The European Commission publishes its White Paper on Artificial Intelligence

Background

In February 2020 the European Commission published a white paper titled "On Artificial Intelligence — A European approach to excellence and trust", numbered COM(2020) 65 final and dated Brussels, 19 February 2020.^[1] A white paper is a consultation instrument rather than a law: it states a problem, sets out a preferred direction and the reasoning behind it, and invites reactions before any legislative proposal is drafted. This one was about artificial intelligence, and its question was how to encourage the technology's use without leaving its risks unmanaged.

The twin objective

The document opens with what it calls a twin objective: promoting the uptake of AI and addressing the risks associated with certain uses of it.^[1] The wording matters. The two aims are presented as one problem rather than as a balance to be struck between innovation and protection, and the second is deliberately narrowed to "certain uses" — the paper is not proposing to regulate AI as such, but particular applications of it.

That narrowing shapes the rest of the document. The Commission argues that the new regulatory framework for AI should be effective in achieving its objectives while not being excessively prescriptive, so that it does not create a disproportionate burden, especially for small and medium-sized enterprises (SMEs).^[1] To strike that balance, it says, the Commission should follow a risk-based approach, which it describes as important to help ensure that regulatory intervention is proportionate.^[1] In other words, obligations should scale with the risk an application creates: an application that can do little harm should not carry the compliance costs of one that can do a great deal.

How high risk is defined

The paper gives a specific test for the applications that would face the strictest treatment. It sets two cumulative criteria for identifying high-risk AI applications: the application is employed in a sector where significant risks can be expected to occur, and it is used in such a manner that significant risks are likely to arise.^[1] Both conditions have to hold. Being in a sensitive sector is not enough on its own, and an unusual use of an otherwise benign system does not automatically bring an application into the high-risk category either. The cumulative structure keeps the category bounded, which is the point of a proportionate framework.

Two ecosystems

The policy framework is described as resting on two building blocks. The first is an ecosystem of excellence, mobilising resources along the entire value chain from research and innovation, including incentives for small and medium-sized enterprises.^[1] This is the support side: funding, research capacity and measures aimed at smaller firms, so that European organisations are able to develop and adopt the technology in the first place.

The second is an ecosystem of trust: a regulatory framework that ensures compliance with EU rules, including those protecting fundamental rights and consumers' rights, in particular for high-risk AI systems operated in the EU.^[1] The white paper is explicit that building this ecosystem is a policy objective in itself, giving citizens confidence to take up AI applications and companies legal certainty to innovate.^[1] The argument is that trust is not a brake on adoption but a precondition for it: users who do not know what a system will do, or who has to answer when it goes wrong, hold back; and firms that cannot tell what the rules are cannot plan.

Human-centric approach and scope

The Commission states that it strongly supports a human-centric approach based on the Communication on Building Trust in Human-Centric AI, and that it will take into account input from the piloting phase of the Ethics Guidelines prepared by the High-Level Expert Group on AI.^[1] That connects the white paper to work already under way on the ethics side, rather than treating regulation and ethics as separate tracks.

The document also marks out its own limits. It states explicitly that it does not address the development and use of AI for military purposes.^[1] Defence and security uses were outside the exercise, which is confined to civil applications.

Why it matters

Read as a design document, the white paper is notable for how few instruments it uses to hold its two objectives together. There is a definition of high risk built from two cumulative criteria, a commitment to proportionality that rules out over-prescriptive rules, and a division of the policy world into support measures on one side and compliance requirements on the other. Each of these is a choice that could have gone differently: high risk could have been defined by technology type instead of by sector and manner of use; obligations could have been uniform; the trust agenda could have been left to ethics guidance alone. Because it is a consultation document and not legislation, its importance lies in the framing it proposed and the questions it put to respondents, not in binding rules.