Regulating AI: It’s Not a Question of If, But How

Recently, I was invited by the GIZ (Gesellschaft für Internationale Zusammenarbeit) and the German Federal Ministry for Digital and Transport (BMDV) to share my thoughts with a delegation from Brazil’s Ministry of Science, Technology, and Innovation (MCTI). These were my key points:

1. Transparency and Control: Emphasizing transparency and control is essential in both the development and deployment of AI systems.
2. Regulate the Use, Not the Technology: Demanding explainability from a large language model (LLM) is technologically challenging, if not impossible. What we should focus on regulating is the way AI is used.
3. Leverage Existing Frameworks: We already have regulatory frameworks in place to protect data and ensure fairness. These can be applied to AI without needing to reinvent the wheel.

AI regulation doesn’t happen in isolation—it requires balancing the interests of citizens, businesses, and governments. Every regulation has its trade-offs. Take, for example, the EU’s car insurance regulation, which mandates equal premiums for men and women, even though men statistically have more accidents. From a data science perspective, this doesn’t make sense, but from a societal point of view, it reflects our desire to avoid gender discrimination. However, this decision has consequences: it disrupts risk management and leads to higher overall costs. Men’s premiums fell by 10%, while women’s premiums rose by 30%.

Such trade-offs may be worth it, reflecting society’s values, but regulation is never a zero-cost game. The current regulatory environment in Europe has pushed many AI innovations to be used elsewhere. Perhaps that’s intentional? Or perhaps it opens doors for countries like Brazil to capitalize and catch up in AI innovation.