New IViR study says EU position on source code endangers AI regulation
8 February 2021
Dr Irion's study concludes that the EU position on source code in international trade agreements limits the EU's ability to regulate AI in the interests of consumers. This is particularly the case in the ongoing e-commerce negotiations at the WTO. These findings are relevant for future EU regulations that allow algorithmic controls requiring access to the source code or access to the interfaces of an AI system. On 26 January, Dr Irion presented the findings of her study to members of the European Parliament and the European Commission. Several media have covered the study, including Euractiv, Politico and Tagesspiegel.
Today, algorithmic decision-making is an integral part of many digital services, such as online shopping and social media applications. It is also used in financial services. To protect consumers from harmful artificial intelligence practices, such as (price) discrimination, inaccurate consumer information and algorithmic bias, the European Commission will propose rules on AI transparency early this year.
As AI applications are often delivered across borders from outside the EU, it is not only the EU regulations themselves that matter, but also the design of trade disciplines. Members of the World Trade Organisation are currently discussing plurilateral rules on e-commerce. These negotiations also include agreements on the non-disclosure of source code. Although these agreements are mainly mentioned as a tool against forced technology transfer, they should also ensure that national and European rules on AI transparency cannot be obstructed.
The Institute for Information Law was commissioned by the German consumer organisation Verbraucherzentrale Bundesverband to carry out a study on the cross-border use of AI technologies and its impact on consumer law in the EU. In the current negotiations, the EU supports the inclusion of a special clause. This clause prohibits participating countries from including in national legislation measures requiring access to or transfer of the source code of software, with some exceptions. This is a cause for concern because if such a clause is not carefully conditioned, it may prevent future EU legislation on harmful AI.
The study concludes that the source code clause in trade law indeed restricts the EU's right to regulate AI policy in several important ways. This conclusion is surprising given that EU trade policy documents do not refer to AI but only to e-commerce and no direct link has been made between the software source code clause and algorithms. The investigation raises an important EU policy issue that needs to be democratically examined and debated. This must be done before the EU agrees to a new software source code clause in a plurilateral WTO agreement on electronic commerce.
Digitalisation is leading to more and more digital artefacts consisting of software source code. AI technology can create new risks for individuals and society, while trade law remains largely static after ratification. The source code clause is too broad for domestic digital policies that must build on the ability of systems to exchange data (interoperability), accountability and verifiability of digital technologies.
The study makes two recommendations: