Recording of webinar on 'Digitalization and AI decision-making in administrative law proceedings'

The Centre for Global Law and Innovation of the University of Bristol Law School and the Faculty of Law at Universidade Católica Portuguesa co-organised an online workshop to discuss emerging issues in digitalization and AI decision-making in administrative law proceedings. I had the great pleasure of chairing it and I think quite a few important issues for further discussion and research were identified. The speakers kindly agreed to share a recording of the session (available here), of which details follow:

Digitalization and AI decision-making in administrative law proceedings

This is a hot area of legal and policy development that has seen an acceleration in the context of the covid-19 pandemic. Emerging research finds points of friction in the simple transposition of administrative law and existing procedures to the AI context, as well as challenges and shortcomings in the judicial review of decisions supported (or delegated) to an AI.

While more and more attention is paid to the use of AI by the public sector, key regulatory proposals such as the European Commission’s Proposal for an Artificial Intelligence Act would largely leave this area to (self)regulation via codes of practice, with the exception of public assistance benefits and services. Self-regulation is also largely the approach taken by the UK in its Guide to using artificial intelligence in the public sector, and the UK courts seem reluctant to engage with the technology underpinning automated decision-making. It is thus arguable that a regulatory gap is increasingly visible and that new solutions and regulatory approaches are required.

The panellists in this workshop covered a range of topics concerning transparency, data protection, automation of decision-making, and judicial review. The panel included (in order of participation):

• Dr Marta Vaz Canavarro Portocarrero de Carvalho, Assistant Professor at the Faculty of Law of Universidade Católica Portuguesa, specialising in administrative law, and member of the Centro de Arbitragem Administrativa (Portuguese Administrative Law Arbitration Centre).

• Dr Filipa Calvão, President of the Comissão Nacional de Proteção de Dados (Portuguese Data Protection Authority) since 2012, and Associate Professor at the Faculty of Law of Universidade Católica Portuguesa.

• Dr Pedro Cerqueira Gomes, Assistant Professor at Universidade Católica Portuguesa and Lawyer at Cerqueira Gomes & Associados, RL, specialising in administrative law and public procurement, and author of EU Public Procurement and Innovation - the innovation partnership procedure and harmonization challenges (Edward Elgar 2021).

• Mr Kit Fotheringham, Teaching Associate and postgraduate research student at the University of Bristol Law School. His doctoral thesis is on administrative law, specifically relating to the use of algorithms, machine learning and other artificial intelligence technologies by public bodies in automated decision-making procedures.

Is the ESPD the enemy of procurement automation in the EU (quick thoughts)

I have started to watch the three-session series on Intelligent Automation in US Federal Procurement hosted by the GW Law Government Procurement Law Program over the last few weeks (worth watching!), as part of my research for a paper on AI and corruption in procurement. The first session in the series focuses in large part on the intelligent automation of information gathering for the purposes of what in the EU context are the processes of exclusion and qualitative selection of economic providers. And this got me thinking about how it would (or not) be possible to replicate some of the projects in an EU jurisdiction (or even at EU-wide level).

And, once again, the issue of the lack of data on which to train algorithms, as well as the lack of representative/comprehensive databases from which to automatically extract information came up. But somehow it seems like the ESPD and the underlying regulatory approach may be making things more difficult.

In the EU, automating mandatory exclusion (not necessarily to have AI adopt decisions, but to have it prepare reports capable of supporting independent decision-making by contracting authorities) would primarily be a matter of checking against databases of prior criminal convictions, which is not only difficult to do due to the absence of structured databases themselves, but also due to the diversity of legal regimes and the languages involved, as well as the pervasive problem of beneficial ownership and (dis)continuity in corporate personality.

Similarly, for discretionary exclusion, automation would primarily be based on retrieving information concerning grounds not easily or routinely captured in existing databases (eg conflicts of interest), as well as limited by increasingly constraining CJEU case law demanding case-by-case assessments by the contracting authority in ways that diminish the advantages of automating eg red flags based on decisions taken by a different contracting authority (or centralised authority).

Finally, automating qualitative selection would be almost impossible, as it is currently mostly based on the self-certification implicit in the ESPD. Here, the 2014 Public Procurement Directives tried to achieve administrative simplification not through the once only principle (which would be useful in creating databases supporting automatisation of some parts of the project, but on which a 2017 project does not seem to have provided many advances), but rather through the ‘tell us only if successful’ (or suspected) principle. This naturally diminishes the amount of information the public buyer (and the broader public sector) holds, with repeat tenderers being completely invisible for the purposes of automation so long as they are not awarded contracts.

All of this leads me to think that there is a big blind spot in the current EU approach to open procurement data as the solution/enabler of automatisation in the context of EU public procurement practice. In fact, most of the crucial (back office) functions — and especially those relating to probity and quality screenings relating to tenderers — will not be susceptible of automation until (or rather unless) different databases are created and advanced mechanisms of interconnection of national databases are created at EU level. And creating those databases will be difficult (or simply not happen in practice) for as long as the ESPD is in place, unless a parallel system of registration (based on the once only principle) is developed for the purposes of registering onto and using eProcurement platforms (which seems to also raise some issues).

So, all in all, it would seem that more than ever we need to concentrate on the baby step of creating a suitable data architecture if we want to reap the benefits of AI (and robotic process automation in particular) any time soon. As other jurisdictions are starting to move (or crawl, to keep with the metaphor), we should not be wasting our time.