Archives par mot-clé : AI

Computer program “AlphaGo” defeats a human professional player at the game of Go

Google DeepMind announced yesterday that the program « AlphaGo » managed to defeat the European Champion at the game of Go.

Demis Hassabis, the Google engineer who makes the announce starts by saying that « games are a kind of microcosmos of the outside world », which tends to make us believe that only computational progress would be needed to go from mastering the microcosmos to mastering the outside world – there would be no qualitative difference, only a difference of scale.

He follows on saying that « another way of viewing the complexity of Go is the number of possible configurations of the board is more than the number of atoms in the universe ».

Indeed, the problem seems linked with the number of atoms in the universe. With a game of Go, we take only the board into account, and we leave aside what happens around. With a « game of the outside world », we would need to take everything into account.

That make us wonder (as Kurzweil does): what if we could take the universe as a game of Go, what if we had the computing power to take into account all the atoms of the universe and assess all their resulting combinations?

Quentin Meillassoux, in his conference Temps et surgissement ex nihilo (in French) discusses the possibility of calculating probabilities at the scale of the universe.

Comparing the set of possible universes a suite of transfinite numbers (according to the definition of Cantor), he infers that the set of possible universes is not a countable totality – which makes impossible any calculation of probability on it:

« Or nous savons, nous qui venons après la révolution cantorienne de l’ensemblisme, que rien ne nous permet d’affirmer de la sorte que le concevable est nécessairement totalisable. L’événement Cantor est la détotalisation du nombre, détotalisation dont l’autre nom est transfini.

Qu’est-ce que le transfini? La pluralisation incloturable des quantités infinies. L’ensemble B des parties d’un ensemble A est toujours plus grand que A. A fut-il infini.


La force du théorème cantorien c’est d’avoir généralisé ce débordement d’un nombre d’éléments par le nombre de ses regroupements. Chaque fois que vous avez un ensemble, comptez ses parties, vous obtenez, et vous êtes assurés de l’existence d’un ensemble qui aura plus d’éléments que ce premier ensemble, parce qu’il sera composé des parties, des regroupements possibles du premier ensemble. On peut donc construire une suite illimitée, non pas simplement d’ensembles finis, mais d’ensembles infinis, car la théorie des ensembles pose l’existence d’un premier infini, et à partir de ce premier infini, et à l’aide notamment du théorème de Cantor, nous pouvons construire une succession de nombres de cardinaux transfinis, car cette succession d’infinis plus grand les uns que les autres on les appelle les nombres cardinaux transfinis, ou suite des alephs. Et cette successsion n’admet aucun achèvement, aucun bouclage, il n’y a pas d’ensemble ultime, il n’y a pas d’ensemble de tous les ensembles, c’est à dire que c’est une succession d’infinités, plus grande les unes que les autres, qui n’admet pas de recollection ultime. Il n’y a pas d’ensemble tout.

Ce qui est énoncé comme impensable c’est l’ensemble du concevable. Si on admet que le concevable est ce qui est logiquement et mathématiquement concevable et non pas donné dans l’expérience, eh bien le concevable n’est pas totalisable. Et l’inférence probabiliste n’est pas légitime. »

More about DeepMind breakthrough here

Listen to the conference « Temps et surgissement ex nihilo » here


The Policeman’s Beard Is Half Constructed

The Policeman’s Beard Is Half Constructed is a novel wrote in 1983 by Racter, a computer program made by William Chamberlain and Thomas Etter.

« Wrote » might be an exaggerated word as the text wasn’t generated but computed using templating.

MobyGames reviewed its code:

« Racter strings together words according to « syntax directives », and the illusion of coherence is increased by repeated re-use of text variables. This gives the appearance that Racter can actually have a conversation with the user that makes some sense, unlike Eliza, which just spits back what you type at it. Of course, such a program has not been written to perfection yet, but Racter comes somewhat close.

Since some of the syntactical mistakes that Racter tends to make cannot be avoided, the decision was made to market the game in a humorous vein, which the marketing department at Mindscape dubbed « tongue-in-chip software » and « artificial insanity ». Not a true « AI » by any stretch of the word, but a unique program that is well worth a look as an indication of where the field of artificial intelligence was heading in 1984.

According to the Racter FAQ, co-designer William Chamberlain even released a book called « The Policeman’s Beard is Half Constructed » (Warner Books, NY. 0-446-38051-2, paper $9.95) before the release of the program, the authorship of which he attributed solely to Racter. »

You can read the novel on Ubu or talk to Racter.

A conversation with Racter
A conversation with Racter