FUTUREID3 : Identification in the era of Automated Decision Making

Tuesday, 19 March, 2019 (All day)
Add to Calendar

Jesus and St Johns Colleges, Cambridge, March 18 - 21, 2019

Panels will take place on March 19 & 20 only.

Call for Participation

This workshop takes its focus from an important, new problem for economics, governance and justice in countries with histories of weak state administration. Does machine-based identification strengthen or weaken the dispersed and delegated processes of registration of the poor and the marginal?  This key problem unsurprisingly has led to questions in the realm of law and policy, in civil registration in general, and at the interface of machine learning, particularly with fast-developing private credit markets in poor countries.

In the realm of law/policy, there appears to be consensus at least in development policy circles on the benefits of stronger identification systems—shown notably by the adoption of SDG Target 16.9 that states should, by 2030, “provide legal identity for all, including birth registration”, and by the World Bank’s decision to establish a program on “identification for development”.

This policy agenda is founded on solid academic scholarship investigating the importance of registration and identification for state consolidation and effectiveness (for example, Caplan and Torpey 2001; Szreter 2007; Breckenridge and Szreter 2012).  State consolidation and effectiveness, however, also presents political dangers. Other scholars have emphasized the equal contribution that registers of people and property have made towards the enabling of authoritarian regimes (most influentially, Scott 1999), and the uses of identification for surveillance and control (Torpey 2000; Bennett and Lyon 2008).  As formerly administratively passive states adopt identification systems driven by standardised biometrics and machine learning this dichotomy between surveillance and recognition is becoming pronounced.

There is still considerable uncertainty about the impact of the SDG target. Some of this doubt stems from the ambiguity of the term “legal identity”.  While the concept of birth registration is well-understood, and has been the subject of international agreements and technical guidance over some decades, legal identity as a concept is less clear. “Legal identity” has no definition in international law and there is no clarity on what its delivery would require (or indeed if state involvement is necessary at all).  In seeking to tilt the balance towards benign outcomes, development policy interventions have tended to give their attention to the new challenges raised by digital identification, especially the need for regulation of privacy and data protection.

There has been less focus on the implications of digitalisation for longstanding issues around the governance of identification systems more generally. Perhaps most important of these are the basis on which eligibility for a particular identity is determined (including whether a person is a citizen or not) and the scope for independent (court and other) oversight of those decisions. If these rules are non-existent or very restrictive, the stronger identification systems fostered by new forms of automation are likely rapidly to reinforce exclusion rather than promote inclusion.

In the last two years it has become clear that the tensions between automated identification and administrative registration have to be confronted in two further ways.  The first concerns the relationship between civil registration in general and automated biometric registration.  Comparative historical research suggests that this is a matter of supporting distributed agency -- of providing easily accessible tools that ordinary individuals, and especially the poor, can use to engage with each other, with non-state institutions and with the state over resources and entitlements. There is evidence from the Indian, South African, and Kenyan experiences to suggest that biometric registration can do this work to strengthen the efforts of long neglected populations to make themselves known administratively. There is also evidence that these schemes can be expensive and wasteful, and politically destructive, weakening the entitlement claims of the poor and the marginal, their representative institutions and organisations and diverting precious resources from the basic, and slow, work of administrative registration. Large-scale, centralised biometric identification programmes that unlock all other rights (not just legal presence in the country, but also access to land, health care, education) also require us to pay careful attention to the relationship of such programmes to relevant rules of access and appeals: who decides, what matters can courts and other oversight institutions review, and, taking into account the institutional and political-economic contexts that shape these outcomes, how are issues in these programmes shaped (or not) into critical questions.

A second problem, which has become much more influential over the last two years, concerns the application of machine-learning decision making processes to the old puzzles of identification and risk assessment, and their effects on the administrative fabric of identification and registration.  Moving from the engineering realm to popular discourse, artificial Intelligence is a cliché but also more than that. As Nigeria’s Biometric Verification Number and IndiaStack both show, FATF guidelines for Know Your Customer requirements and, probably more importantly, lucrative credit risk assessments linked to robust forms of identification have motivated governments and banks to require biometric registration for account holders.  These banking programmes have supported very large scale, and comparatively successful, campaigns of registration. Recently they have begun to change significantly. The application of algorithmic and tiered credit-scoring risk assessments, like Safaricom’s popular M-Shwari credit system, can reduce the registration burden imposed on the poor and the administrative costs of providing small amounts credit. Whether these new, efficient and convenient tools for assessment and decision making will weaken or strengthen registration processes in general -- releasing pressure from above and from below on registration agencies, and creating a large and obvious pool of weakly identified people -- is an important new problem for research.

 

Cheap and effective machine learning tools for identification and risk assessment also introduce a number of new and compelling legal and institutional opportunities and problems.  Overlapping with the legal identity question identified above, many of these can be clustered around state power and the development of meaningful privacy regimes in societies without well developed legal protections.  As the Zimbabwean government’s recent sharing of its population registration database with the Chinese firm CloudWalk shows, machine learning may equip the global network of states with powerful instruments for control and subjugation -- even if they do not work well in practice.  Similar questions apply in the purely commercial domain to the relationship between privacy, machine-learning and monopoly firms -- both domestic and international -- in the context of weakly developed regulatory institutions and legal instruments. 

 

In this workshop we will draw on the research of an expert network of scholars and consultants with close familiarity with registration projects around the world to address these problems.  Most of the participants will fund their own travel, accommodation and registration but we are seeking support for specialist researchers and policy-makers who cannot fund their own travel to the workshop.  

 

Proposed panels:

 

1) Global comparisons of automated ID systems and devolved registration

Organiser:  Keith Breckenridge & Edgar Whitley

2) Comparative costs and benefits of identification projects

Organiser:  Keith Breckenridge

3) Competing models of privacy and data regulation : EU, USA, China

Organiser:  Jon Klaaren

4) Statelessness, border-marking and migration

Organiser:  Bronwen Manby

5) Registration and administration

Organiser:  Simon Szreter

6) Digital property and credit surveillance as alternative or platform for civil registration

Organiser:  Edgar Whitley

7) Practical inclusion  -- How to walk the last mile

Organiser: Mia Harbitz

 

Organisers:

  • Keith Breckenridge, WISER, Wits
  • Mia Harbitz, retired IADB, public registries expert
  • Jonathan Klaaren,  Law & WISER, Wits
  • Bronwen Manby, Legal Consultant Human Rights
  • Simon Szreter, History & Policy, Cambridge
  • Jaap van der Straaten,  CRC4D
  • Edgar Whitley, Information Systems, LSE