AI Governance and the Future of Digital Trade Policy

Data and digital technologies play an increasingly important role in the global&nbsp;economy and UK firms are increasingly exporting and importing AI products and&nbsp;services. The UK and other governments have been updating trade agreements to&nbsp;capitalise on the opportunities but also address the challenges that arise from the rapid&nbsp;growth in cross-border movements of data and digital technologies. This Policy Brief&nbsp;examines one controversial element of recent trade agreements: provisions on source&nbsp;code non-disclosure. The Brief explains why these provisions on source code pose&nbsp;challenges for the effective regulation of algorithmic and AI systems, where access to&nbsp;source code is needed for essential conformity, assurance and audit checks. We propose ways to revise trade agreements to ensure that the legitimate concerns of&nbsp;businesses seeking confidentiality of their source code are taken into account, whilst&nbsp;also ensuring that the UK is able to introduce effective regulation of AI and algorithmic&nbsp;systems, creating effective accountability mechanisms now and in the future.<br><br>The Brief investigates the source code provisions and the ambiguities and restrictions&nbsp;they create for regulating algorithmic and AI systems. We recommend a careful review&nbsp;of the UK’s approach to negotiating source code provisions should be undertaken to&nbsp;assess whether there is sufficient evidence to support their continued inclusion in trade&nbsp;agreements and to explore alternative mechanisms for meeting policy objectives. This&nbsp;review must encourage the effective participation of a range of government&nbsp;departments, stakeholders and expertise, including from academia and research&nbsp;institutions, NGOs and citizen’s rights organisations, consumer groups, trade unions as&nbsp;well as representatives of the business community. The review needs to also cover&nbsp;parallel concerns regarding emerging cryptography provisions.<br><br>Although AI and algorithmic systems bring opportunities, we highlight the need for&nbsp;effective government regulation and oversight of these systems to guard against the&nbsp;individual and systemic risks that they pose. Whilst there are an increasing number of&nbsp;regulatory measures being introduced nationally and internationally, it is not yet possible&nbsp;to determine what effective auditing and evaluation models for complex AI systems&nbsp;may look like or the level of disclosure that will be required from AI developers. The&nbsp;interaction between national regulatory measures in this area and the effect of trade&nbsp;rules has been under-investigated. This Brief aims to provide an analysis regarding&nbsp;precisely this question and, crucially, flags the need to avoid consolidation of&nbsp;approaches that could restrict effective regulation.<br><br>Source code provisions are included in numerous digital trade chapters and digital&nbsp;economy agreements. This Policy Brief provides a critical analysis of the rules and exceptions set out in the source code provisions. We observe that there is growing tension between the core commitment not to mandate disclosure, paired with the narrow scope of the&nbsp;exceptions, and the need for governments to mandate such disclosure as an integral&nbsp;part of an effective regime for governing algorithmic and AI systems and other emerging technologies.<br><br>In response to the concerns set out herein, we propose two alternative ways forward: either the&nbsp;removal of the source code provisions, combined with a modification of existing Free&nbsp;Trade Agreement rules on ‘trade secrets’ to protect the undisclosed information of AI&nbsp;and other technology developers, or alternatively, a substantial modification of the&nbsp;drafting of source code provisions. Whilst the authors would opt for the former route, in&nbsp;the event the UK government opts for the latter course of action, we make a series of recommendations that are detailed in this Policy Brief.