The Economics of Artificial Intelligence
Regulating AI
How governments are approaching AI regulation, from the European Union's risk-based AI Act to lighter-touch approaches, and the economic trade-offs involved.
As AI spreads into hiring, lending, health care, education and public services, governments are deciding how to regulate it. Approaches differ widely.
The EU AI Act
The European Union adopted the AI Act in 2024, the first comprehensive AI law by a major jurisdiction. It takes a risk-based approach:
- Unacceptable risk: some uses are banned, such as social scoring by governments and certain manipulative techniques.
- High risk: AI used in areas like hiring, credit scoring, education, law enforcement and critical infrastructure must meet requirements for data quality, transparency, human oversight and accuracy.
- Limited risk: some systems, like chatbots, must tell users they are interacting with AI, and AI-generated content must be labelled in certain cases.
- Minimal risk: most AI uses, like spam filters, face few new rules.
It also sets obligations for providers of general-purpose AI models. Its rules phase in between 2025 and 2027.
Other approaches
- The United States has relied mainly on existing laws, sector regulators and executive actions, which have changed between administrations, with debate about federal versus state rules.
- China introduced rules on specific AI uses, such as recommendation algorithms and generative AI, requiring registration and content controls.
- The United Kingdom has favoured a principles-based approach through existing regulators.
- India has so far favoured a light-touch approach focused on promoting AI while addressing specific harms, including guidelines and advisories on AI and deepfakes.
The economic trade-offs
- Benefits of regulation: protecting people from harms such as discrimination, fraud and unsafe products; building trust that encourages adoption.
- Costs: compliance burdens, especially for smaller firms; the risk of slowing innovation or pushing it elsewhere.
- Uncertainty: rapidly changing technology makes it hard to write lasting rules.
A company uses AI to screen job applications. Under the EU AI Act, this counts as high risk, so the provider must document its training data, test for bias, allow human oversight and keep records. This adds cost, but reduces the chance that the tool unfairly rejects qualified candidates, such as disabled applicants whose CVs look different.
Even without AI-specific laws, existing rules on discrimination, consumer protection, privacy, product safety and copyright already apply to many AI uses. New laws add rules specific to AI's risks.
- The EU's 2024 AI Act takes a risk-based approach, banning some uses and regulating high-risk ones.
- The U.S., China, UK and India have taken different approaches.
- Regulation can build trust and prevent harm but adds compliance costs.
- Existing laws on discrimination, privacy and consumer protection already apply to AI.
No recording for this one yet - EconReader can read it aloud for you.