By Professor Albert Sanchez-Graells, Professor of Economic Law and Co-Director of the Centre for Global Law and Innovation (University of Bristol Law School)
Preventing, detecting, and sanctioning corruption in public procurement is one of the main goals of all systems of regulation applicable to the expenditure of public funds via contract (see eg Williams-Elegbe, 2012). Despite constant and regularly renewed efforts to fight procurement corruption at an international (such as the UN Convention against Corruption, or the 2016 OECD’s Preventing Corruption in Public Procurement report) and domestic level (see eg the UK’s 2020 ‘Local government procurement: fraud and corruption risk review’), corruption remains a pervasive problem in any given jurisdiction. Of course, there are different forms and degrees of corruption infiltration in different procurement systems but – if any evidence was needed that no system is corruption-free – pandemic-related procurement served as a clear reminder that this is the case (see eg Transparency International, 2021; as well as Good Law Project v Cabinet Office  EWHC 1569 (TCC)). It should then not be surprising that the possibility that artificial intelligence (AI) could ‘change the rules of the game’ (eg Santiso, 2019) and bring procurement corruption to an end is receiving significant attention. In a recent paper*, I critically assess the contribution that AI can make to anti-corruption efforts in the public procurement context and find that, while it could make a positive incremental contribution, it will not transform this area of regulation and, in any case, AI’s potential is significantly constrained by existing data architectures and due process requirements.
In the paper, I argue that the expectations around the deployment of AI as an anti-corruption tool in procurement need to be tamed. Even if, under the right conditions, AI can be faster, more consistent and more accurate than human decisionmakers, most narrow AI applications (eg robotic process automation of anti-corruption checks, or machine learning aimed at predicting corruption risk) cannot perform cognitive functions that humans would not also be able to carry out given sufficient time. As such, AI can contribute incrementally to current anti-corruption efforts in procurement, but it cannot significantly alter (or substitute for) existing oversight and enforcement architectures. To put it simply, AI can deliver more screening for potential corruption and discharge human officials from that administrative burden so they can reorient their efforts to more value-added activities, but that screening cannot be based on rules or information that would not be available to a human anti-corruption official. Consequently, AI cannot generate a revolution in the way corruption is prevented, identified, and sanctioned. It can only generate an increase in the volume of anti-corruption checks that are carried out, as well as speed them up, which can also allow for earlier interventions. Of course, these are desirable improvements, but they should induce lower levels of expectation than the hopes for an AI-based transformation of anti-corruption mechanisms.
Moreover, the extent to which AI can deliver such improvements is highly dependent on existing limitations in data availability and quality. Advanced forms of AI (eg unsupervised machine learning, including natural language processing techniques) currently cannot be adequately developed (or, rather, trained) on the basis of existing procurement and other relevant data and, even if enhanced data architectures were created, there are significant questions around the possibility of deploying them in unbiased ways that do not perpetuate current social, economic and political structures that could entrench, or even worsen, anti-corruption efforts. It should be stressed that building AI solutions based on past data carries the inherent risk of extending the shortcomings of existing oversight and enforcement architectures into the future and, what is worse, to insulate them from appropriate scrutiny due to the aura of objectivity and infallibility that can be ascribed to AI. It should also be stressed that new AI solutions in turn create new corruption risks (eg data poisoning, or adversarial attacks) that may be difficult to identify and remedy, which need to be dealt with at design stage, and also be balanced against their potential contribution to anti-corruption efforts at the point of making a decision whether or not to implement them.
Finally, even if the general obstacles concerning procurement data could be overcome, given the need to embed AI anti-corruption approaches into existing legal frameworks, it should be stressed that there are fundamental due process-based constraints that will continue to limit the potential use cases of AI. Some of the main constraints result from the duty to provide reasons in administrative law contexts, as well as the more stringent requirements for the imposition of (criminal) sanctions resulting from corruption in procurement. In the absence of as yet unlikely developments in the explainability of AI, legal requirements will continue to demand the presence of a ‘human-in-the-loop’ or, at the very least, an actively involved ‘human-over-the-loop’ in all AI anti-corruption procurement solutions. This requires close consideration of the AI-human interaction and the ways in which it will be necessary to create additional ‘dead-driver’ vigilance devices to prevent the mindless rubberstamping of AI-generated proposed decisions, as well as the creation of undesirable feedback loops and perverse dynamics in the adjustment of the algorithms supporting human decision-making. And, of course, for as long as there are human officials making AI-supported decisions or with the power to override AI-proposed decisions, the traditional risks of corruption will continue to demand oversight and enforcement, and these will have to be adapted to new corruption risks derived from the AI implementations.
All of this leads to two main conclusions for policymakers considering the deployment of AI-based anti-corruption solutions in a procurement context: first, prioritising improvements in procurement data capture, curation and interconnection is a necessary but insufficient step. Taking corruption in procurement seriously requires a broader and more comprehensive approach to the digitalisation of other databases, both public and private. Second, existing anti-corruption oversight and enforcement architectures need to be maintained or even expanded, and there is a need to ensure that the training and upskilling required to make use of AI solutions does not come at the cost of core capabilities. Moreover, the deployment of AI solutions for anti-corruption purposes in the procurement context (and more generally) creates its own governance and corruption risks, and these too require the development of additional layers of monitoring and oversight that will need resourcing. Therefore, investment in anti-corruption AI cannot be seen as a substitute for traditional investment in these efforts because existing and foreseeable AI solutions can act as a complement, but not a substitute of current approaches to the prevention, detection, and sanction of corruption in public procurement – and they come with their own governance challenges.
* You can read a more detailed analysis in the full paper: A Sanchez-Graells, ‘Procurement corruption and artificial intelligence: between the potential of enabling data architectures and the constraints of due process requirements’, to be published in S. Williams-Elegbe & J. Tillipman (eds), Routledge Handbook of Public Procurement Corruption (Routledge, forthcoming). Available at SSRN: https://ssrn.com/abstract=3952665.