Responsibly Buying Artificial Intelligence: A ‘Regulatory Hallucination’ -- draft paper for comment

© Matt Lowe/LinkedIn.

Following yesterday’s Current Legal Problems Lecture, I have uploaded the current full draft of the paper on SSRN. I would be very grateful for any comments in the next few weeks, as I plan to do a final revision and to submit it for peer-review in early 2024. Thanks in advance for those who take the time. As always, you can reach me at a.sanchez-graells@bristol.ac.uk.

The abstract of the paper is as follows:

Here, I focus on the UK’s approach to regulating public sector procurement and use of artificial intelligence (AI) in the context of the broader ‘pro-innovation’ approach to AI regulation. Borrowing from the description of AI ‘hallucinations’ as plausible but incorrect answers given with high confidence by AI systems, I argue that UK policymaking is trapped in a ‘regulatory hallucination.’ Despite having embraced the plausible ‘pro-innovation’ regulatory approach with high confidence, that is the incorrect answer to the challenge of regulating AI procurement and use by the public sector. I conceptualise the current strategy as one of ‘regulation by contract’ and identify two of its underpinning presumptions that make its deployment in the digital context particularly challenging. I show how neither the presumption of superiority of the public buyer over the public contractor, nor the related presumption that the public buyer is the rule-maker and the public contractor is the rule-taker, necessarily hold in this context. Public buyer superiority is undermined by the two-sided gatekeeping required to simultaneously discipline the behaviour of the public sector AI user and the tech provider. The public buyer’s rule-making role is also undermined by its reliance on industry-led standards, as well as by the tech provider’s upper hand in setting contractual benchmarks and controlling the ensuing self-assessments. In view of the ineffectiveness of regulating public sector AI use by contract, I then sketch an alternative strategy to boost the effectiveness of the goals of AI regulation and the protection of individual rights and collective interests through the creation of an independent authority.

Sanchez-Graells, Albert, ‘Responsibly Buying Artificial Intelligence: A “Regulatory Hallucination”’ (November 24, 2023). Current Legal Problems 2023-24, Available at SSRN: https://ssrn.com/abstract=4643273.

Responsibly Buying Artificial Intelligence: A Regulatory Hallucination?

I look forward to delivering the lecture ‘Responsibly Buying Artificial Intelligence: A Regulatory Hallucination?’ as part of the Current Legal Problems Lecture Series 2023-24 organised by UCL Laws. The lecture will be this Thursday 23 November 2023 at 6pm GMT and you can still register to participate (either online or in person). These are the slides I will be using, in case you want to take a sneak peek. I will post a draft version of the paper after the lecture. Comments welcome!