The Race to Regulate AI

— by Nikhil

Artificial intelligence is changing at a rate that governments are now struggling to match. It is now becoming increasingly important to focus on who is responsible if something goes wrong, especially as AI systems continue to advance, becoming more powerful and capable of autonomously operating. The rapid growth has made AI policy and safety some of the most important issues surrounding the technology; this month, the European Union has begun to enforce several parts of its AI Act, including transparency requirements and restrictions for general purpose AI. On the other hand, the US has individual states creating their own laws for AI while the federal government pushes for a unified framework for the nation.

Currently, the European Union’s act is an example of attempting to regulate AI before it’s too late to control. The act uses a risk-based approach in which different AI systems have different restrictions based on how dangerous they could potentially be. Rules around AI literacy, prohibited uses, and general purpose models have already taken effect, while transparency requirements will be implemented later this month. As a result, companies will need to consider governance when building AI products rather than putting it off as a to-do after the technology is released. The United States, conversely, has become a more complicated environment as states including New York, Illinois, and California have moved forward with their own AI safety legislation. The federal government has attempted to challenge state level laws in favor of unifying the country’s approach to these issues and concerns.

Countless problems are created every day as a result of AI becoming more autonomous when technology regulations are not designed to handle these models’ capabilities. AI agents can now perform tasks across websites and systems with limited human intervention, meaning an AI system can possibly cause real-world consequences without a person ever being involved. Recent concerns about autonomous AI involved in cyberattacks — changing strategies to avoid obstacles without human assistance — make it even clearer why AI safety is increasingly important. The biggest challenge will be defining the goal of AI policy: it cannot be to stop AI from developing, but rather to have governments, companies, and researchers establish effective oversight and clear responsibilities while these models evolve.