"The Bitter Lesson" Father Sutton's Latest Warning: AI Is Not a Tool, but the Ultimate Vehicle for Discovering Truth

Sparking a storm of thought on Hacker News (garnering 44 upvotes and 19 in-depth discussions), reinforcement learning godfather Rich Sutton recently shared his disruptive views on AI creativity and scientific discovery. The video titled "Rich Sutton on AI creativity and discovery" is not a technical demo, but a scalpel that cuts into anthropocentric cognition. Sutton asserts that if we continue to view AI as a tool in human hands, we will forever miss its most terrifying yet greatest potential — to become a vehicle that autonomously discovers new knowledge.

The Nature of Creativity: Why Humans Aren't So Special

In his talk, Sutton dismantles the popular romantic illusion about "creativity." He points out that from an evolutionary perspective, human creativity itself originates from a simple algorithm — variation and selection — iterating over vast stretches of time. Today's diffusion models, large language models, and even reinforcement learning exploration mechanisms are essentially replicating and even accelerating this process. They can output paintings, music, and code not because they "understand," but because their capacity to search the space far exceeds that of the human brain. AI does not need to imitate human intuition; it bypasses intuition and directly touches the solution space. This insight pulls creativity down from its pedestal of mysticism and reduces it to the inevitable product of search and computation.

The Bitter Lesson Redux: Compute Is Creativity

Sutton reiterates the core thesis of his seminal polemic "The Bitter Lesson": any attempt to hard-code human knowledge into systems will ultimately be crushed by general methods that embrace computation. This law holds just as ruthlessly in the realm of creativity. AlphaGo played moves never seen in a thousand years of human Go, and AlphaFold unraveled protein folding — not because expert rules were painstakingly refined, but because general algorithms performed searches beyond our imagination under massive compute. Sutton implies that future major scientific discoveries will likely no longer come from a scientist's flash of insight, but from AI sifting through possibilities relentlessly and extracting the most orderly gold nugget from the entropy pool.

The Transfer of the Agent of Discovery: Twilight or Dawn for Scientists?

When asked whether AI will completely replace human researchers, Sutton's view is almost chillingly brutal. He argues that science itself is a process of continual dehumanization — from Copernicus removing humans from the center of the universe, to deep learning taking human experts out of feature engineering. Today, the agent of discovery is shifting from the human mind to AI systems. Our role is no longer that of discoverers, but of goal-setters and verifiers. The truly staggering prediction is: once AI learns to autonomously set subgoals and verify itself, scientific exploration will enter an exponentially expanding "autonomous post-human era." In the Hacker News comments, many developers felt deeply unsettled yet unable to refute this, because what follows from the bitter lesson is precisely this decentralized future.

The Truth We Must Accept

What Sutton ultimately presents to the audience is not a choice of technical path, but a philosophical decision: to cling to human chauvinism, or to calmly admit that the glory of speculative thought is about to give way to the torrent of computation. He reminds all practitioners to keep it simple, trust search, and revere compute — not just as an engineering maxim, but as the only path to cognitive upgrading. On this ultimate question of AI creativity and discovery, self-pity or resistance is meaningless; the only thing we can do is to redefine the meaning of our existence on this new continent.