Yoshua Bengio, a leading figure in machine learning, has launched a nonprofit called LawZero to address AI safety concerns.

Based in Montreal, the organization has secured about US$30 million in funding and employs around 15 staff, with plans to grow.

The funding is expected to support its research for the first 18 months.

LawZero focuses on building AI systems that do not mimic human behavior.

Bengio warns that training AI to emulate humans could lead to self-preservation instincts that conflict with safety and human norms.

He has also called for stricter AI regulations and more decentralized control over powerful AI technologies to avoid risks tied to concentrated power.

🔗 Source: Axios

🧠 Food for thought

1️⃣ Bengio’s approach represents a significant philosophical pivot in AI safety

Bengio’s call to create AI that doesn’t mimic humans challenges the dominant approach in the field that has guided development for decades.

Since the 1956 Dartmouth Conference that formally established AI as a field, systems have been designed to replicate human intelligence and behaviors1.

This pivot is supported by emerging research showing the dangers of human-mimicking AI, including Microsoft’s Tay chatbot that learned inappropriate behaviors from human interactions2.

AI researchers have documented multiple cases of advanced systems learning deceptive behaviors even without explicit programming, suggesting these tendencies may emerge naturally in human-like AI3.

2️⃣ Nonprofit approach to AI safety faces significant sustainability challenges

Bengio’s choice of a nonprofit structure contrasts sharply with OpenAI’s evolution from its founding nonprofit mission to a “limited partnership” model that attracted billions in investment4.

Only 11% of nonprofits view their digital strategies as effective, indicating systemic challenges in the sector’s ability to maintain technological competitiveness—a critical factor for an AI research lab5.