National Disability Insurance Scheme Amendment (Getting the NDIS Back on Track No. 1) Bill 2024 [Provisions]
Senate Standing Committees on
Community Affairs Inquiry
National Disability Insurance Scheme Amendment
( Getting the NDIS Back on Track No. 1 ) Bill 2024 [ Provisions ]
Supplementary Submission to Submission Number 15
2 June 2024
Submission By Ms Marie Johnson CEO Centre for Digital Business Pty Limited
National Disability Insurance Scheme Amendment (Getting the NDIS Back on Track No. 1) Bill 2024 [Provisions]
I would like to provide this additional information as a supplementary submission, to my main submission (number 15).
This additional information constitutes two articles published in the past few days.
The first article is persuasive for the opinions presented by two esteemed scholars, one being Terry Carney AO, whose opinions and articles held weight at the RoboDebt Royal Commission.
The first article is “Decoding the algorithmic operations of Australia’s National Disability Insurance Scheme” by Georgia van Toorn, Terry Carney AO. First published: 30 May 2024, in the Australian Journal of Social Issues.
Georgia van Toorn is a lecturer in the School of Social Sciences at the University of New South Wales and an Associate Investigator at the ARC Centre of Excellence for Automated Decision-Making & Society (ADM+S).
Terry Carney AO is Emeritus Professor at the Law School University of Sydney, a Fellow of the Australian Academy of Law and a past President (2005–2007) of the International Academy of Law and Mental Health.
Some of my work has been referenced in this article.
I believe the van Toorn/Carney article will become as widely known and as influential as the original Carney article, which was so persuasive in the RoboDebt Royal Commission.
Given the seriousness of the legal and ethical questions raised in this article, I believe that it is incumbent for Members of the Committee to be aware of this article and the potential legal ramifications of the connections being observed between RoboDebt and the NDIA’s use of algorithms.
I pose the question: Just how can the NDIS Bill be passed, which is not only non-compliant with the UNCRPD, but rests on such hidden yet-to-be designed algorithms (which I described in my main submission, number 15)? The van Toorn/Carney article refers to these as subterranean systems, “because their workings are neither publicly known nor amenable to legal rectification…”
Abstract
In recent years, Australia has embarked on a digital transformation of its social services, with the primary goal of creating user-centric services that are more attentive to the needs of citizens. This article examines operational and technological changes within Australia’s National Disability Insurance Scheme (NDIS) as a result of this comprehensive government digital transformation strategy. It discusses the effectiveness of these changes in enhancing outcomes for users of the scheme. Specifically, the focus is on the National Disability Insurance Agency’s (NDIA) use of algorithmic decision support systems to aid in the development of personalised support plans.
This administrative process, we show, incorporates several automated elements that raise concerns about substantive fairness, accountability, transparency and participation in decision making. The
Marie Johnson CEO, Centre for Digital Business Pty Limited Supplementary Submission: National Disability Insurance Scheme Amendment (Getting the NDIS Back on Track No. 1) Bill 2024 [Provisions] 2 June 2024
National Disability Insurance Scheme Amendment (Getting the NDIS Back on Track No. 1) Bill 2024 [Provisions]
The conclusion drawn is that algorithmic systems exercise various forms of state power, but in this case, their subterranean administrative character positions them as “algorithmic grey holes”—spaces effectively beyond recourse to legal remedies and more suited to redress by holistic and systemic accountability reforms advocated by algorithmic justice scholarship.
“This article reveals that NDIS algorithmic systems, used as supporting inputs in determining the content and funding of participants’ plans, risk replicating well-documented instances of algorithmic harm experienced in other areas of social welfare provision. We refer to these systems, and the nature of their decision-making powers, as “subterranean,” because their workings are neither publicly known nor amenable to legal rectification in the way Australia’s Robodebt or the Dutch Childcare Benefit harms were rendered accountable
These systems exercise forms of state power, yet their subterranean administrative character positions them in an “algorithmic grey hole” effectively beyond the reach of legal remedies; and one which significantly compromises the delivery of substantive algorithmic fairness by failing to routinely engage the contextual complexities of the disability experience.
A key contribution of the article is to illustrate how barriers to internal and external (judicial or tribunal) review mechanisms and lack of systemic correction, leaves the vast bulk of decision making beyond understanding or prospects of challenge. Excessive reliance on unknowable algorithms empties review of its substantive merit, transforming it into an “algorithmic grey hole.”
The second article is “Exclusive: Shorten Revives the Coalition’s Failed NDIS Reforms” by the award-winning journalist Rick Morton, The Saturday Paper Senior Reporter, published 1 June 2024.
Synopsis
The Rick Morton article is critical for the Committee’s deliberations, for it describes in detail, amongst other things, connections between the NDIS Bill, RoboDebt, debt raising, automated assessments, and the god powers of the CEO.
The article references disability advocates Uli Cartwright and Cat Walker, who told this Committee that this Bill “locks in procedurally unfair processes which echo robo-debt governance”.
Cost-saving reforms to the NDIS will allow debts to be raised against disabled people and give extraordinary new powers to cancel support entirely.
There is now mounting evidence of the connections between RoboDebt practices and the use of algorithms by the NDIA (RoboNDIS), and serious questions of legality including the operation of the access to justice, and unresolved ethical issues including the risk of harm and death.
As examined in my main submission (Number 15), there is a naïveté and a reckless indifference as to what is actually involved in implementation. Compounding this extraordinary risk of the lack of capability to implement, the commentary in the van Toorn/Carney article, presents a clarion warning as to the shifting legal minefield in the era of algorithmic decision making, which cannot be ignored.
Marie Johnson CEO, Centre for Digital Business Pty Limited Supplementary Submission: National Disability Insurance Scheme Amendment (Getting the NDIS Back on Track No. 1) Bill 2024 [Provisions] 2 June 2024
National Disability Insurance Scheme Amendment (Getting the NDI$ Back on Track No. 1) Bill 2024 [Provisions]
Both articles are reproduced in full in the following pages.
Marie Johnson 2 June 2024
Marie Johnson CEO, Centre for Digital Business Pty Limited Supplementary Submission: National Disability Insurance Scheme Amendment (Getting the NDIS Back on Track No. 1) Bill 2024 [Provisions] 2 June 2024
3
Decoding the Algorithmic Operations of Australia’s National Disability Insurance Scheme
Georgia van Toorn$^\text{1}$,$^\text{2}$ | Terry Carney$^\text{2}$,$^\text{3}$
Abstract
In recent years, Australia has embarked on a digital transformation of its social services, with the primary goal of creating user-centric services that are more attentive to the needs of citizens. This article examines operational and technological changes within Australia’s National Disability Insurance Scheme (NDIS) as a result of this comprehensive government digital transformation strategy. It discusses the effectiveness of these changes in enhancing outcomes for users of the scheme.
Specifically, the focus is on the National Disability Insurance Agency’s (NDIA) use of algorithmic decision support systems to aid in the development of personalised support plans. This administrative process, we show, incorporates several automated elements that raise concerns about substantive fairness, accountability, transparency and participation in decision making.
The conclusion drawn is that algorithmic systems exercise various forms of state power, but in this case, their subterranean administrative character positions them as “algorithmic grey holes”—spaces effectively beyond recourse to legal remedies and more suited to redress by holistic and systemic accountability reforms advocated by algorithmic justice scholarship.
Keywords
algorithm, algorithmic grey hole, algorithmic justice, automated decision-making, disability, NDIS
This is an open access article under the terms of the Creative Commons Attribution-NonCommercial License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited and is not used for commercial purposes. © 2024 The Author(s). Australian Journal of Social Issues published by John Wiley & Sons Australia, Ltd on behalf of Australian Social Policy Association. Aust J Soc Issues. 2024;00:1–19.
INTRODUCTION
In Australia, several recent high-profile scandals have highlighted the risks and challenges associated with the digital transformation and datafication of medical, corporate and social service infrastructures. One of the most egregious examples of data harms is the widely publicised case of robodebt, an automated debt assessment and recovery scheme implemented by Services Australia as part of its income compliance programme (Carney, 2018a, 2018b). The impact of robodebt, in terms of the personal and societal trauma inflicted upon millions of Australians wrongly accused of welfare fraud, has been extensively documented (Braithwaite, 2020; Graycar & Masters, 2022; Nikidehaghani et al., 2023; O’Donovan, 2019), thanks in part to a nation-wide Royal Commission (Robodebt Royal Commission, 2023). These incidents have raised awareness both with the Australian government and among the general public about the diverse risks, harms and injustices arising from the government use of automated decision-making (ADM) systems.
While there is increasing concern about automation’s societal consequences, certain areas of social provisioning have received minimal scrutiny. In this article, we examine a different “subterranean” form of automation as an “input” into decision making—the implementation of computer algorithms within the NDIS, which, despite being greatly impacted by the government’s digital transformation initiatives, has received limited attention from scholars and the media. The NDIS is one of Australia’s largest and most expensive social programmes, providing support to over 630,000 participants at an annual cost of $48.7 billion (Government, 2024). Upon applying to enter the scheme, each participant undergoes an assessment to determine their eligibility, and if deemed eligible, their needs are evaluated to develop a personalised support plan. Algorithmic technologies have become integral to NDIS assessment, planning and review processes (Australian National Audit Office [ANAO], 2020; Taylor-Fry, 2021). However, there is a notable gap in understanding regarding the functioning and consequences of these algorithms. Currently, there is very little publicly available information about the role algorithms play in the National Disability Insurance Agency’s (NDIA’s) administrative procedures. Important details remain unknown, such as the data used to inform the planning algorithms, how different data points are weighted, and where the line is drawn between automation and human discretion in individual cases. Additionally, it is unclear who is responsible for constructing these algorithms and what types of evidence and data were used in the development process.
As highlighted by some critical commentators, it is both surprising and concerning that, despite the significant impact and scale of the NDIS, there remains a dearth of understanding regarding its internal technical operations (Johnson, 2022; van Toorn et al., 2022).
In this article, we consolidate the available public evidence regarding the nature of NDIS planning algorithms, aiming to provide initial insights into this opaque process. Our specific concern lies with issues of fairness, accountability and legality associated with the implementation of algorithmic decision making in the scheme. The academic literature on algorithmic fairness has a rich history, exploring diverse definitions of fairness (Friedler et al., 2021; Hoffmann, 2019; Trewin et al., 2019). Recent research emphasises the need for a shift from formal mathematical models of “fair” decision making to more comprehensive evaluations, assessing whether algorithms can promote or deny justice in practice (Bennett & Keyes, 2019; Green, 2022b). Green, for example, contends that the problem of unfairness extends beyond formal equality in terms of “equal treatment for individuals based on their attributes or behavior at a particular decision point” (Green, 2022b, p. 4). Substantive fairness, he argues, concerns the broader landscape of unjust norms and institutions that shape the nature and consequences of decisions guided by algorithms, resulting in certain individuals having their life chances determined by “a small number of ‘zero-sum, high-stakes’” calculations (Fishkin, 2014, p. 131, quoted in Green, 2022b, p. 10)
VAN TOORN AND CARNEY
| 18394655, ———————WILEY- 0,
Our purpose in this article is to explore whether NDIS planning algorithms satisfy stan-
dards of substantive algorithmic fairness. Adopting Green’s conceptual framework, our ap-proach involves placing the legal and institutional structures surrounding NDIS planning
algorithms at the core of conversations about fairness. The agency overseeing the scheme, the
NDIA, has thus far kept the details of its data-driven, algorithmic planning and assessment
processes concealed from public view. Due to the restricted availability of information about
these administrative processes, we focus on secondary sources, including consultancy and ex-
ternal auditors’ reports as well as the limited documentation published by the NDIA.
Through our discussion, we highlight several technical and socio-legal aspects of the
scheme’s decision-making infrastructure that subvert substantive algorithmic fairness. These https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342
include, first, the absence of any policy or legal mechanism to ensure algorithmic inputs are
reasonably balanced against the views and preferences of people with disability; second, the National
manner in which computational processes oversimplify the concept of disablement, imposing Health
on people standardising data categories that can distort their experiences of disability and And
Corresponding support needs; and third, the limited recourse options for NDIS participants Medical
to challenge decisions informed by algorithms. In particular, we note the absence of legal con- Research
tsraints governing the use of algorithms in NDIS planning. These findings indicate the pres-
ence of what legal scholars term an “algorithmic grey hole” (Solow-Niederman, 2023). In this Council,
scenario, the application of algorithms by governments is subject to certain legal limitations, Wiley
but these constraints are so minimal that they effectively permit government agencies to act Online
without meaningful restraint. A key contribution of the article is to illustrate how barriers to Library
internal and external (judicial or tribunal) review mechanisms and lack of systemic correc- on
ion, leaves the vast bulk of decision making beyond understanding or prospects of challenge. [31/05/2024]. Excessive reliance on unknowable algorithms empties review of its substantive merit, trans-
torming it into an “algorithmic grey hole.” Seethe
The argument is developed in three main sections. The following section, Section 2, provides Terms
some background to the NDIS, explaining how a combination of financial and administrative and
pressures have led to the adoption of algorithmic decision support tools within the scheme. Conditions
In Section 3, we explore the nature and purpose of these algorithmic processes, their role in
formulating individual support plans, and issues of human bias, inconsistency, inequity and
cost overruns that an algorithmic approach is intended to address. In Section 4, we highlight
a number of tensions between the principles established in the NDIS Act and the undisclosed
use of algorithms to guide the planning process. Disability experiences and support needs, we
argue, are highly unique to each individual, and can be influenced by complex social and cul-
tural factors, making a rigid algorithmic approach inappropriate and potentially harmful. In
addition, we explore accountability, legality and procedural fairness aspects of NDIS decision (https://onlinelibrary.wiley.com/terms-and-conditions)
making. Our concluding remarks call for immediate action to ensure clear paths for review, on
transparency in algorithmic processes and socio-cultural shift that involves the public, partic- Wiley
ularly people with disabilities, in rethinking the role of technology in alignment with principles Online
of disability justice. Library
2 | RESHAPING RESOURCE ALLOCATION: THE use;OA
EMERGENCE OF NDIS AUTOMATION articles
are governed by the applicable Creative
The NDIS is a creature of the complications of a federal system of government that divides responsibilities between the national and the states/territory levels of government. It was by
to conceived on advice of the Productivity Commission as an additional and generous national the
“capstone” programme in which funding for individualised support would be targeted to a
small group of people with “severe and permanent” disability (Productivity Commission, 2011,
esp. 10–21). Known as Tier 3 assistance, this funded support would constitute the majority of
VAN TOORN AND CARNEY
NDIS Expenditure and Policy Challenges
NDIS expenditure, which for the Productivity Commission signalled the need for robust and nationally standardised tools to evaluate eligibility. Jointly funded by both levels of government (most extensively by the national government), the scheme was premised on retention of state/territory funding of existing services and support arrangements for an estimated 4 million people with “lower level or shorter-term disabilities” (known as “Tier 2” participants). This assumption was fragile because it relied on two dubious mechanisms: federal-state bilateral agreements and successful legislative drafting to confine NDIS expenditures to Tier 3 participants.
Federal-state bilateral funding agreements have proven to be especially weak policy instruments in Australia. This is mainly because of the fiscal imbalance between the wealthier national government with greater revenues and fewer responsibilities and the cash-strapped states and territories with carriage of most services. This means there is usually an additional revenue cost to the national government as a foundation for requiring a level of matching state expenditure, but government-to-government agreements are difficult to draft with any precision as well as being beset by fraught federal-state political tensions (Carney et al., 2019, p. 799). For its part, key provisions of the NDIS legislation are crafted so as to quarantine sup- port to Tier 3 participants. For example, eligibility is framed to exclude people experiencing lower-level impacts from a disability (the eligibility threshold); participants are not funded for supports available in mainstream health and other service programmes or those able to be provided by family or civil society (the boundary provisions); and value for money was written in as a funding guideline (economic sustainability).
While eligibility boundaries proved difficult to incorporate or administer given the greater imperative to honour principles of individualisation and control (Venning et al., 2021, pp. 103–105), those scheme elements have for the most part succeeded in concentrating NDIS resources in the way intended once someone is admitted as a participant (though numbers of participants have exceeded expectations). What failed almost completely in the overall design was not anticipating that states and territories would close most Tier 2 existing disability programmes (banking-associated expenditure savings) and that the neoliberal assumption of a competitive “market” to actualise participant choice and control of supports would prove to be a mirage (Considine, 2022, pp. 113–140; Dickinson & Yates, 2023; van Toorn, 2021). Rectifying these design failures was one key challenge of the Bonyhady/Paul inquiry (Campanella et al., 2023).
The underlying policy drivers of the NDIS are a combination of the familiar (individual- isation and client control; preservation of financial viability) and the rarely encountered—namely a philosophy of investing public support under a “life-course” version of the insurance principle. Popularised as maximising financial and other gains, including increasing disabled people’s social and economic participation, this principle involves consistently assessing levels of individual support provided within a frame of a (projected) whole-of-life trajectory. In combination with an obligation to ensure scheme viability across all participants, this results among other things in desirable outcomes such as favouring early capacity-building when assess-ing funding entitlements (Ng & Gray, 2022, pp. 654–655), but at heart poses a fundamental tension that is difficult to resolve (Needham & Dickinson, 2018).
The switch from the previous disability model of take-it or leave-it access to a standard offering of one-size-fits-all government designed and delivered disability services, to the NDIS model of tailored funding packages for the purchase of required supports, called for a radical new form of legislation and administration. Under this ideal, no longer could simple so-called “bright line” eligibility rules be enacted and objectively administered. Scheme eligibility instead had to be cast in more subjective language, and likewise the resource entitlements of participants, resulting in language such as “reasonable and necessary supports.” The complexity and unfamiliarity of this legislative architecture (Ardill & Jenkins, 2020) is reflected in difficulties encountered by the Administrative Appeals Tribunal (AAT) on review of NDIS decisions (Venning et al., 2021). Fuzziness of eligibility (numbers and needs of participants)
ALGORITHMIC DECISION-MAKING TOOLS IN THE NDIS
From the outset, the Productivity Commission envisaged a significant role for yet-to-be-developed tools, writing that:
Assessments [sh]ould be designed to be as objective as possible. The people making assessments would need to be independent from the client (unlike treating general practitioners), be properly trained in the use of the tools and be approved or appointed by the National Disability Insurance Agency for the purpose of conducting NDIS assessments. The agency would monitor assessors for their appropriate use of the assessment tools. ‘Hard’ assessments would be unfair on the client. Assessment ‘softness’ could jeopardise the scheme…
(Productivity Commission, 2011, pp. 20–21)
In theory, such rating tools could be injected at the application for entry to the scheme or any of the subsequent main stages of NDIS decision making.
Suitable objective tools for determining scheme eligibility were found not to exist at the time of that inquiry (Productivity Commission, 2017). Early NDIS roll-out stimulated interest in developing tools as under-staffing and under-qualified staff with too little time, saw indicative plans imposed and re-used without face-to-face refinement, access decisions favouring some disabilities over others, and both access and package value favouring applicants with good access to social capital of informed family/advocates (Carney et al., 2019; Dickinson et al., 2021, p. 35). From the outset, there were concerns that the absence of objective tools to direct funding allocations could lead to the scheme becoming financially unviable.
VAN TOORN AND CARNEY
6 |
| 18394655, -WILEY |
The Productivity Commission (2011, pp. 338–339) recommended a “transition to a fully- fledged assessment toolbox” drawing on “multiple sources of information.” Experts at the time warned against employing “ready made tools” for the sake of expediency (Productivity Commission, 2011, p. 338).
Following an initial trial of a functional assessment tool developed internally by the NDIA, however, the agency changed its approach and adopted a set of generic assessment instruments for 11 primary disability types, along with the World Health Organization Disability Assessment Schedule version II (WHODAS II) for people with disabilities not listed (including mental health conditions/psychosocial disabilities) (National Disability Insurance Agency [NDIA], 2015; see also Productivity Commission, 2017, p. 156). The assessments produced a “severity” score to inform calculations for both determining eligibility and allocating funding.
These points-based scoring tools essentially convert information about a person’s disability into numerical metrics, serving as crucial data inputs for eligibility and planning decisions.
Since the early implementation of the scheme, it was envisioned that these decisions would adhere to a “guided” decision-making process integrated into an overall resource allocation framework supported by actuarial modelling of scheme costs (NDIA, 2016; Productivity Commission, 2011). This process relies on automatically categorising each individual into a “reference group” using a computation that takes into account their level of functioning (the “severity” of their disability), age and type of disability (Productivity Commission, 2017).
Once grouped, the person is then assigned a typical support package (TSP) based on their corresponding reference group. The plan is able to be adjusted manually, either adding or subtracting entitlements, if it is judged not to meet specific needs, taking into account what supports the individual already has in place, and whether these are considered sufficient (See Figure 1). As the NDIA explained in 2016:
The idea is to try to disaggregate the participant population into homogeneous subgroups where each person in a subgroup is expected to have a similar a priori package cost (although there will inevitably be a distribution within each sub- group) but where, at the same time, the subgroups are big enough to have some statistical credibility. (NDIA, 2016, p. 19)
Use of TSPs is not a rigid requirement (planners can build a plan without using the tool), but the approach is strongly encouraged to establish consistency, enabling the prediction of an individual’s potential funding needs, and also aiding in the estimation and control of overall scheme costs. Nonetheless, it relies on a form of circular reasoning: the algorithm generates a standardised plan, which is later modified according to individual situations, and the data from these plans are then used to benchmark future iterations of the algorithm.
VAN TOORN and CARNEY 7 | 18394655, ———————WILEY- 0,
being added to the support package, leading to an increase in both actual and projected Downloaded
scheme costs. It was precisely this concern that prompted the government to propose from
outsourcing the assessment process to “independent” private contractors and replac-
ing the TSP with a more data- intensive social profiling process (Joint Standing C'tee on
NDIS, 2021b).
Akin to the role of impairment tables in Disability Support Pension and Carer pay-
ments assessment since the 1990s (Carney, 2006, Ch. 8; van Toorn & Soldatic, 2024), the
proposal envisaged using independent assessors and functional assessment scores in asso-
ciation with other undisclosed data firstly, to ascertain eligibility for NDIS participation (access), and secondly, to automatically match each individual to one of 400 “profiles” https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342
(NDIA, 2021). Similar to the existing reference group model, each profile had a correspond- by
ing indicative funding package, enabling an automated calculation of “a reasonable and National
necessary funding amount for participants with similar functional capacity and life cir- Health
cumstances” (NDIA, 2021, p. 16). This approach was touted by government as a corrective And
for scheme inequities (favouring of some disabilities over others; barriers to access for ap- Medical
plicants without access to a good portfolio of recent medical reports), as well as reducing Research
staffing overheads and processing time (Dickinson et al., 2021; Johnson, 2021). Crucially,
the automated components were said to address issues of “sympathy bias” in a scheme Council,
heavily dependent on “individual public servants' judgement and their natural empathy” Wiley
(Australian Government, 2021, p. 32). Heavily criticised on a host of grounds (van Toorn & Online
Scully, 2023), including its oversimplification of complex planning (Dickinson et al., 2021), Library
perceived cost- minimisation function and violation of the United Nations Convention on on
the Rights of Persons with Disabilities (Gooding & West, 2023), the proposals were shelved [31/05/2024]. on 9 July 2021 (Henriques- Gomes, 2021).
While the independent assessment proposal never progressed beyond trial phase, the Seethe
TSP continues to serve as a reference point for guiding planners when making detailed de- Terms
cisions about the content and funding levels of supports. Their discretion comes into play and
once functional disability information has placed a participant within a reference group. Conditions
Due to the significant impact of the TSP on the total cost of the scheme, there is a growing
emphasis on its evaluation by external auditors and the NDIA (ANAO, 2023). In 2022, the
NDIA initiated a review of the TSP, aiming to address the problem of funding packages
being assigned above the value of TSPs. TSPs themselves have experienced inflation over
time due to both rising reference package values (resulting from assessments indicating
lower levels of function) and manual adjustments made by planners as they assess support
needs through guided questions (Taylor- Fry, 2021). To address these “behavioural” aspects,
particularly concerning the problem of excessive human discretion in plan valuations, the (https://onlinelibrary.wiley.com/terms-and-conditions)
Auditor- General advised the NDIA to reconsider its decision to remove a rule that previ- on
ously mandated planners to refer any budgets deviating from the TSP benchmark by over 8 Wiley
per cent to a higher- level decision maker for approval (Taylor- Fry, 2021, p. 60). The NDIA Online
agreed to “update the ICT system controls” and is currently implementing a new ICT infra- Library
structure, “with the aim of iteratively improving TSP modelling over time” (ANAO, 2023, for
rules
p. 60). of
While falling short of transparency and co- design standards recently affirmed by the use;OA
Robodebt Royal Commission (2023, Ch. 17), concerns regarding the current application of articles
algorithmic tools as a planning aid highlight other unresolved issues in relation to substan- are
tive fairness, including fairness in terms of whether the tools themselves accurately capture governed
the experiences and impacts of disability for effective support planning; whether legal and by
institutional mechanisms exist to ensure that a human decision maker is sufficiently engaged the
and accountable; and whether the decision- making process meets standards of legality and applicable
procedural fairness. Creative
VAN TOORN AND CARNEY
| 18394655, -WILEY——————— | | | |
FAIRNESS, HUMAN ACCOUNTABILITY AND LEGALITY OF NDIS AUTOMATION
4.1 Fairness
In computer science research, algorithmic fairness is understood as an approach to developing methods that address biases and discrimination inherent in algorithms (see for example, Bellamy et al., 2019). Researchers explore techniques that pinpoint biases present in the algorithm’s training data, aiming for equitable treatment across diverse demographic groups. The formal equality principle, as articulated by Aristotle, emphasises the idea that comparable situations should be treated in a consistent and equitable manner: “treat like cases as like”(Gosepath 2021, n.p., cited in Green, 2022b). As detailed below, this approach delineates the NDIA’s actuarial strategy of profiling individuals to predict their support needs, relying on a statistical average of what individuals in that category with similar characteristics typically receive.
Scholars in the social sciences and humanities contend that formal algorithmic fairness hinges on a limited analytical framework focused solely on specific decision points, detached from the broader context surrounding those decisions (Green, 2022b; Hanna et al., 2020; Hoffmann, 2019). Fairness, as outlined by Bennett and Keyes (2019) in their discussion of fairness in the context of disability justice, involves addressing social hierarchies that disad- vantage people with disability prior to and beyond the point of decision, suggesting that failure to do so may perpetuate discrimination. This aligns with Green’s notion of substantive algo- rithmic fairness (Green, 2022b). Unlike formal algorithmic fairness, which focuses on equal treatment of similar cases, substantive fairness considers broader social relationships and in-stitutional arrangements. It aims to reduce disparities rooted in social hierarchy upstream from a decision-making process. Additionally, substantive fairness advocates for a structural response downstream, suggesting the renovation of socio-legal, political and administrative structures to limit technologically mediated harm to individuals. In the context of disability, substantive fairness requires an approach that not only comprehensively considers the societal impact of technologies on the opportunities of people with disability but also addresses the social and institutional factors contributing to inequities, rather than focusing solely on tech-nical aspects.
Here, we employ a substantive algorithmic fairness perspective, directing our focus toward the legal framework governing the use of NDIS algorithms (a topic we revisit later) and whether, in this case, algorithmic decision making reproduces social/administrative processes that hamper equal participation and self-determination for people with disability. In rela- tion to fairness, a major concern expressed by scheme participants is that the use of algo-rithmic processing of participant data to create a reference package runs counter to the core principles of person-centred planning and the fundamental values of individual choice and self-determination, which underpin the scheme (Joint Standing C’tee on NDIS, 2021b). Data-driven approaches claim to offer personalised solutions, but often fail to live up to this prom-isе. These methods rely on clustering techniques to understand individuals and predict their needs and preferences for service delivery. However, their effectiveness depends on the quality of the data they receive, as the accuracy and utility of the insights generated are directly influenced by the data’s completeness and, in this case, how well they capture lived experiences and impacts of disability. Reference packages use demographic and diagnostic information to categorise people and predict their support needs. This process assumes specific needs and support preferences derived from the reference package, neglecting crucial factors such as en-vironment, social relationships, gender, sexual orientation, cultural characteristics and social/ class disadvantage, all of which play a significant role in shaping experiences of disablement. While both functional assessments and guided questions are supposed to incorporate these
VAN TOORN AND CARNEY
9
| 18394655, ———————WILEY-
factors into the planning process, research indicates that for many people this does happen in Downloaded practice (van Toorn & Scully, 2023). The complexity of these social and cultural factors, intri- from cately linked to lived experiences of disability, is not adequately considered in the algorithmic process, which can result in unsuitable plans or inadequate funding. These concerns were raised in testimonies during a parliamentary inquiry focused on in- dependent assessments. Testimonies included remarks from one of the early architects of the scheme, Bruce Bonyhady, who highlighted that algorithmic approaches “put people in boxes before they have had a chance to outline what they would like to achieve or the ways in which they hope their lives change” (Bonyhady, 2021, p. 15). Furthermore, a parent voiced their con- cern that such approaches remove the human element from the planning process and revert https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342 to a medical model of disability, placing excessive emphasis on functional capacity (Joint by Standing C’tee on NDIS, 2021a, p. 39). A more recent criticism highlighted in a submission National to another parliamentary inquiry, examining the culture and capability of the NDIA, points Health out that despite the NDIS legislation stressing the individualisation and participant- directed And nature of plans, “the application of statistical averages in automated roboplanning eliminates Medical the individual person and their needs, transmuted instead to a fictional average, a fictional Research ‘persona’” (Johnson, 2023b, p. 8). The lack of transparency in the planning process and the use of algorithmic tools without full or detailed disclosure make it difficult to determine whether Council, planners indeed overly rely on reference plans over the input and preferences of NDIA partic- Wiley ipants themselves. However, at the very least, it is problematic if participants perceive a lack Online of agency in decision making, as this can lead to a loss of trust and disengagement from the Library process. The key concern in this context is whether there exists enough “statistical credibility” on to justify using reference packages as templates, considering the known issues with the under- lying data on which they rely (van Toorn & Scully, 2023). Indeed, equity in the form of treating statistically “like cases alike” based on simplistic or inaccurate data (simple standardisation) Seethe is a poor substitute for deeper engagement with what substantive fairness and justice require in Terms this area of high- stakes decision making (Gooding & West, 2023). and For instance, significant fairness issues arise concerning the way the NDIA’s algorithmic Conditions infrastructure codes disability as a single impairment. When individuals apply for the scheme, they are obliged to specify a “primary” disability and differentiate it from any other disabili- dies or health/mental health conditions they experience. They must fulfil the eligibility criteria for each of these conditions independently. This process has been criticised for its “dangerous and arbitrary splitting out of diagnosis,” disregarding the full circumstances of the person (Johnson, 2023b, p. 13) and its legality is questioned (Johnson, 2023a). The likely reason for it relates to the construction of the reference package, wherein the algorithm lacks the ability to commodate more than one category of impairment. Consequently, the potential interaction https://onlinelibrary.wiley.com/terms-and-conditions of multiple disability or health/mental health conditions is not taken into account. This has on consequences for the nature and degree of entitlement. The concept of defining an individual’s support needs solely based on a “primary” disability Online is not only ableist in its reductionism but also heavily gendered (Charitsis & Lehtiniemi, 2023, Library p. 12). For instance, women with disabilities experience multiple and gendered forms of dis- able, some intertwined with chronic illness, pain, violence- related injury, psychosocial dis- ability and mental health issues (Yates et al., 2021). These disabilities may overlap and change for over time, making it difficult to designate a single “primary” disability. Unfortunately, it may rules not come as a shock that women have a 26 per cent lower chance of qualifying for NDIS sup- of ports (NDIA, 2019), as the scheme’s algorithmic architecture fails to adequately capture the use;OA complex interplay of gender and disability. The problem with this data- driven approach is articles that it is impossible to extrapolate from a generalised profile what supports a person needs to are to flourish in life, especially if that person doesn’t conform to the statistical “norm.” People with governed disability represent a highly diverse group, with various patterns of impairment intersecting the with gendered, ethnic, racial, religious, sexual and class identities, adding layers of complexity
VAN TOORN AND CARNEY
redacted: s47F - Personal privacy
To the disability experience. Assessing individuals based on their “primary” disability perpetuates a reductionist and simplistic view of disability. It raises the bar for access for those with intersecting health and mental health conditions and physical and intellectual disabilities. But it is not only these groups affected, for as Charitsis and Lehtiniemi argue, when social benefits and public services become increasingly conditional on individuals conforming to rigid data categories that belie their multifaceted identities and life situations, “almost anyone can fail to access services and end up at the receiving end of disciplinary and control mechanisms” (Charitsis & Lehtiniemi, 2023, p. 12).
by | Human decision makers and accountability | National Health And Medical Research Council, Wiley Online Library onlinelibrary.wiley.com/doi/10.1002/ajs4.342 [31/05/2024].
As the Robodebt Royal Commission observed, human engagement is critical for many citi- zens unaccustomed to dealing with government or experiencing vulnerabilities, leading it to recommend “More ‘face- to- face’ customer service support options should be available for vulnerable recipients” (rec. [13.3]). Because of well-known local and international fiascos with algorithms and automation in welfare (Bouwmeester, 2023), heightened scrutiny of algorithms and automation in the NDIS is certainly called for. Across diverse areas of automation, this scrutiny often centres around ensuring adequate human involvement in decision making to ensure accountability. The concept of the “human in the loop” is seen as an important oversight mechanism and risk mitigator for medium and high-risk systems (Jones, 2017), aligning with a broader trend in the automation literature that scrutinises and emphasises the role of human agency, especially in areas of high-stakes decision making (Gillespie, 2014; Kitchin, 2017; Pasquale, 2015; Wagner, 2019). Accountability is closely tied to the presence and participation of human deci-sion makers, aiming to ensure that human judgement plays a meaningful role in the decision-making process (Wagner, 2019). This idea is expressed in the Productivity Commission’s recommendation that disability assessments should not merely be accepted without consideration (“rubber stamped”), and any plans that deviate from the established “benchmark” of assessed needs for individuals with similar characteristics must undergo review by a more senior NDIA administrator (Productivity Commission, 2011, pp. 20–21). Hence, the decision process most certainly involves human discretion, as an NDIA planner is responsible for making adjustments and approving the final plan. Nevertheless, this casts doubt on whether having a human in the loop is adequate to guarantee fairness and accountability (for a critical examination of policies requiring a human in the loop, see Green, 2022a). Reuben Binns (2022, pp. 205, 208) argues persuasively that in the case of individualisation, algorithmic input undermines and is incompatible with exercise of the required human dis-cretion, particularly with lower-level decision makers. While “statistical credibility” might be thought to be achievable in this context, external review mechanisms like the AAT provide some insight into the practical adequacy and acceptability of decisions. The increasing num- er of AAT appeals in recent years suggests that despite human involvement, decisions are being made in an inconsistent and, some argue, unfair manner, neglecting the specific needs and preferences of individuals involved (Collings et al., 2016; Henriques-Gomes, 2022; Perry et al., 2019; Smith-Merry et al., 2023; St Guillaume et al., 2021). In this case, as in numerous other domains employing decision support tools, the issue of accountability becomes complex due to administrators’ enmeshment within algorithmic processes, making it unclear where the line lies between human discretion and automation. Because planners are closely enmeshed within algorithmic processes, it makes little sense to attribute flaws in decision making solely to either the human or machine components, fol-low-ing the traditional binary liability model (Marchant & Lindor, 2012). As Wagner (2019,
Legality, procedural fairness and review on the merits
At first blush, it appears perverse to argue that the use of algorithms in NDIS decision making constitutes an “algorithmic grey hole” in the sense of providing the appearance but not the substance of rights of legal redress (Solow-Niederman, 2023). Australia rightly prides itself for its robust judicial avenues for testing whether decisions are legal and whether affected parties received a fair and unbiased hearing (once called “natural justice” now procedural fairness).
And citizens have ready access to an independent tribunal (currently called the “AAT” until renovated in 2024) whose task is to “step into the shoes” of the original decision maker to review the decision afresh and on its merits to impose (unless overturned on appeal) its own assessment of what is the “correct and preferrable” decision. AAT merits review is conducted not in a passive but instead in a more “inquisitorial” manner. The hearing is more flexible, free of unduly technical rules about evidence or procedure applying in courts, but nevertheless is obliged to accord procedural fairness.
For example, it was a denial of procedural fairness for the AAT to doubt, without hearing from an NDIS applicant in person, that funding his mother was the only acceptable way of providing overnight care and management of his seizure risk, and thus the only “effective and beneficial” form of reasonable and necessary support (Klewer, 2023, esp. para [113]).
As we seek to show, all of these avenues in theory are open as a check on NDIS decision making. Yet, all three avenues are quite illusory in practice, constituting classic examples of algorithmic grey holes. Consideration of legal accountability, procedural fairness scrutiny or AAT merits review of the role of algorithms in NDIS decision making is only possible be- cause the NDIS Act does not contain a provision validating a decision made solely by computer. Consequently, all NDIS decisions are open to judicial challenge or AAT review (Ng & O’Sullivan, 2019). However, neither courts nor the AAT can consider a case until a “final” decision has crystallised—preliminary inputs, such as the contribution of TSPs toward a final NDIS plan, do not qualify as a “decision” (the High Court ruling in Bond, 1990).
Use of algorithms as an aid to decision making is expressly legitimated by NDIA operational guidelines (the Supports Rule, 2013), and no suggestion has been made that this is unlawfully in excess of the rule-making powers of the Act (it is not suspect as being ultra vires).
As now discussed, however, few if any of the remaining avenues hold prospects of meaningful restraint on NDIS use of algorithms. In non-technical terms, this is because: (a) the contribution of NDIS algorithms toward typical budgets (including nomination of a primary disability) is too preliminary to qualify for judicial review and any contribution toward ultimate legal error too difficult to prove, with any unlawful fettering of caseplanner’s discretionary decisions being too subtle or remote, and there is no way of knowing or evidencing any fettering; (b) it is not possible to establish any breach of the duty to fully fund “reasonable and necessary supports”; and (c) while AAT merits review itself is never affected by earlier algorithmic contributions to primary decision making because it is a de novo reconsideration, rectification of an individual case provides no systemic recourse to substantive fairness for those in a similar position unless each also exercises their appeal rights.
4.3 | Legal considerations
4.3.1 | The legality of NDIS “typical budget” planning
Concern has been expressed by Parliamentary committees and others that NDIS practice of requiring applicants to nominate their primary and any secondary impairments is contrary to law (Johnson, 2023a), or that the typical budget generated by the algorithms constrains or leads to miscarriage of planners’ ultimate decision making. With regard to the distribution of responsibility (or “division of labour”) between an algorithmic input and the role of NDIS planners, the legal position is that the decision maker must not be subject to a “fetter” that takes away the choices (or “discretions”) legally required to be considered (Chiam, 2019). However, an administrative arrangement that gives an initial case planner authority to manually vary the presumptive TSP by a given percent- age (in either direction) and refer larger variations for consideration by a more senior planner would not breach that rule. The same is true of purely human systems: a “blan- et” administrative directive insisting that planners always choose one of several legally available options would breach the rule, but gross under- resourcing and overworking of administrators—such as that experienced during the early transition of the scheme up to mid-2017 as the volume of work resulted in perfunctory telephone “planning” (Carney et al., 2019, p. 786, 794)—would not do so. Likewise for concerns that planners are in- centivised by Key Performance Indicators to conduct planning meetings quickly, with limited interaction with participants, leading to the use of automated reference packages to reduce caseload burden, and prioritising efficiency over customising support plans (Johnson, 2023b). Even where decisions are subject to human review, algorithmic sys- tems and the surrounding organisational environment will still “nudge and influence the judgement of overworked [case] workers, as well how they justify their discretion to themselves and others” (Hong, 2023, p. 6).
4.3.2 | The “reasonable and necessary support” test
For individual NDIS participants, what matters is what is included as a funded support in their plan and how adequately it is funded. The legislation frames this as a judgement about what are “reasonable and necessary supports” (NDIS Act s. 34). The rules made under the Act elaborate the meaning of the phrase (NDIS Supports Rule, 2013) and their inclusion in a plan (NDIS Plan Management Rules, 2013, Pt 6). So, can legal recourse be obtained on the basis that algorithmically generated typical plans are incompatible with this obligation to ensure “reasonable and necessary supports,” such as by expecting families or mainstream services to meet the portion of the costs (so- called NDIS “boundary issues”)? The most definitive guidance on the phrase came from Mortimer J at first instance in McGarrigle, writing that it takes its meaning from its context, “especially … s 4(11), which sets out what reasonable and necessary supports should enable and empower people with a dis- ableity to do, read with s 14 which sets out the purposes for which funding for reasonable and necessary supports is provided” (McGarrigle, 2017a, para [41]). However, this does not mean that alternative sources of support—such as from mainstream health and welfare services, Tier 2 state and territory services, or voluntary family and civil society supports—may not be more appropriate in a given case. The Full Court, in endorsing referral back for reconsideration by the AAT, merely remarked that “:[g]iven the potential systemic importance of the issues sought to be raised” a three member AAT panel headed by a Presidential Member might be appro- riate (McGarrigle, 2017b, para [8]). Mortimer’s ruling stands. Simply stated it is that once a decision is made that an item is a reasonable and necessary inclusion in a plan, this serves as a “gateway” into fully funding that item of support. It would be a legal error to then reduce
VAN TOORN AND CARNEY
13
| | 18394655, ———————WILEY- |
to be funded on the basis of, say, a capacity for family contributions (McGarrigle, 2017a, para [95] (Mortimer J)).
Recourse to the courts on the basis that the contribution of TSPs to the final funding de- cision contravenes this ruling is not in any practical sense open, because those algorithmic inputs generate no legally relevant errors in the ultimate funding decision either with respect to the “boundary” issues, or the full funding of reasonable and necessary supports.
Remedying Algorithmic Injustice Through AAT Merits Review
External review by the AAT of access or planning decisions grapples afresh with the merits or otherwise of the factual and legally relevant dimensions of the decision. The review is in no way constrained by any preceding unfairness or other discriminatory impacts of NDIA algorithms; the algorithms are not relevant to or in play in AAT review, but nor would it be practically possible to interrogate their role in the primary decision being reviewed.
AAT merits review necessarily includes reconsidering the “weight” to be given to com- eting factors, such as operationalisation of the “Goldilocks” tension in NDIS design be- tween personalisation and sustainability under the lifecourse “insurance principle”—a balance between planning decisions that are overly hard (inflexible/arbitrary, ousting per- onalisation) and overly soft (idiosyncratic, erratic and lacking consistency) (Productivity Commission, 2011, pp. 20–21). That tension is reflected in the two arms of NDIS Supports Rule (2013, r 3.1(b), (c)), that:
calculation whether proposed supports are ‘reasonable and necessary’ under the scheme, the decision- maker must consider, amongst other matters: ‘whether there is evidence that the support will substantially improve the life stage outcomes for, and be of longterm benefit to, the participant’; and ‘whether funding or provision of the support is likely to reduce the cost of the funding of supports for the partic- ipant in the long term’. (Ng & Gray, 2022, p. 654)
However, all that is legally required is consideration of both factors, not that any particular weight be given to either. Recourse to the AAT as a response to NDIA algorithms is an illusory (grey hole) remedy beyond justice for that individual, because unlike a judicial ruling, AAT review does not result in any systemic change in the way algorithms are used. It currently offers no more than nor- mative guidance reconciling conflicting values and principles of the NDIS. Despite manifold other concerns detailed earlier, algorithmic inputs into NDIS case planning do not speak at all to “boundary” questions at the AAT. Nor do they disadvantage applicants when developing an evidence-based case for additional or better-funded reasonable and necessary supports being sought on review. Instead, for a more “vulnerable” participant such as someone lacking the social capital to advocate strongly for their needs, or the socioeconomic resources to come to the NDIS armed with a suite of recent medical and other reports, it is the inequity and unfairness of being short-changed by NDIS primary decision making. Putting it another way, the unfairness lies in needing to apply to the AAT at all. Substantive unfairness stems from the way, in the absence of other documentary/evidentiary material or avenues for obtaining it, undue weight given to algo- rithmic inputs (uncorrected for by information elicited by human case planners with adequate time and expertise) deprives those participants from obtaining their just entitlements under the personalised planning process.
CONCLUSION: TOWARD ACCEPTABLE NDIS ALGORITHMS?
This article reveals that NDIS algorithmic systems, used as supporting inputs in determining the content and funding of participants’ plans, risk replicating well-documented instances of algorithmic harm experienced in other areas of social welfare provision. We refer to these systems, and the nature of their decision-making powers, as “subterranean,” because their workings are neither publicly known nor amenable to legal rectification in the way Australia’s Robodebt or the Dutch Childcare Benefit harms were rendered accountable (Bouwmeester, 2023). These systems exercise forms of state power, yet their subterranean administrative character positions them in an “algorithmic grey hole” (Solow-Niederman, 2023) effectively beyond the reach of legal remedies; and one which significantly compromises the delivery of substantive algorithmic fairness by failing to routinely engage the contextual complexities of the disability experience.
Subterranean systems present significant threats of algorithmic harm in various social ser- vices, including Australia’s use of the Job Seeker Classification Instrument used in employ- ment services (DEWR, 2020). Further research is needed to develop a deeper understanding of the many dimensions of subterranean risk to the realisation of substantive fairness, whether within the NDIA’s administrative procedures or more generally, and how best to address that danger. In this context, “formal” algorithmic fairness, which focuses on treating individuals equally based on their attributes, seems insufficient in meeting the standards of “substantive” algorithmic fairness (Green, 2022b), which instead seeks to address social injustices, specifically the exclusion or marginalisation of people with disabilities from decision making. The as yet nascent concept of “algorithmic justice” offers insights into what might be entailed in realising that ideal (Marjanovic et al., 2022). Algorithmic justice broadens consideration in at least three ways: (i) in its focus on ADM harms and an accompanying theory of justice going beyond traditional formal equality of distributional access to “social goods”; (ii) in its concentration on systemic as distinct from merely individual impacts (picking up not only economic but socio-cultural and political dimensions of ADM); and (iii) in its interest in the algorithmic process rather than just various distortions in data sets, and in restoration of the full human subject to counter “datafication” of subjects (ibid, 281).
Three preliminary conclusions about the way algorithmic justice and substantive algo- rithmic fairness might be advanced for NDIS participants can tentatively be drawn from our scoping inquiry in this article. Our first conclusion is that immediate action is required to en- sure that primary decision-making processes at minimum reflect the precepts laid down in the Robodebt Royal Commission (2023, rec [17.1]) and since accepted by the government, namely that: there should be a “clear path” of review for those affected by ADM; that agency Websites “contain information advising that automated decision-making is used and explaining in plain language how the process works”; and that “business rules and algorithms should be made available, to enable independent expert scrutiny.” The path for obtaining AAT merits review of NDIA decisions has been available from the inception, and access to it is now a “clear” route following the removal of previous barriers.⁉ However, more could be done to ensure easy, accessible and effective internal NDIA review. Our examination shows a close to zero compliance with the second and third of these standards. NDIA Websites do not currently apprise readers that automation is used in decision making, or the processes involved, much less in “plain lan-guage.” The complete veil of secrecy around NDIS “business rules and algorithms” is another serious deficiency, giving rise to actual and perceived bias and discrimination, particularly against people with multiple marginalised identities (van Toorn & Scully, 2023).
Second, given the lack of any, much less any genuine “co-design” (Dickinson & Yates, 2023) in the development of existing NDIA algorithms and the credible claims that tailoring NDIS participants’ support to their unique and diverse circumstances can only be accomplished by
VAN TOORN and CARNEY 15 | 18394655, ———————WILEY- 0,
skilled human case planners (Binns, 2022), precautionary measures are warranted. NDIA op- Downloaded
erational guidelines should be reviewed to ensure retention of adequate human decision mak- from
ing, and independent research should urgently be commissioned into the inner workings of
current arrangements, and the findings made public as a basis for public debate on continued
use of such algorithms.
Third, scrutiny of algorithms alone is unlikely to affect meaningful change if it neglects
the socio- bureaucratic context in which they operate. The NDIS functions as a large, bu-
reaucratic infrastructure, guided largely by principles of standardisation and efficiency in
its approach to technological innovation. While human planners do retain some discretion, NDIS processes and systems essentially conform to a well- established technocratic logic https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342
where the pursuit of fairness is simplistically equated with straightforward standardisation by
in how individuals are processed (or treated equally) by algorithms. It is essential, we argue, National
to spotlight these subterranean operational aspects, with a specific focus on involving the Health
public, especially people with disability, in rethinking the role of algorithms. This goes And
beyond surface attempts at co- design. Rather, it necessitates people with disability having Medical
a meaningful say in negotiations around NDIS resources and frameworks governing their Research
distribution. It is only in such circumstances that systems of resource allocation, whether
algorithmically augmented or otherwise, can genuinely align with the principles of disabil- Council,
ity justice. Wiley
Due to the subterranean and pervasive character of NDIA algorithms giving rise to Online
algorithmic grey holes, we conclude that they are less suited to redress through traditional Library
judicial or tribunal accountability reforms and are more adequately rendered accountable on
under rendered accountable through the structural changes advocated by algorithmic jus- [31/05/2024]. tice scholarship.
See
the
AUTHOR CONTRIBUTIONS Terms
Georgia van Toorn: Conceptualisation; writing – original draft; writing – review and editing. and
Terry Carney: Conceptualisation; writing – original draft; writing – review and editing. Conditions
ACKNOWLEDGEMENTS
We are deeply indebted to incisive suggestions by two anonymous referees which sharpened
the conceptualisation and form of our argument, though we bear responsibility for any
errors. Open access publishing facilitated by University of New South Wales, as part of the
Wiley - University of New South Wales agreement via the Council of Australian University
Librarians. (https://onlinelibrary.wiley.com/terms-and-conditions)
FUNDING INFORMATION on
This research was funded by the Australian Research Council Centre of Excellence for Wiley
Automated Decision-Making and Society (grant number: CE200100005). Online
Library
CONFLICT OF INTEREST STATEMENT for
rules
The authors have no conflicts of interest to declare. All co- authors have seen and agree with of
the contents of the manuscript and there is no financial interest to report. use;OA
articles
ENDNOTES are
1 As expressed in the Explanatory Memorandum to the 2012 Bill: ‘The criteria listed in clause 34 balances what support is governed
‘necessary’ to assist the participant to pursue their goals (in accordance with the participant's statement of goals and aspi- by
rations) and to facilitate their social and economic participation, with what is ‘reasonable’, including whether the cost of the
the support represents value for money and is reasonable, the efficacy of the support, whether it is not reasonable to expect applicable families and carers to provide the support, and whether the support would be more appropriately provided by other main-
stream services’: Explanatory Memorandum, National Disability Insurance Scheme Bill 2012 available at https://www.legis Creative
lation.gov.au/Details/C2012B00230/Explanatory%20Memorandum/Text. Commons
16 VAN TOORN and CARNEY | 18394655, -WILEY——————— 0, 2 A 2015 ruling of the Federal Court in Mulligan in a loose sense does insist that supports be assessed on the basis of how the NDIS “finds” the participant. This obligation, however, applies to later stages of the planning process in Downloaded which the case planner assesses the functional impacts of disability for each participant. To establish legal error, it from would be necessary to show that an initial nomination of a primary impairment results in a TSP and ultimate fund- ing package that fails to consider impacts in all relevant functional domains (communication; social- interaction; learning; mobility; self- care; self- management) or unlawfully narrows consideration of how the impairment ‘af- fect[s] the person’s capacity for social or economic participation’. In other words, the nomination of one impair- ment for the purpose of running the algorithm (or for any other purpose) must be shown to detract from/impact on the ultimate responsibility for crafting an assessment package that meets all the legally relevant functional dimen- sions laid down in Mulligan. In a lay sense, this may appear achievable, but proof to the legal standard is doubtful, in part due to the opacity of the process. 3 A key barrier in this respect was the inability to review a plan superseded by a later one on its expiration. https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342 by National REFERENCES Health Ardill, A. & Jenkins, B. (2020) Navigating the Australian National Disability Insurance Scheme: a scheme of big And ideas and big challenges. Journal of Law and Medicine, 28(1), 145–164. Medical Australian Government. (2021) National Disability Insurance Scheme oversight, Tuesday 18 May, 2021, Canberra. Australian Parliament. Available from: https:// parli nfo. aph.gov. au/ parlI nfo/search/displ ay/displ ay.w3p;query= Research Id% 3A% 22com mitte es% 2Fcom mjnt% 2Ff02 5947f- e54b- 4d95- a2b8- 68dde e7dec c3% 2F0000% 22 [Accessed 6th Council, June 2022]. Australian National Audit Office [ANAO]. (2020) Decision- making controls for NDIS participant plans. Canberra, Wiley ACT: Australian National Audit Office. Available from: https:// www. anao.gov. au/ work/ perfo rmanc e- audit/ Online decis ion- makin g- contr ols- ndis- parti cipant- plans [Accessed 3rd December 2020]. Library Australian National Audit Office [ANAO]. (2023) Effectiveness of the National Disability Insurance Agency’s management on of assistance with daily life supports. Canberra, ACT: Australian National Audit Office. Available from: https://www. anao.gov.au/sites/defau lt/files/2023- 06/Auditor- Gener al_ Report_ 2022-23_ 43.pdf [Accessed 17th October 2023]. [31/05/2024]. Bellamy, R.K.E., Dey, K., Hind, M., Hoffman, S.C., Houde, S., Kannan, K. et al. (2019) AI Fairness 360: An extensible toolkit for detecting and mitigating algorithmic bias. IBM Journal of Research and Development, Seethe 63(4/5), 15. Terms Bennett, C. & Keyes, O. (2019) What is the point of fairness? disability, AI and the complexity of justice. ACM and SIGACCESS Accessibility and Computing, 125, Article 5. Binns, R. (2022) Human Judgment in algorithmic loops: Individual justice and automated decision- making. Conditions Regulation & Governance, 16(1), 197–211. Bond. (1990) Australian Broadcasting Tribunal v Bond and others (1990) 170 CLR 321. Bonyhady, B. (2021) An analysis of the NDIA’s proposed approach to independent assessments. Melbourne, VIC: Melbourne Disability Institute. Available from: https:// www. aph.gov. au/ Docum entSt ore. ashx? id=29b30 746- 40dd- 45ab- b16b- a5ff4 0f24e e3& subId =705928 [Accessed 8th September 2021]. Bouwmeester, M. (2023) System Failure in the Digital Welfare State: Exploring parliamentary and judicial control in the Dutch childcare benefits scandal. Recht der Werkelijkheid, 44(2), 13–37. Braithwaite, V. (2020) Beyond the Bubble that is Robodebt: How governments that lose integrity threaten democ- racy. Australian Journal of Social Issues, 55(3), 242–259. Campanella, N., Thorne, L. & Edmonds, C. (2023) NDIS treated as a “Magic Pudding” by governments and disabil- (https://onlinelibrary.wiley.com/terms-and-conditions) ity service providers, review co- chair says. ABC News, 22 August 2023. Available from; https:// www. abc. net. au/ on Wiley news/ 2023- 08-22/conce rns- about- ndis- raise d- by- co- chair- of- review/102758296 Carney, T. (2006) Social security law and policy. Sydney, NSW: Federation Press. Online Carney, T. (2018a) The New Digital Future for Welfare: Debts without legal proofs or moral authority? UNSW Law Library Journal Forum, 2018, 1–16. for Carney, T. (2018b) Robo- Debt Illegality: A failure of rule of law protections. AUSPUBLAW. Available from: https:// rules auspu blaw.org/ 2018/ 04/ robo- debt- illeg ality/ [Accessed 8th November 2023]. of use; Carney, T., Then, S.-N., Bigby, C., Wiesel, I. & Douglas, J. (2019) National Disability Insurance Scheme Decision- OA making: Or when tailor- made caseplanning met taylorism and the algorithms? Melbourne University Law articles Review, 42(3), 780–812. are Charitsis, V. & Lehtiniemi, T. (2023) Data ableism: ability expectations and marginalization in automated societies. governed 24(1), 3–18. Television and New Media, Chiam, C. (2019) The future of the fettering rule in judicial review. University of Tasmania law Review, 38(1), bythe 27–46. Collings, S., Dew, A. & Dowse, L. (2016) Support planning with people with intellectual disability and complex applicable support needs in the australian national disability insurance scheme. Journal of Intellectual and Developmental Creative Disability, 41(3), 272–276.
VAN TOORN and CARNEY 17 | 18394655, ———————WILEY- 0,
Considine, M. (2022) The Careless State: reforming Australia's social services. Melbourne, VIC: Melbourne University Press. Downloaded
DEWR. (2020) Assessments guideline – Job Seeker Classification Instrument (JSCI) and Employment Services from
Assessment (ESAt). Canberra, ACT: Department of Employment and Workforce Relations. Available from:
https:// www. dewr. gov. au/ new- emplo yment - servi ces- model/ resou rces/ new- emplo yment - servi ces- trial - nest-
enhan ced- servi ces- asses sment s- guide line- job- seeke r- class ifica tion [Accessed 8th November 2023].
Dickinson, H. & Yates, S. (2023) A decade on: the achievements and challenges of the National Disability Insurance
Scheme's implementation. Australian Journal of Social Issues, 58, 460–475. Available from: https://doi.org/10.
1002/ajs4. 277
Dickinson, H., Yates, S., Smith, C. & Doyle, A. (2021) Avoiding simple solutions to complex problems: independent
assessments are not the way to a fairer NDIS. Melbourne, VIC: Children and Young People with Disability Australia. Available from: https://apo.org. au/sites/defau lt/ files/ resou rce- files/ 2021- 05/apo- nid312281.pdf https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342
Fishkin, J. (2014) Bottlenecks: a new theory of equal opportunity. Oxford: Oxford University Press. by
Foster, M., Henman, P., Tilse, C., Fleming, J., Allen, S. & Harrington, R. (2016) ‘Reasonable and necessary’ National
care: the challenge of operationalising the NDIS policy principle in allocating disability care in Australia.
Australian Journal of Social Issues, 51, 27–46. Available from: https://doi. org/10.1002/j.1839- 4655. 2016. tb003 Health
63. x And
Friedler, S.A., Scheidegger, C. & Venkatasubramanian, S. (2021) The (Im)possibility of fairness: different value Medical
systems require different mechanisms for fair decision making. 64(4), 136–143. Communications of the ACM,
Gillespie, T. (2014) The relevance of algorithms. In: Gillespie, T., Boczkowski, P. & Foot, K. (Eds.) Media technolo- Research
gies: essays on communication, materiality, and society. Cambridge, MA: MIT Press, pp. 167–194. Council, Gooding, P. & West, A. (2023) Disability assessments and the algorithmic veil: lessons from the abandoned ‘indepen-
dent assessments’ proposal for the National Disability Insurance Scheme. Australian Journal of Human Rights, Wiley
29(1), 44–64. Available from: https://doi.org/10.1080/1323238X. 2023. 2203536 Online
Gosepath, S. (2021) Equality. The stanford encyclopedia of philosophy. Available from: https://plato.stanford.edu/ Library
entries/equality/ [Accessed 4th October 2024]. on
Government. (2024) Budget Paper no. 1, 14 May. Available from: https:// budget.gov. au/conte nt/ bp1/downl oad/ bp1_
2024-25.pdf [Accessed 19 May 2024]. [31/05/2024]. Graycar, A. & Masters, A. (2022) Bureaucratic Bastardry: Robodebt/debt recovery, AI and the stigmatisation of
See
citizens by machines and systems. International Journal of Public Policy, 16(5–6), 333–344. the
Green, B. (2022a) The flaws of policies requiring human oversight of government algorithms. Computer Law and Terms
Security Review, 45, 105681. and
Green, D. (2022b) Escaping the impossibility of fairness: from formal to substantive algorithmic fairness. Philosophy
and Technology, 35, Article 90. Conditions
Hanna, A., Denton, E., Smart, A. & Smith-Loud, J. (2020) Towards a critical race methodology in algorithmic fairness.
A paper presented at the Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency,
Barcelona, Spain.
Henriques- Gomes, L. (2021) NDIS independent assessments should not proceed in current form, Coalition's own
advisory council says. The Guardian Australia. Available from: https:// www.thegu ardian. com/austr alia- news/
2021/ jul/ 08/ ndis- indep enden t- asses sment s- shoul d- not- proce ed- in- curre nt- form- coali tions - own- advis ory-
counc il- says [Accessed 8th July 2021].
Henriques- Gomes, L. (2022) NDIS minister downplays major increase in appeals since 2021 as “Tiny Fraction of
Participants”. The Guardian Australia. Available from: https:// www.thegu ardian. com/socie ty/ 2022/ may/ 09/ ndis- minis ter- downp lays-244- incre ase- in- appea ls- since -2021- as- tiny- fract ion- of- parti cipants [Accessed 9th (https://onlinelibrary.wiley.com/terms-and-conditions)
May 2022]. on
Wiley Hoffmann, A.L. (2019) Where fairness fails: Data, algorithms, and the limits of antidiscrimination discourse.
Information, Communication & Society, 22(7), 900–915. Online
Hong, S.H. (2023) Prediction as extraction of discretion. Big Data & Society, 10(1). Available from: https://doi.org/ Library
10.1177/ 20539517231171053 for
Johnson, M. (2021) “Citizen- centric” demolished by NDIS algorithms. InnovationAus. Available from: https:// www. rules
innov ation aus. com/citiz en- centr ic- demol ished - by- ndis- algor ithms/ [Accessed 15th November 2022]. ofuse;
Johnson, M. (2022) The Hunger Games created by NDIS algorithms. InnovationAus. Available from: https:// www. OA
innov ation aus. com/ the- hunge r- games - creat ed- by- ndis- algor ithms/ [Accessed 8th November 2023]. articles
Johnson, M. (2023a) Committee bombshell: NDIA has been operating unlawfully. InnovationAus. Available from: are
https:// www. innov ation aus. com/ commi ttee- bombs hell- ndia- has- been- opera ting- unlaw fully/ ? trk= feed- detail_ governed main- feed- card_ feed- artic le- content [Accessed 8th November 2023].
Johnson, M. (2023b) Supplementary submission to Joint Parliamentary Standing Committee on the NDIS Senate bythe
Canberra, ACT. Available from: https:// www. aph. Inquiry “Capability and Culture of the NDIA”. gov. au/ applicable Docum entSt ore. ashx? db365 fea- d327- 4683- a9f2- 1f72d e0cbf 74& [Accessed 8th November id= subId =722398
2023]. Creative
References
VAN TOORN and CARNEY | 18394655, -WILEY——————— 0,
Joint Standing C’tee on NDIS. (2021a) Independent assessments. Canberra, ACT: Joint Standing Committee on the National Disabilty Insurance Scheme. Available from: https:// parli nfo. aph.gov. au/ parlI nfo/downl oad/commi Downloaded ttees/ commj nt/ 41d2f 136- 9426- 4d05- affc- 4e12f fee45 4c/ toc_ pdf/ Joint% 20Sta nding% 20Com mittee% 20on% from 20the% 20Nat ional% 20Dis abili ty% 20Ins urance% 20Sch eme_ 2021_ 05_ 04_ 8734_ Offic ial. pdf; fileType= appli cation% 2Fpdf# searc h=% 22com mitte es/commj nt/41d2f 136-9426- 4d05- affc- 4e12f fee45 4c/ 0000% 22 [Accessed 30th December 2021]. Joint Standing C’tee on NDIS. (2021b) Independent assessments. Canberra, ACT: Joint Standing Committee on the National Disability Insurance Scheme. Available from: https:// parli nfo. aph.gov. au/ parlI nfo/downl oad/commi ttees/ repor tjnt/ 024622/ toc_ pdf/ Indep endentAsse ssments.pdf; fileType= appli cation% 2Fpdf [Accessed 30th December 2021]. Jones, M. (2017) Right to a human in the loop: political constructions of computer automation & personhood from data banks to algorithms. Social Studies of Science, 47(2), 216–239. https://onlinelibrary.wiley.com/doi/10.1002/ajs4.342 Kitchin, R. (2017) Thinking critically about and researching algorithms. Information, Communication & Society, 20(1), 14–19. Klewer. (2023) Klewer v National Disability Insurance Agency [2023] FCA 630. Available from: https:// www. austl ii. Health edu. au/ cgi- bin/ viewd oc/au/ cases/ cth/ FCA/ 2023/ 630. html? conte xt=1;query = klewer; mask_ path= au/ cases/ cth/ FCA [Accessed 8th November 2023]. And Marchant, G. & Lindor, R. (2012) The coming collision between autonomous vehicles and the liability system. Santa Medical 52(4), 1321–1340. Clara Law Review, Marjanovic, O., Cecez-Kecmanovic, D. & Vidgen, R. (2022) Theorising algorithmic justice. European Journal of Research Information Systems, 31(3), 269–287. Council, McGarrigle. (2017a) McGarrigle v National Disability Insurance Agency [2017] FCA 308. Available from: https:// Wiley www. austl ii. edu. au/ cgi- bin/ viewd oc/ au/ cases/ cth/ FCA/ 2017/ 308. html? conte xt=1; query = McGar rigle; mask_ path= au/cases/cth/ FCA Online McGarrigle. (2017b) National Disability Insurance Agency v McGarrigle [2017] FCAFC 132 (Kenny, Robertson and Library Kerr JJ). Available from: https:// www. austl ii. edu. au/cgi- bin/ viewd oc/au/cases/cth/ FCAFC/2017/132. html?conte on xt=1;query =Natio nal% 20Dis abili ty% 20Ins urance% 20Age ncy% 20v% 20McG arrig le; mask_ path= au/cases/cth/ FCAFC [Accessed 8th November 2023]. [31/05/2024]. National Disability Insurance Agency [NDIA]. (2015) National Disability Insurance Scheme Annual financial sus- See tainability report 2014–15. Canberra, ACT: National Disability Insurance Agency. Available from: https://data. ndis.gov. au/ media/3474/downl oad?attac hment [Accessed 8th November 2023]. Terms National Disability Insurance Agency [NDIA]. (2016) Insurance principles and financial sustainability manual. Canberra, ACT: National Disability Insurance Agency. National Disability Insurance Agency [NDIA]. (2019) Analysis of participants by gender. Canberra, ACT: National Conditions Disability Insurance Agency. Available from: https://data. ndis.gov. au/ media/2120/downl oad?attac hment [Accessed 8th September 2021]. National Disability Insurance Agency [NDIA]. (2021) Personalised Budgets Proposal for a new NDIS budget model. Technical information paper. NDIS Plan Management Rules. (2013) National Disability Insurance Scheme (Plan management) rules 2013. Available from: https:// www. legis lation.gov. au/ Detai ls/ F2017C00101 [Accessed 13th September 2014]. NDIS Supports Rule. (2013) National Disability Insurance Scheme (Supports for participants) rules 2013. Available from: https:// www. legis lation.gov. au/ Detai ls/ F2013L01063 [Accessed 13th September 2014]. Needham, C. & Dickinson, H. (2018) “Any one of us could be among that number”: comparing the policy narratives for individualized disability funding in Australia and England. Social Policy & Administration, 52(3), 731–749. (https://onlinelibrary.wiley.com/terms-and-conditions) Ng, Y.- F. & Gray, S. (2022) Disadvantage and the automated decision. Adelaide Law Review, 43(2), 641–677. on Wiley Ng, Y.-F. & O’Sullivan, M. (2019) Deliberation and automation – when is a decision a decision? Australian Journal of Administrative Law, 26(1), 21–34. Online Nikidehaghani, M., Andrew, J. & Cortese, C. (2023) Algorithmic accountability: robodebt and the making of welfare Library cheats. Accounting, Auditing & Accountability Journal, 36(2), 677–711. for O’Donovan, D. (2019) Righting the wrongs of robodebt. Eureka Street, 29(23), 11–13. Pasquale, F. (2015) The Black Box Society: the secret algorithms that control money and information. Cambridge, MA: Harvard University Press. Perry, E., Waters, R. & Buchanan, A. (2019) Experiences of National Disability Insurance Scheme planning from the articles perspective of adults with intellectual disability. Australian Journal of Social Issues, 54(3), 210–223. are Productivity Commission. (2011) Canberra, ACT: Productivity Commission. Available Disability care and support. from: https:// www. pc.gov. au/ inqui ries/ compl eted/disab ility- suppo rt/ report/ disab ility - suppo rt-volum e1.pdf governed [Accessed 13th September 2014]. bythe Productivity Commission. (2017) Canberra, ACT: Productivity National Disability Insurance Scheme (NDIS) costs. Commission. Available from: http:// www. pc.gov. au/ inqui ries/curre nt/ ndis- costs/ posit ion/ ndis- costs - posit ion. applicable pdf [Accessed 21st February 2019]. Creative
VAN TOORN AND CARNEY
19
| | 18394655, —WILEY- | Robodebt Royal Commission. (2023) Report: Royal Commission into the Robodebt Scheme. Canberra, ACT. Available from: hhttps://robod ebt.royal commi ssion.gov. au/system/ files/2023- 07/repor t- of- the- royal - commisson - into- the- Downloaded robod ebt- scheme.pdf [Accessed 18th August 2022].
Smith-Merry, J., Gilroy, J. & Watharow, A. (2023) The NDIS at ten years: designing an equitable scheme for the next decade. Medical Journal of Australia,
218, 291–294.
Solow-Niederman, A. (2023) Algorithmic grey holes. Journal of Law & Innovation, 5(1), 117–139.
St Guillaume, L., Coe, G. & Murray, M. (2021) An evaluation of NDIS appeals and review: experiences of culturally and linguistically diverse people with disability, their families and carers. Sydney: The Multicultural Disability Advocacy Organisation NSW/University of Notre Dame Australia. Available from:
hhttp://mdaa.org.au/wp- conte nt/uploa ds/
2021/09/NDIS-Review-Appea ls-Proce ss-Report_ FINAL_ SEPTEMBER-2021.pdf [Accessed 18th August
2021].
Taylor-Fry. (2021) Review of NDIA actuarial forecast model and drivers of Scheme costs. Canberra: Department of Social Services. Available from:
hhttps:// www.dss.gov. au/sites/defau lt/ files/docum ents/ 01_ 2022/ revie w- ndia-
actua rial- forec ast- model - and- drive rs- schem e- costs.pdf [Accessed 15th November 2022]. by
Trewin, S., Basson, S., Muller, M., Branham, S., Treviranus, J., Gruen, D. et al. (2019)
Considerations for AI fairness National
for people with disabilities. AI Matters, 5(3), 40–63.
van Toorn, G. (2021) The new political economy of disability: transnational networks and individualised funding in the Health
age of neoliberalism. London: Routledge.
And
van Toorn, G. & Scully, J.L. (2023) Unveiling algorithmic power: exploring the impact of automated systems on Medical
disabled people’s engagement with social services. Disability & Society,
1–26. Available from:
hhttps://doi.org/10. Research
1080/ 09687 599.09682 023.02233684
van Toorn, G. & Soldatic, K. (2024) Disablism, racism and the spectre of eugenics in digital welfare. Journal of Council,
Sociology, 1–17. Available from:
hhttps://doi.org/10.1177/14407833241244828
van Toorn, G., Soldatic, K. & Scully, J.L. (2022) NDIS plans rely on algorithms to judge need –
The upcoming review Wiley
should change that. The Conversation. Available from:hhttps:// theco nvers ation.com/
dis- plans - rely- on- algor Online
ithms - to- judge - need- the- upcom ing- revie w- shoul d- chang e- that-193106 [Accessed 27th October
2022]. Library
Venning, A., Hummell, E., Foster, M., Burns, K. & Harris Rimmer, S. (2021)
Adjudicating reasonable and necessary on
funded supports in the National Disability Insurance Scheme: A critical review of the values and priorities indicated
in the decisions of the Administrative Appeals Tribunal.
Australian Journal of Public Administration, 80(1), 97–113.
Wagner, B. (2019) Liable, but not in control? Ensuring meaningful human agency in automated decision- making systems. Policy & Internet, 11(1), 104–122. See the Terms and Conditions AUTHOR BIOGRAPHIES Georgia van Toorn is a lecturer in the School of Social Sciences at the University of New South Wales and an Associate Investigator at the ARC Centre of Excellence for Automated Decision-Making & Society (ADM+S). Specialising in international social policy, politics, disablement and social justice, she explores global transformations in welfare governance, with a particular focus on processes of marketisation, the commodification of social care and data analytics and algorithmic decision- making in the public sector. Terry Carney AO is Emeritus Professor at the Law School University of Sydney, a Fellow Online Library of the Australian Academy of Law and a past President (2005–2007) of the International rules Academy of Law and Mental Health.He is currently an Associate Investigator at the ARC for Centre of Excellence for Automated Decision-making and Society, writing on issues of au- tomation in social security law, welfare services and the NDIS. use;OA articles are governed by How to cite this article: van Toorn, G. & Carney, T. (2024) Decoding the algorithmic the applicable Creative Commons l operations of Australia’s National Disability Insurance Scheme. Australian Journal of Social Issues, 00, 1–19. Available from: hhttps://doi.org/10.1002/ajs4.342
National Disability Insurance Scheme
JUNE Amendment – 1 (Getting Back on Track No. 1) Bill 2024 [Provisions]
THE SATURDAY PAPER | THE MONTHLY | QUARTERLY ESSAY | AUSTRALIAN FOREIGN AFFAIRS | SEARCH | SUBSCRIBE | LOGIN
NEWS COMMENT CULTURE LIFE FOOD PUZZLES SPORT NEWSLETTERS PODCASTS EDITIONS JUNE 1 – 7, 2024 | № 502
News
Cost-saving reforms to the NDIS will allow debts to be raised against disabled people and give extraordinary new powers to cancel support entirely. By Rick Morton.
Exclusive: Shorten revives the Coalition’s failed NDIS reforms Minister for Government Services Bill Shorten. CREDIT: AAP IMAGE / MICK TSIKAS Bill Shorten’s reform of the National Disability Insurance Scheme, forecast to cut $14.4 billion in growth over the next five years, reanimates critical elements of the previous Coalition government’s faulty overhaul of the system - originally developed with assistance from one of the architects of robodebt. The reform revives an explicit threat to punish disabled people with debts even after they have accessed support a measure Shorten campaigned against in opposition It also gives the chief executive of the agency delivering the scheme extraordinary powers to cancel support entirely. At the end of March Shorten surprised disability advocates with new legislation despite promising they would be given an opportunity to co-design scheme reforms Nobody had seen the bill Many were told to sign non-disclosure agreements and even then were only briefed on the bill half an hour before the minister for government services and the NDIS rose in the House of Representatives to speak to it Members of the governments NDIS Participant Reference Group such as Cat Walker and Uli Cartwright who had also signed confidentiality agreements weren’t briefed until the following day Since then the bills implications have slowly become clear While the disability community has cautiously welcomed the opportunity for reform elements of the new legislation have contributed to a rising sense of concern Whether NDIS reforms succeed or fail this moment represents a key turning point in the life of the landmark scheme Shortsens ministerial prospects also depend on it as the May budget papers declared NDIS spending growth will shrink to an additional $1.5 billion over the next five years instead of almost $16 billion This reduction was predicated on the bill and subsequent changes to NDIS rules and other legislative instruments being passed by the parliament I can proudly say that this government and myself have worked every day with the disability sector to do everything we can to make life on the NDIS and life for people with disability in Australia betterShorten told the parliament on March 27 We promised to make the NDIS a priority and not penalise people with disability for wanting to live fulfilled lives And we promised to restore trust in the scheme The bill however has been introduced before the government has even responded to its own NDIS review led by the schemes grandfather Bruce Bonyhady and top public servant Lisa Paul Entire sections of the new legislation claim to give effect to recommendations from the review but are in fact at odds with what was recommended Some are years old dating back to a wish list of amendments from the Coalition era After weeks of pressure the federal government has already had to redraft sections of the legislation which had been passed by the House of Representatives because a new definition of “NDIS supports” was so badly worded it would have pulled funding from eligible participants Shorten told The Saturday Paper that never in the modern history of Australia has a government listened to and worked more closely with people with disability There will be more amendments to come he said I’m only interested in the best outcomes for participants and the scheme There is zero ego here If further changes are needed well make them Legal rights groups disability advocates and administrative law specialists say there are other flaws with the bill and some are significant At its core the amendments to the National Disability Insurance Scheme Act would still redefine who is and isn’t eligible for support make it easier for the agency and government to declare what kinds of support will be funded and create extraordinary new powers for the agency chief executive to revoke funding plans All of this would be baked into the law while the detail of what will and won’t be funded or who will and won’t be granted access will be left to future governments to determine via the easier route of rule-making and legislative instruments that need only be waved through the Senate The promised ecosystem of foundational supports which has always been envisioned to sit outside the NDIS for those who are not eligible and that would be jointly funded by all states and territories and the Commonwealth has never eventuated Now there is a new agreement to deliver these sprawling arrangements meant to deliver a soft landing to those who need support but do not qualify for the disability scheme over the coming years In short the promise of the NDIS review will arrive at some later uncertain date but the parts ceding more control to the agency and government will be passed within weeks if the bill is supported following its consideration by the Community Affairs Legislation Committee at the end of this month National Legal Aid told the inquiry elements of Shortsens reform go further than the legislation he helped kill in opposition the so-called independent assessments that were backed by then minister Stuart Robert Shorten is introducing a new needs assessment that says Legal Aid appears to be the only determinant of the supports a person will receive under the scheme and there is little scope for review of the needs assessment report This needs assessment has not been further fleshed out in the legislation There is no explanation for how it will be conducted whether the participant will even be involved or whether they can see it The bill allows the agency to request a new assessment if it is not satisfied although the participant will not have this option The respected Villamanta Disability Rights Legal Service has issued a blistering rebuke of the legislations drafters arguing this administrative efficiency is the “NDIS equivalent of Robo-debt” Collectively this raises the very real possibility that the intended assessment is simply using the category of disability recorded in the Agencies database (assigned by the Agency at access but never disclosed to the participant) entering it into a formula and generating the funding budget the groups submission to the inquiry says These assessments would also enshrine a longstanding but murky agency practice of picking a primary disability category under which it will fund support for people in the NDIS while refusing to deal with secondary conditions that may also be disabling A separate parliamentary inquiry found in November last year the arbitrary distinction likely had no basis in governing legislation or the reality of participants lives If passed in its current form the Bill will create a horrific maze of red tape extreme administrative burden and excessive checks and balances for people with disability with no rights to targeted review of decisions While Shorten told the parliament your needs assessment will look at your support needs as a whole - and we won’t distinguish between primary and secondary disabilities any longer the bill contains provisions that repeatedly and explicitly state only impairments for which a person meets the eligibility requirements should be assessed Taken together the changes proposed in the new legislation create a stunning new matrix of uncertainty that disabled people or their carers will be expected to navigate They will do so under the specific threat of being punished The proposed amendments including the introduction of flexible budgets make it significantly more likely that debts will be raised against participants including for actions taken by third parties National Legal Aid says In particular section 46(1A) has the effect of deeming a support payment made to a third party as an ‘NDIS amount’ that can be pursued from the participant under section 182 Even if a NDIS participant has received support funding and paid for a service the agency can decide it was not an appropriate support and collect the debt from the participant even though they no longer have the money In a statement a DSS spokesperson said this section was intended to support and safeguard participants It would only be in extreme circumstances such as where a person has deliberately misused NDIS funding that consideration of raising a debt under section 182 may occur which is the case now the statement said In 2021 when Stuart Robert was attempting to ram through similarly wide-ranging legislative changes to the NDIS The Saturday Paper obtained a leaked draft of the proposed changes alongside the comments and sections worked on by National Disability Insurance Agency employees One of those was a former compliance director from the then department of human services who helped conceive and implement the vast and illegal robodebt scheme in 2014 and 2015 The subsequent royal commission found he was aware that the proposal was inconsistent with social security legislation and policy but was not a lawyer and therefore did not appreciate the legal problems Like his manager at the time the compliance director went on to work in the senior management of the National Disability Insurance Agency where the job of reining in scheme costs has been a multi-year project with little change in emphasis His comments on the 2021 internal redraft of the NDIS legislation indicated he and the compliance division were particularly enthusiastic about creating a link between s.46 of the act and the existing debt provision in s.182 In part due to an opposition campaign waged by Bill Shorten that suite of legislative amendments was abandoned Now with Shorten as minister the explicit link between the two sections has been resurrected In the margins of the expanded s.46 as they intended it in January 2021 the former compliance director wrote Funding not to be used for certain purposes - the clear reference an amount spent in contravention of this subsection is a debt is a positive The drafting of the rules of what is an ‘ordinary living expense’ or ‘goods or services specified in the rules’ should provide us with a basis for making determinations of non-compliance and remove some of the questions of what is allowable / not allowable with the removal of the term ‘reasonable and necessary,’ he wrote Shortsens current reform differs in largely preserving the language of reasonable and necessary In 2021 the former compliance director also commented approvingly on a suggested change to s.182 that it provides clarity in terms of who the debt is against (the recipient who receives the money and does not spend in accordance with the Act) Following his notes he was appointed as the senior officer responsible for integrity of scheme outlays at the NDIA in January 2022 RELATED READING During the 2021 review Stuart Robert and the agency also tried to introduce a new power that would give the chief executive of the NDIA the ability to request any information they wanted from a participant under the threat of stopping their support funding despite the fact all participants are assessed and provide reams of material as part of that process in the first place My reading of this is that it provides a clear legislative basis for stopping future payments until we have received previously requested information/documentation, the former compliance director noted This power too is back under Shorten These provisions will allow the CEO to request information from a participant or other person if the CEO is considering revoking a persons status as a participant in the NDIS the new bills explanatory memorandum says Currently there is no ability for the CEO to request information for the purposes of considering the revocation of a persons status as a participant It is important for the CEO to be able to request and receive information from participants There are legitimate uses for all of these new powers including assessing whether people granted access under early intervention requirements will continue to need some form of support but advocates and legal aid representatives fear they could be used to constrain scheme costs arbitrarily The ability to kick people off is clearly what they want and clearly its people with psychosocial disability a legal source involved with the scheme tells The Saturday Paper And the problem is their systems aren’t good enough for them to possibly have any handle on who are the people that they’re talking about what are their situations and what needs to happen So it will be completely random as to who gets kicked off and in what circumstances and that is extremely dangerous Disability advocates Uli Cartwright and Cat Walker told the parliamentary inquiry into the bill the proposed legislation locks in procedurally unfair processes which echo robo-debt governance Making concerning elements lawful won’t make them ethical, they wrote If passed in its current form the Bill will create a horrific maze of red tape extreme administrative burden and excessive checks and balances for people with disability with no rights to targeted review of decisions Crucially they say the move to flexible budgets sounds good on paper but introduces a new regime at the same time as the golden thread linking life goals with funding outcomes is cut The weakening of the criteria connecting our goals and aspirations to our funding and spending will make it impossibly hard to defend supports solutions which might be unique to us Cartwright and Walker wrote If recognition of the collective and compounding impacts of multiple impairments is not explicitly formalised participants will be unable to identify which supports meet their needs without taking a scalpel to their lived experience and too frightened of debts being raised if they are unable to do so More than a year ago the federal government commissioned Labor-aligned research and political strategy outfit RedBridge Group to conduct a series of focus groups and narrative tests about potential and actual NDIS reform Ever since Shorten has been adopting the language of its findings There was also significant respondent concern regarding the threat that spiralling costs might pose to the Scheme - giving its critics a weak spot to attack RedBridge says in its October 2023 report Indeed when we presented rorts fraud and unreasonable pricing as posing an existential threat to the NDIS we were able to create an environment in which respondents were amenable to reforms designed to counter these things A separate element of the NDIS reforms reducing supported independent living (SIL) funding so disabled people with significant, 24/7 support needs will be forced to live in groups of at least three countering a recommendation from the disability royal commission to phase out group homes was also stress tested A significant cohort of dodgy providers have been abusing disabled people and their SIL funding provided by the disability scheme but advocates argue this is a reason to increase funding for some people so they can live how they want RedBridge found the government could cut services if it had a believable story to tell about protecting participants After presenting respondents with the above proof points that provide vital believable context of reforms designed to benefit participants we were able to elicit a degree of qualified tolerance for reforms that would restrict either eligibility for the Scheme raising the threshold for children with developmental delay or the amount of support high-need participants would receive moving to a 1:3 care model for Supported Independent Living participants the research says A concurrent push coming from the NDIS review - to eliminate the use of unregistered providers has split the disability sector Some participants are warning they will lose fundamental choice and control if they are forced to hire approved workers from approved providers and that in the past these workers have offered impersonal or substandard care At a press conference in Brisbane on Friday May 24 Shorten continued to follow the guidebook from RedBridge Fraud and rorting spiralling costs - these were the lens through which reform was sold We want to make the scheme fairer more transparent and more accountable It is growing too fast We believe that if we can moderate the growth of the scheme it will be there for future generations, Shorten told reporters We’ve now got some initial legislation to start the journey of improving the outcomes in the scheme for people with disabilities This is after a year-plus review and implementing the changes that the review has proposed We are not arrogant We will take on board the submissions Standing next to him was Mike Phelan his handpicked acting NDIS Quality and Safeguards Commissioner with a background in policing and criminal intelligence Phelan doubled down on a claim he first made as the head of the Australian Criminal Intelligence Commission that fraud was running at as much as 20 per cent in the NDIS Now that I’ve been involved in working with the NDIS its arguable I was a little bit conservative he said particularly given some of the things that are happening It is not that participants don’t want to fight rorting in the scheme Their concern is that the central agency delivering the disability scheme has weaponised a culture of hostile complexity that it now looks to have codified in law In our experience NDIA decision-making relies heavily on Operational Guidelines even when the [Administrative Appeals Tribunal] has found that the guidelines are inconsistent with the NDIS Act or Rules National Legal Aid says Indeed there are Federal Court precedents dating back to 2017 that the NDIA has simply ignored Now the new legislation seeks to embed this way of doing business Shorten attempted to strike a supportive tone when he introduced the new legislation to the parliament in late March I know that much-needed - and indeed much-wanted change can produce anxiety, he said Talk of any change to a family battling to make ends meet can sound like a problem not an opportunity and I can respect nervousness which might be caused by this discussion I just want to reassure these people who’ve battled hard to create an NDIS and to get their packages of support we will work with you to make sure that people are getting the right support in the right way For some however the echoes of a previous battle against the Coalition are too hard to ignore “Shorten was involved in fighting that,” one disability legal source says. And how lazy is that, to just bring it back when you need it for budget purposes? Minister Shorten said many voices in the sector he had spoken with or heard from have been very clear they do not characterise this as reheated 2021 laws This is the first of two tranches of legislation the first being able to close a significant loophole that shouldn’t be ignored any longer Secondly it sets up the legal authority and rules to begin the co-design process The Senate Community Affairs Legislation Committee is due to report on June 20 This article was first published in the print edition of The Saturday Paper on June 1, 2024 as Exclusive: Shorten revives the Coalitions failed NDIS reforms. Thanks Thanks Thanks Thanks forforforfor readingreadingreadingreading thisthisthisthis freefreefreefree article.article.article.article. For almost a decade The Saturday Paper has published Australia’s leading writers and thinkers We have pursued stories that are ignored elsewhere covering them with sensitivity and depth We have done this on refugee policy on government integrity on robodebt on aged care on climate change on the pandemic All our journalism is fiercely independent It relies on the support of readers By subscribing to The Saturday Paper you are ensuring that we can continue to produce essential issue-defining coverage to dig out stories that take time to doggedly hold to account politicians and the political class There are very few titles that have the freedom and the space to produce journalism like this In a country with a concentration of media ownership unlike anything else in the world it is vitally important Your subscription helps make it possible SelectSelectSelectSelect youryouryouryour digitaldigitaldigitaldigital subscriptionsubscriptionsubscriptionssubscription 3 months 6 months 12 months $55 $79 $109 See all subscription options Most read across The Saturday Paper NEWS News Corp’s deal with OpenAI a blow to journalism by Tim Dunlop Exclusive: Shorten revives the Coalition’s failed NDIS reforms by Rick Morton NEWS Fossil fuels are unreliable backstops in the energy transition by Bianca Nogrady