Bengio, alongside other AI leaders like Google DeepMind CEO Demis Hassabis, discussed the importance of mitigating risks associated with AI agents. They highlighted the potential for AI to contribute to scientific discovery without being agentic, citing examples like DeepMind's protein folding breakthrough. Both Bengio and Hassabis agree on the necessity of cybersecurity measures and controlled experimentation with AI agents. However, they acknowledge the economic pressures driving the development of agentic systems, as businesses recognize their potential for tangible returns on investment.
Key takeaways:
- Yoshua Bengio warns against AI agents, suggesting they could lead to catastrophic scenarios if they achieve AGI.
- Bengio advocates for building non-agentic systems to control agentic systems and mitigate risks.
- There is a push for national regulation to ensure AI companies prove the safety of agentic models before development.
- Bengio and Demis Hassabis agree on the need for measures to mitigate AI risks, but note the economic drive towards agentic systems.