The Creator of AlphaGo Believes AI Is Heading in the Wrong Direction

David Silver introduced the concept of superintelligence to the world.

In 2016, he developed an AI program at Google DeepMind called AlphaGo, which mastered the notoriously challenging game of Go, surpassing mere mimicry.

Since then, Silver has launched his own venture, Ineffable Intelligence, aimed at creating more generalized forms of AI superintelligence. As Silver explains, this will involve focusing on reinforcement learning, where AI models gain new skills through trial and error. The goal is to develop “superlearners” that exceed human intelligence across various fields.

This strategy differs from the methods employed by many AI companies, which typically rely on leveraging the coding and research capabilities of large-language models.

Speaking to WIRED from his London office, Silver expresses skepticism towards this approach. He acknowledges the impressive nature of large-language models but argues they learn from existing human intelligence instead of independently cultivating their own.

“Human data acts like fossil fuel, offering a significant shortcut,” Silver states. “In contrast, systems that learn autonomously can be considered a renewable resource—capable of learning indefinitely and without constraints,” he adds.

Having met Silver on various occasions, I find him remarkably humble in the AI field. When discussing ideas he deems outlandish, he often breaks into a mischievous grin. Right now, however, he conveys a serious tone.

“I envision our mission as establishing first contact with superintelligence,” he says. “By superintelligence, I mean something extraordinary that should independently discover novel forms of science, technology, governance, or economics.”

Five years ago, such an objective may have seemed absurd. Nowadays, tech CEOs frequently discuss machines surpassing human intelligence and taking over entire job sectors. The notion that a new technological breakthrough could unlock superhuman AI potential has led to numerous billion-dollar startups.

Ineffable Intelligence has raised $1.1 billion in seed funding to date, boasting a valuation of $5.1 billion— a significant achievement by European AI metrics. Silver has also attracted leading AI researchers from Google DeepMind and other cutting-edge laboratories to support his mission.

Silver states his intention to donate all earnings from his equity stake in Ineffable Intelligence—potentially amounting to billions if successful—to charitable causes.

“Building a company centered on superintelligence carries a tremendous responsibility,” he shares. “It’s crucial that this effort benefits humanity, and any profits I earn from Ineffable will support high-impact charities aimed at saving as many lives as possible.”

Total Focus

Silver met Demis Hassabis, CEO of Google DeepMind, during a chess tournament as children, forging a lifelong friendship and collaborative relationship.

They maintained their connection even after Silver departed from Google DeepMind, a decision he made to pursue a completely new direction. “It’s essential that there exists an elite AI lab that is fully dedicated to this approach,” he believes. “It shouldn’t merely be a segment of a larger organization focused on LLMs.”

The limitations of the LLM-centric approach are evident, according to Silver, through a simple thought experiment. Picture releasing a large language model into a society that believed in a flat Earth. Lacking real-world interaction, Silver claims, the system would remain an enthusiastic flat-earther, despite advancements in its programming.

Conversely, an AI that can learn independently about the world could achieve its own scientific breakthroughs.