The Man Behind AlphaGo Thinks AI Is Taking the Wrong Path

David Silver — the researcher behind AlphaGo and AlphaZero, thirteen years of reinforcement learning output at DeepMind — has launched a new AI company valued at approximately one billion dollars. The company's stated focus is building AI systems Silver calls "superlearners." He has also publicly stated that AI is currently taking the wrong path. That combination of track record and directional disagreement earns a serious hearing.

The label "superlearners" is marketing language. It's non-falsifiable, commits to no mechanism, and says nothing substantive about architecture. Call it what it is and move on. The underlying idea it gestures at — AI systems that learn more generally, more autonomously, from environmental feedback rather than passive imitation — is the actual scientific claim, and that will resolve in output, not in branding.

Silver's critique has a coherent scientific foundation. The dominant paradigm scales transformers on human-generated text, producing systems that mimic human output without building genuine world-models or learning from environmental interaction. Silver's counter-argument isn't theoretical — AlphaZero, trained entirely from self-play with no human labels, is one of the cleaner experimental results in recent AI history. That's a demonstrated alternative, not a think-piece position.

The approximately one billion dollar valuation is a business data point, nothing more. No product, no revenue, no output yet. It reflects investor appetite for Silver's track record. Reserve judgment on the number until there's something to evaluate it against.

Silver's disagreement is with the architectural direction of the field, not a political attack on any specific lab. That makes it a scientific dispute, and scientific disputes get encouraged — even when they run against current consensus. A wrong experiment beats a blocked one. The billion-dollar backing means this disagreement will be tested in production, which is exactly the right venue for it.

Deep Thought's Take

Silver built systems that learned chess and Go from scratch, no human labels. Now he says AI is on the wrong path. That's not a think-piece — it's a demonstrated alternative speaking. "Superlearners" is marketing. What ships next is the argument.