Yoshua Bengio, one of the three renowned “AI godfathers” who pioneered deep learning and neural networks, delivered a stark warning at the World Economic Forum in Davos this week about the dangers of AI agents—artificial intelligence systems capable of acting independently without human input. Bengio’s message comes at a critical juncture as AI agents have become one of the hottest topics at this year’s gathering in Switzerland.
The Canadian research scientist, whose foundational work laid the groundwork for today’s AI boom alongside Geoffrey Hinton and Yann LeCun, argued that “all of the catastrophic scenarios with AGI or superintelligence happen if we have agents.” He emphasized that achieving artificial general intelligence (AGI)—machines that can reason as well as humans—is possible without building agentic systems. According to Bengio, applications in science and medicine that people value most are non-agentic, and more powerful non-agentic systems can continue to be developed.
The timing of Bengio’s warning is particularly significant as businesses increasingly recognize the tangible return on investment offered by AI agents, which could meaningfully enter the workforce as soon as 2025. OpenAI recently unveiled an AI agent capable of surfing the web and performing tasks like booking restaurants or adding groceries to shopping carts, while Google has previewed similar capabilities.
Bengio acknowledged the challenge ahead: companies and countries will likely continue building agents regardless of warnings, driven by competitive pressures and fear of falling behind. However, he proposed solutions, including using non-agentic systems as monitors to control agentic ones, though this would require substantial investment. He also called for national regulation requiring AI companies to prove system safety before building agentic models.
During a panel discussion with Google DeepMind CEO Demis Hassabis, Bengio raised a “red flag” about AI agents, calling them “the most dangerous path.” He pointed to DeepMind’s protein folding breakthrough as an example of profound AI applications that don’t require agency. While Hassabis agreed on risk mitigation measures like cybersecurity protections and simulation testing, he noted the economic incentives driving agentic development, asking why users wouldn’t want systems that can both recommend and book restaurants automatically.
Key Quotes
All of the catastrophic scenarios with AGI or superintelligence happen if we have agents.
Yoshua Bengio stated this during his interview at Davos, emphasizing his core concern that AI agents pose the greatest existential risk as the industry moves toward artificial general intelligence.
All of the AI for science and medicine, all the things people care about, is not agentic. And we can continue building more powerful systems that are non-agentic.
Bengio made this argument to demonstrate that valuable AI applications don’t require agency, suggesting an alternative path for AI development that avoids the risks associated with autonomous systems.
I want to raise a red flag. This is the most dangerous path.
Bengio delivered this warning during a panel discussion with Google DeepMind CEO Demis Hassabis, directly confronting the audience about the risks of pursuing agentic AI development.
Unfortunately I do think there’s an economic gradient, beyond science and workers, that people want for their systems to be agentic. When you say ‘recommend me a restaurant,’ why would you not want the next step, which is, book the table.
Demis Hassabis acknowledged the commercial pressures driving agentic AI development, highlighting the practical consumer demand that makes it difficult to halt progress despite safety concerns.
Our Take
Bengio’s warning carries particular weight given his status as an AI pioneer rather than an outside critic. His distinction between agentic and non-agentic AI offers a nuanced framework that the industry desperately needs as it grapples with safety concerns. However, Hassabis’s counterpoint about economic incentives reveals the fundamental challenge: consumer demand and competitive pressures create powerful momentum toward agentic systems. The proposal to use non-agentic AI as monitors is intriguing but raises questions about feasibility and whether such safeguards can keep pace with rapidly advancing agent capabilities. The real test will be whether governments can implement meaningful regulation before market forces make agentic AI ubiquitous. With OpenAI and Google already launching agent products, we may be past the point of prevention and into the realm of damage control.
Why This Matters
This warning from one of AI’s most respected pioneers represents a critical inflection point for the industry. As AI agents transition from theoretical concepts to practical business tools in 2025, Bengio’s concerns highlight the tension between commercial incentives and existential risks. His call for regulation and safety measures comes as major tech companies race to deploy agentic AI systems that could fundamentally transform how businesses operate and how humans interact with technology.
The debate at Davos underscores a broader challenge facing the AI industry: balancing innovation with responsibility. With businesses already seeing ROI from AI agents and companies like OpenAI and Google launching web-browsing agents, the momentum toward agentic AI appears unstoppable. Bengio’s proposal to use non-agentic AI systems as safeguards offers a potential middle path, but requires coordinated investment and regulatory frameworks that don’t yet exist. The disagreement between Bengio and Hassabis also reveals divisions within AI leadership about whether agentic systems can be safely developed or should be avoided entirely—a debate that will shape AI development for years to come.
Recommended Reading
For those interested in learning more about artificial intelligence, machine learning, and effective AI communication, here are some excellent resources:
Recommended Reading
Related Stories
- Artificial General Intelligence Could Arrive by 2024, According to AI Experts
- OpenAI CEO Sam Altman’s Predictions on How AI Could Change the World by 2025
- Outlook Uncertain as US Government Pivots to Full AI Regulations
- The AI Hype Cycle: Reality Check and Future Expectations
- The Artificial Intelligence Race: Rivalry Bathing the World in Data
Source: https://www.businessinsider.com/yoshua-bengio-ai-godfather-agents-2025-1