The AI Act Should Be Technology-Neutral
The AI Act’s broad definition of AI penalizes technologies that do not pose novel risks. To resolve this, policymakers should revise the definition of AI to only apply to specific AI approaches that create significant challenges.
Introduction
Policymakers have long valued the principle of technology neutrality, which holds that laws and regulations should avoid privileging or penalizing one set of technologies over another. Technology neutrality does not necessarily demand that the exact same rules apply to different technologies. For example, if policymakers believe AI systems present novel risks not found in non-AI systems, they can and should address those risks.
This report shows that the AI Act is not, despite the intention of the European Commission, technology-neutral. Instead of addressing unique concerns about uninterpretable machine learning (ML) systems—a subset of AI systems—the Act would apply to a much broader set of AI systems that do not need regulatory intervention. The result is legislation that would create significant overreach and potential harm to the EU’s AI ecosystem. A better definition would limit the scope of the proposed law to only those technologies that pose novel risks.
AI development is not linear—it has gone and continues to go through various periods of flourishing (“springs”) and stagnations (“winters”). The last “AI winter” has passed, but the EU is falling behind its global competitors—China and the United States—in AI research, investment, and adoption. If the EU’s AI Act limits innovation, as the current version of the bill is likely to do, the next AI winter could be strictly European.
