Don’t be fooled—LLMs don’t reason

On an afternoon in Seoul in March 2016, I watched a program I helped build put a stone on the fifth line of a Go board in what looked like a gift to its human opponent. Move 37 in game two of the five-game match looked so absurd that some commentators thought it was a programming glitch. It wasn’t. AlphaGo won the game, ultimately triumphing 4-1 over Lee Sedol, one of the greatest professional Go players of all time. “I thought AlphaGo was based on probability calculation and that it was merely a machine,” Lee said afterwards. “But when I saw this move, I changed my mind. ” When Deep Blue defeated then reigning world chess champion Garry Kasparov in 1997, it did so by looking six to eight moves ahead per player and evaluating 200 million chess positions per second, using rules hard-coded by humans. Go is a vastly more complex game. A stone’s worth depends on how distant groups and territory unfold over dozens of moves. Computing even a fraction of the possible outcomes would take a supercomputer billions of years. To win, AlphaGo had to sense who was ahead at a glance and even invent moves no human had thought to play. That is why many accounts of AlphaGo’s match against Lee portray move 37 as a flash of pure machine intuition. But that is a misunderstanding. It was actually AlphaGo’s powers of reasoning that made this creative choice—and these are powers that today’s AI lacks. If we want future AI systems to produce trustworthy results and really novel insights in fields like science and medicine, we need to equip them with genuine reasoning capabilities of this kind. AlphaGo is made up of two systems. The first, its policy network, was trained to guess what move a strong human would play. This “intuitive” part regarded move 37 as nothing special—a play that had a roughly one in 10,000 chance of being made by an expert human player. What made AlphaGo choose it was the program’s search machinery, which looked beyond immediate plausibility and weighed the future consequences of proposed moves. It explicitly constructed and searched a game tree with thousands of branches, each representing a different possible future. A well-known theory in the behavioral sciences, popularized by Daniel Kahneman, distinguishes between two modes of human thought: System 1 is fast, gut-level, effortless; system 2, slow, step-by-step, and deliberative. AlphaGo offered a striking machine analogue of that split. Its networks supplied the hunches—this move looks promising, this position looks won—and its search supplied the deliberation, testing those hunches against the moves and countermoves that would follow. As in human cognition, neither half works alone. Intuition alone would never have opted for move 37, and brute-force search would have struggled to sieve through all the many possible moves. This is strikingly different from the way today’s AI models work. A large language model picks the next token, over and over. That amounts to system 1 in action—fast, associative, and surprisingly good pattern completion across almost every subject people write about. Not long after ChatGPT debuted, the field realized that language fluency alone falls short of true usefulness. The apparent solution was to make models that deliberate: Instead of answering immediately, they can now generate intermediate steps that decompose a problem, carry forward partial results, and influence subsequent reasoning—a process known as chain of thought. The gains have proved real, above all in mathematics and coding. But unlike AlphaGo’s search, this does not introduce a genuinely separate reasoning mechanism: The intermediate reasoning is still produced by the same next-token prediction process, iterated for longer before the model commits to an answer. Three shortcomings prevent what chatbots do from qualifying as reasoning (in a way that a scientist might recognize). First, these models typically maintain no explicit, persistent, and inspectable epistemic state. There is no open ledger that lays out the hypotheses a model is considering, the confidence it has in various explanations, the evidence it’s weighing, and the unresolved questions it’s holding onto—all things that should be systematically revised as new information arrives. Second, they lack a clean separation between what the system knows and how it manipulates that knowledge. Knowledge and reasoning are inextricably interwoven in the weights of the neural network—there is no independent, explicitly represented set of beliefs. Third, while the chains of thought chatbots produce look like deliberation, research has demonstrated that the bots often concoct them after the fact, reaching an answer by one route but reporting another. This is a problem because in the high-stakes applications we all care about, such as medicine, engineering, and scientific research, it matters not only what a system concludes but also how it arrives at its conclusion. When mistakes happen—for example, in medical diagnosis and treatment—we need to be able to pinpoint what went wrong: Was the system’s reasoning at fault, did it draw on invalid evidence, or did it make incorrect assumptions? This is why I recently left my position at Google DeepMind. I believe we need a fresh approach to machine reasoning—one that draws on AlphaGo’s architecture. AlphaGo maintains a record of what it knows about a given position: the game tree. This data structure contains all the variations, the possible futures, that AlphaGo has considered, each move and position being annotated with judgments made by its neural networks. As its reasoning progresses, AlphaGo updates the game tree and eventually synthesizes the information in it to decide which move to make. Similarly, for general reasoning a system should maintain an epistemic state that represents what the system holds as settled, what it doubts, what it has ruled out, which questions stay open. Reasoning can then be understood as a sequence of moves that change the epistemic state to advance knowledge and reduce uncertainty: deducing consequences, breaking problems into parts, and—crucially—deciding what question to ask, calculation to perform, or experiment to run next. Of course, open-world reasoning is harder than playing a board game such as Go or chess. In the real world the current state of affairs is only partially known, the set of available actions is large and variable, and the consequences of actions are stochastic or unknown. But recent advances in LLMs and other neural models now give us the capability to take on such general reasoning problems. For example, LLMs can suggest ways of tackling a problem on the basis of what is known and what resources are available. They can interact with tools via APIs or code and help assess whether a claim is supported by available evidence. Most important, to keep the system honest, an independent part of the system must evaluate each move by how much it actually resolves uncertainty, updating beliefs only when the change is backed by evidence. Once these rules are enforced, the model can accumulate certified knowledge and improve its reasoning policy by learning from past reasoning experiences. You can think of such a system as the scientific method on steroids, with the purpose of producing knowledge that can withstand scrutiny. I do not think we reach trustworthy machine intelligence by making system 1 bigger. Scale sharpens intuition, but it does not make intuition more deliberative. Move 37 mattered because a machine held a position, weighed the possible futures, and chose the move its artificial instincts would likely have rejected. Society needs such creative moves in drug discovery, materials, climate, diagnosis—fields where the board looks nothing like a Go board and nobody hands us the rules. We will get such insights only from systems that reason—systems whose conclusions arise from an auditable sequence of evidence, inference, and belief revision rather than from a convincing story told after the fact. Thore Graepel is chair of machine learning at University College London. He was a core member of the AlphaGo team at DeepMind and works to ensure that AI benefits human flourishing. Deep DiveStay connectedIllustration by Rose WongGet the latest updates fromMIT Technology ReviewDiscover special offers, top stories, upcoming events, and more.