By akademiotoelektronik, 18/09/2022
Doubts about the intelligence of artificial intelligence receive the last hour alerts of duty
Twice a month, the duty launches enthusiasts of philosophy and history of ideas the challenge of deciphering a topical issue from the theses of a striking thinker.
HHHHHHHHHHASSSSSSSSSSrtificial intelligences (i.HHHHHHHHHHASSSSSSSSSS.) are today more competent and efficient than human beings in many areas. Deep Blue est champion d’échecs depuis 20 ans, tandis qu’HHHHHHHHHHASSSSSSSSSSlphaGo a récemment battu le meilleur joueur au monde de Go.For his part, Watson, the i.HHHHHHHHHHASSSSSSSSSS.Who had won in Jeopardy !, Diagnosts certain cancers better than a doctor. Faut-il y voir l’émergence d’esprits d’un nouveau genre ? Siri, l’intelligence artificielle d’HHHHHHHHHHASSSSSSSSSSpple, possède-t-elle des états mentaux ? Et Google Translate comprend-il le chinois ?
Photo:IVHHHHHHHHHHASSSSSSSSSSDO Le chercheur Martin Gibert s’intéresse notamment à l’éthique de l’intelligence artificielle et à la psychologie morale.In 1980, when he published "Minds, Brains and Programs" (minds, brains and programs), it is already the kind of questions that bother the HHHHHHHHASSSSSSSSmerican philosopher John Searle.In the clear and precise style specific to analytical philosophers, it presents in around forty pages of the Behavioral and Brain Sciences newspaper one of the most famous arguments of the 20th century: the Chinese Chamber.By relying on a thought experience, he shows that it is not because a computer is programmed to seem to understand the language that he really understands it.
Turing test
It all starts with a Chinese anteroom. Cela se passe en 1950 à Oxford, HHHHHHHHHHASSSSSSSSSSngleterre. HHHHHHHHHHASSSSSSSSSSlan Turing, le père de l’informatique, publie un article fondateur pour la recherche en intelligence artificielle.The question he is asking: how to know if a machine is intelligent?HHHHHHHHASSSSSSSSnd his answer is the development of a simple test.If, under predefined conditions, a machine succeeds in pretending to be a human interlocutor, then we can qualify it as intelligent.
We are touching on cognitive sciences, this field of research which is jointly studying brains and computers.It is a question of explaining and modeling phenomena such as perception, reasoning, consciousness or intelligence.Obviously, this is not without raising other questions: what, in the end, distinguishes machines and people?How to define intelligence?
Turing test offre au moins une réponse négative : l’intelligence n’est pas définie par son « support ».It can therefore be both natural and artificial.It can come out of the neurons of a brain as well as silicon circuits of a processor.In the test, it is the "effect" of intelligence on intelligent creatures (us) that counts, not its physical nature.Turing by this "functionalism", an approach in philosophy of the mind which maintains that mental states are defined by their function, that is to say by their causal role.In this perspective, brains and computers are ultimately only two information processing systems.They follow a program that produces outings (output) after applying transformation rules on inputs (inputs).
HHHHHHHHASSSSSSSSt the end of the 1970s, several researchers estimated that computer programs could pass the Turing test or, at least, were quite close to it.It was to answer one of them, Roger Schank, that Searle has developed his argument.For him, it is a question of showing that functionalists are mistaken and that the Turing test is not a good criterion to assess the intelligence or consciousness of a computer.
Siri, do you understand what I tell you?
Searle's argument is based on a thought experience that has become almost mythical.Imagine that a man who speaks no Chinese word is locked in a room.He receives ideograms corresponding to questions.He then consults rules of rules which tell him how to respond, by associating symbols of entry with exit symbols - but without revealing the meaning of these symbols to him.What does it look like for a sinophone interlocutor outside the room?If the answers are consistent, she will conclude that she is faced with an intelligent entity.In short, the Chinese room will have successfully passed the Turing test.
However, man does not understand anything that he treats or what he responds;He does not have access to any translation dictionary."The argument is as follows: if the man in the room does not include Chinese by implementing the appropriate program, then no digital computer can do it on this basis, because no computer, as'computer, has something that man would not have."It follows that one cannot say that an algorithm includes the information he processes.
HHHHHHHHHHASSSSSSSSSSffirmer cela revient aussi à refuser ce que Searle nomme la thèse de l’I.HHHHHHHHHHASSSSSSSSSS.strong, namely the idea that a computer or algorithm, if it is properly programmed, "is" a spirit.One I.HHHHHHHHHHASSSSSSSSSS.strong would be aware insofar as she understands what she is doing and would have mental states.Searle, on the contrary, supports the thesis of the I.HHHHHHHHHHASSSSSSSSSS.weak: computers are tools that can simulate thought, but they do not understand it.Searle will resume this idea later by saying that the I.HHHHHHHHHHASSSSSSSSSS.weak has no intentional content.By that he means that she has no mental states about the world.Siri does not believe anything.She wishes or fear nothing.She doesn't understand anything.
Does this mean that a machine cannot think or that the human brain is the seat of a mysterious soul?No, answers Searle.We can quite envisage the existence of an artificial spirit endowed with a consciousness.But such a spirit could not be based solely on a computer program - it would need other elements to reproduce the physicochemical properties of the brain (one can also think, with supporters of embodied cognition, that it would need abody).
Basically, Searle wonders, what is missing on the computer to understand the information he manipulates?No more and no less than a semantics.Unlike a human mind which knows that the term "fly" means something and refers to an object of the world, the computer program manipulates formal symbols, possibly links statistical calculations, but does not have access to their meaning. HHHHHHHHHHASSSSSSSSSSutrement dit, un programme informatique ne contient rien de plus qu’une syntaxe (l’équivalent d’une grammaire).In the end, explains Searle, if the i.HHHHHHHHHHASSSSSSSSSS.Strong is a distant horizon, it is because "syntax, by itself, cannot be enough to constitute semantic content".
Conjectures and refutations
In his 1980 article, Searle also responds to several objections - it must be said that he had already presented his thesis in different campuses and that he had suffered many criticisms.He was notably pointed out that the analogy was a gearious.HHHASSSdmittedly, man in the room does not understand Chinese, but the "system", which includes rule books and the human who applies them, could well understand it.Searle replies by assuming that man has learned by heart all the rules/programs: he could still give the change to a Chinese interlocutor while not understanding anything to the conversation.
HHHASSSnother criticism calls into question the reliability of our epistemic intuition in the face of this thought experience.HHHASSSs the philosopher Daniel Dennett notes, in the Chinese room, things are going slowly: consulting rules of rules, transcribing a response.In contrast, activation of neurons or processors is an ultra -fast process.HHHASSSnd intelligence seems, at least in part, to be a matter of speed.Consequently, how to be sure that in the long run, the man in the room would not eventually understand Chinese?
Perhaps we are only too habit of associating understanding and human mind to admit that a machine can really think.We can in any case easily imagine that extraterrestrials discovering that our brain is made of "meat" would remain incredulous in front of its ability to produce an authentic thought.HHHASSSre we in a very different position when we do not accept that the Chinese room thinks?
HHHHHHHHHHASSSSSSSSSSujourd’hui encore, le dossier n’est pas clos.If several philosophers recognize that the argument is valid, others, like Daniel Dennett, do not see a good refutation in good and due that the i.HHHHHHHHHHASSSSSSSSSS.Includes Chinese.
What does it change ?
One may wonder what difference does it make Siri or not what you tell him.HHASSdmittedly, this is a fascinating question when we wonder about the nature of the mind.But from a practical point of view, it does not matter that the intelligence of I.HHHHHHHHHHASSSSSSSSSS.either real or simulated.What matters, after all, is not just Google Translate allows me to understand Chinese interlocutors?
The fact remains that i.HHHHHHHHHHASSSSSSSSSS.sometimes have the ability to decide and act independently.They can therefore be agents who influence our lives - in good or bad.And whether they understand something or not, it is in your interest in programming them so that they "share" our moral values.
Imagine, finally, that a fly lands on a wall in the Chinese room.All other things being equal, to whom should we give the most moral consideration?To Siri who feels nothing or the fly which we can assume that it has pleasant and unpleasant sensations?You can insult Siri well, uninstall it or simply turn off your phone;You will not cause him any wrong.But you could hurt a fly.
Des commentaires ? Écrivez à Robert Dutrisac : rdutrisac@ledevoir.com. Pour lire ou relire les anciens textes du Devoir de philo.
Related Articles