A chatbot is a computer program capable of interacting with a user through a textual or voice interface. The program imitates the verbal behavior that a human being might display during a similar conversation.

The idea that a computer program can simulate one side of a conversation in this way is the idea behind the Imitation Game. Not the film, but the game at the heart of the test devised by Alan Turing to replace the question, “Are machines capable of thinking?” To be precise, Turing did not expect the test to allow us to give a definitive answer to this question. He simply proposed setting the question aside and considering the test as a practical alternative in the search for an answer. Obviously, the film takes its name from this very game. The game is very simple: one player writes questions and receives answers from two other players. Of the three players, only two are human: the first player and one of the other two. The first player does not know which of the other two is human and which is a machine. Turing’s test consists in observing how often a machine manages to deceive the first player into believing that it is human. Obviously, the frequency with which the machine deceives the first player is directly proportional to the machine’s ability to communicate like a thinking being.

Also in response to Turing’s proposal, in 1966 Joseph Weizenbaum developed the first chatbot: ELIZA. This program’s verbal interactions simulate, in a rather parodic way, the communicative behavior of a psychotherapist conducting an initial interview with a patient. And the choice is not random: the psychotherapeutic context makes a large number of automatic responses used by ELIZA to keep the conversation going seem credible, at least superficially, despite its inability to interact meaningfully. “What do you think this might mean?”, “Can you give me an example?”, “Can you tell me why what you just said is important to you?” … And the trick worked. Although today ELIZA’s rigidity of behavior is obvious, and most of the patterns it uses are easily recognizable, at the time it was common to believe that ELIZA could actually understand the user’s sentences.

Although today’s commercially available chatbots based on LLMs (Large Language Models, that is, large-scale language models), such as ChatGPT and DeepSeek, do not need the clumsy clichés employed by ELIZA, the question of whether it is possible and appropriate to consider expanded versions of the notion of thought that might also include the computational activity of these systems has no obvious answer.

The philosopher John Searle addressed this question in his 1980 article “Minds, Brains and Programs.” He did so through a thought experiment known as the Chinese Room (that is, the Chinese room; the 1968 film has nothing to do with it this time). The thought experiment consists in imagining a person who can neither speak, read, nor write Chinese, locked in a large room containing an enormous number of tablets engraved with Chinese characters and a large instruction book. The instructions simply associate sequences of Chinese characters with sequences of Chinese characters, like this:

 你好吗 → 我很好，谢谢。

我今天很累。→ 很抱歉。

…

There is a slot in one of the room’s walls. When tablets bearing Chinese characters are inserted through the slot, the person in the room takes them, looks up the Chinese characters engraved on them in the instruction book, and returns through the slot the tablets that, according to the book, should be used to answer those received. It is important to emphasize that the room, the number of tablets, and the instruction book must be truly enormous, since they must allow the person to respond sensibly to every possible sentence that can be formulated in Chinese. But this is a thought experiment, so their being enormous is not a problem. Well, it is not impossible to imagine (assuming that the enormity of the objects in question is sufficient, that the person in the room never runs out of time, and that other physical conditions bordering on the impossible are met) that the person in the room manages to hold a more than satisfactory conversation with an interlocutor who knows Chinese (certainly more satisfactory for the interlocutor than for the person in the room). Searle therefore asks whether we must admit that the person in the room knows Chinese. There are various answers to this question. Searle, for example, argues that the person does not know Chinese, but is merely mechanically following the book’s instructions. Some argue that, even if it is true that the person does not know Chinese, one can instead say that the room does. The problem clearly admits of no definitive solution, and this is not the place to discuss it further. What is very interesting, however, is that we no longer need to imagine poor individuals locked in rooms full of tablets, consulting instruction books for endless periods of time. A machine that, by carrying out a mechanical process (however sophisticated), manages to hold a satisfactory conversation with a human interlocutor exists.

Francesco A. Genco

(translated by MIRAI)