous Affairs (2007) 46 AAR 208, 216–18 [23]–[28] (‘Re
White’) (namely: (i) a presumption of reinstatement; (ii) equity of case management between
dismissals; and (iii) the merits of the case if reinstated). Re White was recently adopted by Senior
Member Kelly in Re Tighe and Secretary, Department of Social Services [2017] AATA 408, [96].
82
Lucy, above n 73, 139.
83
Ibid 125, quoting Re Greenham and Minister for Capital Territory (1979) 2 ALD 137, 141 (The
Tribunal).
16 UNSW Law Journal Forum 2018

general rules, and only those rules, must be applied and enforced’.84 Or as Coglianese and
Lehr observe for the USA,

If machine learning can help regulatory agencies make smarter, more accurate decisions,
the benefits to society could be considerable. But can the prospect of the government
regulating by robot, or adjudicating by algorithm, be accommodated within prevailing legal
norms? Fitting machine learning into the regulatory state may turn out to be one of the
most fundamental challenges facing the …. governmental system in the decades to come.85

This article suggests Australia too is yet to fully come to grips with this foundational
challenge. As Hayne acerbically puts it, ‘[s]howing that something can be done does not
mean that it should be done’.86 Robo-debt tarnishes the reputation of machine learning
and wreaks legal and moral injustice.87 It flouts key design principles laid down by the
Administrative Review Council88 and surely needs to be corrected.

84
KM Hayne, ‘Opinion: “Change”’ (2017) 42 Alternative Law Journal 87, 87.
85
Coglianese and Lehr, above n 2, 1153 (emphasis added).
86
Hayne, above n 84 (emphasis in original).
87
Arguably an even greater injustice than that struck down by the High Court in Green v Daniels
(1977) 13 ALR 1 (an unlawful policy direction not to pay benefits to school leavers, as against the
government OCI program to raise debts without adequate legal foundation).
88
Administrative Review Council, above n 30, viii–xi. See especially principle 4, that ‘any
information … to assist a decision maker in exercising discretion must accurately reflect
government law and policy’: at 16; principle 7, ‘the construction of an expert system must comply
with administrative law standards if decisions made …are to be lawful. Decisions made by or with
the assistance of expert systems must comply with administrative law standards in order to be
lawful’: at 27; and principle 23 external review conducted manually in accordance with the
procedures and practices of the tribunal: at 47.