Against AI Regulation: The Innovation Cost We Cannot Afford
The instinct to regulate AI is understandable. New technologies generate anxiety, and policymakers respond to anxiety with legislation. But good economic policy requires weighing costs against benefits, and the costs of premature, comprehensive AI regulation are substantial - while many of the claimed benefits are speculative.
We are at an early and critical stage of AI development. The systems that exist today are capable but narrow; the systems being built now will be transformative. Locking in regulatory frameworks based on current capabilities risks both stifling beneficial innovation and failing to anticipate the actual risks of future systems.
"Regulating AI today is like regulating the internet in 1993 - we barely understood what it was."
The Innovation Cost
Regulatory compliance has fixed costs. Large incumbents can absorb them; start-ups and academic researchers cannot. Comprehensive AI regulation therefore functions as a moat for the firms already dominant in the space - the exact opposite of the competitive outcome that regulation's advocates claim to want. The European Union's AI Act, well-intentioned as it is, has already driven several AI research groups to relocate to less regulated jurisdictions.
The Information Problem
Regulators face severe information asymmetries. The technical complexity of modern AI systems exceeds the capacity of most regulatory agencies to evaluate them. This creates a regulatory capture risk: rules will be written by those who understand the technology, namely the large firms whose interest lies in creating compliance barriers.
Existing Laws Are Sufficient
Most of the harms attributed to AI - discrimination, fraud, privacy violation - are already illegal. The problem is not absence of law but absence of enforcement. Dedicating regulatory energy to new AI-specific frameworks, rather than enforcing existing consumer protection, competition, and data protection law, is displacement activity.
Conclusion
The economic history of technology regulation is not encouraging. We have consistently overestimated the risks of new technologies and underestimated the costs of constraining them. AI regulation should be targeted, light-touch, and adaptive - not the comprehensive framework being proposed, which will damage the UK's position in the most consequential technological race of our time.
