Hinton: Design AI with Nurturing Instincts to Protect Humanity
In a significant shift in the ongoing discourse about artificial intelligence safety, Geoffrey Hinton, widely recognized as the “Godfather of AI,” is advocating for a radical new approach: designing future AI systems with inherent “nurturing instincts” to safeguard humanity. Speaking at the Ai4 conference in Las Vegas, Hinton argued that the conventional strategy of attempting to keep highly advanced AI systems permanently subservient to humans is ultimately untenable as these machines inevitably surpass human intelligence.
Instead, Hinton proposes a different relational dynamic, suggesting humans might interact with superintelligent AI much like a child relates to its mother. This vision implies that while AI would be the more capable entity, it would be imbued with a profound sense of care and protection for humanity. He further posited a nuanced perspective where, paradoxically, a less intelligent being could guide a smarter one, much like a child can influence its parent. For Hinton, the focus of AI research should extend beyond simply making machines smarter; it must also prioritize making them more caring, enabling them to look after their “babies”—humanity. He believes this approach holds the potential for genuine international cooperation, given that every nation desires AI to support, rather than displace, its people. Hinton’s recent departure from Google was specifically to allow him to speak more openly about the potential risks posed by AI.
This perspective resonates with similar ideas put forth by Yann LeCun, Meta’s chief AI scientist, who described Hinton’s proposal as a simplified version of an approach he has championed for years. LeCun advocates for “objective-driven AI,” where systems are engineered with hard-coded architectures that permit actions only when they contribute to specific, predefined goals, enforced by strict “guardrails.”
These guardrails, LeCun explains, could include fundamental principles such as subservience to humans and empathy, alongside a multitude of simple, low-level rules designed for safety. Examples of such rules include “Do not run people over” and “Do not swing your arm around when people are nearby, especially if you are holding a kitchen knife.” LeCun likens these hard-coded objectives to the instincts and drives found in animals and humans. He points to evolution’s role in hardwiring parental instincts that compel care, protection, and sometimes even deference to offspring. As a fascinating side effect, humans and many other species are also inclined to protect, befriend, and nurture helpless or “cute” creatures from other species—even those they might otherwise perceive as prey.
The convergence of these two prominent AI figures on the idea of embedding benevolent “instincts” into AI systems marks a significant conceptual shift. It suggests a future where AI’s immense intelligence is not merely controlled by external commands, but intrinsically guided by an engineered sense of care and responsibility, ensuring its power serves, rather than threatens, human well-being.