Bengio and Privitera in TIME: A Beneficial AI Roadmap

Philip Fox, 20 October 2023

In a joint op-ed for TIME Magazine, Turing Award winner Prof. Yoshua Bengio and KIRA director Daniel Privitera reject a widespread assumption: that when it comes to developing advanced AI, there is an irresolvable conflict between progress, safety and democratic participation.

With the right policy goals, we can realize all three values. Bengio and Privitera suggest a four-step “Beneficial AI Roadmap” for this:

1. Invest in innovative use cases of existing AI
   Governments should reduce regulatory burdens for SMEs and invest in AI systems that promote innovation, but also values like inclusion and participation.

2. Boost research on trustworthy AI
   The AI research community must clear the imbalance between high budgets for making models more capable and low budgets for AI safety research.

3. Democratize AI oversight
   The decision whether, when, how and to what purpose transformative AI is developed must not fall on just a small number of private tech companies.

4. Establish procedures for monitoring and evaluating AI progress
   Compute monitoring and external evaluations of very capable models increase transparency, accountability and control.

At the upcoming AI Safety Summit in the UK, governments should focus on pragmatic policies like these – and not be mislead by the false assumption that we must choose between progress, safety and participation.