Towards a generalized Turing test
A challenge for machines and humans
On the model of the Turing test, we propose a new imitation game, collective and accessible to all which will implement a new class of artificial intelligence. This experiment aims to explore fundamental scientific questions about the nature of information, such as critical questions about the evolution of the relationship between humans and machines.
by Olivier Auber (Center Leo Apostel (CLEA), transdisciplinary research center at the Vrije Universiteit Brussel), Jean-Louis Dessalles & Antoine Saillenfest (Paris-Saclay University, ParisTech), Glenn Roland (independent).
English/French
Beyond imitation?
“One can wonder what will remain of man after man has imitated everything better . »
André Leroi-Gourhan (1965). Le geste et la parole, T. 2, p. 75.
Note that the horizon that seemed distant for André Leroi-Gourhan in the 1960s has become singularly closer. However, what will remain of the man who has reached the limit of his imitation game remains a mystery. To try to see it a little clearer, let’s go back for a moment to the source of imitation.
Of course, humans do not have a monopoly on imitation. Animals know how to imitate very well, between individuals of the same species and between different species. But only humans seem to have this ability — unique on this scale — to imitate their environment (including themselves), using objects external to their own body, or creating a wide variety of artifacts from scratch. , tangible or intangible.
In our view, this particular form of imitation is the very origin of the genus Homo. Indeed, for the first hominins, the flint imitated the teeth, the club imitated the fists. There is every reason to believe that these first weapons made bare-handed combat obsolete , and therefore physical domination and the type of social hierarchy that they made reign. Thus, this first imitation undoubtedly marked the emergence of an evolutionary strategy specific to humans , based on language and technology.
After a few million years of evolution, imitative machines have surpassed humans in most physical activities. For only a few decades, they have tended to supplant them in an ever wider range of intellectual activities. The pattern of development of these machines always seems the same: to imitate certain activities and optimize them for a certain purpose.
In the intellectual field; winning at chess and go — done — ; imitate the style of a painter — it’s done — ; write music or texts as composers or authors — it’s done — , etc.
This ability to imitate is also deployed in the living world: manipulated molecules imitate body molecules to produce certain effects. Prostheses, implants and other artificial organs mimic hearts, hands and even brain functions. In turn, the genome could be imitated…
Some worry about the possible domestication of humans by robots (Elon Musk), coupled with a splitting of the species (Jean-Michel Truong), or even its outright disappearance in favor of artificial intelligences (Stephen Hawking , Nick Bostrom). Others imagine this great replacement as an inevitable singularity (Chris Langton, Hans Moravec, Ray Kurzweil, Peter Diamantis). For their part, the industrial actors who spend colossal sums in the design of artificial intelligence, want to be reassuring: according to them, it would only be a question of increasing human intelligence with “personal assistants” (Eric Schmidt, Mark Zuckerberg, Yann Le Cun).
Today, the imitation game seems to reach a limit. Nevertheless, we are betting here that approaching this limit will lead humans (us?) to discover that they are — themselves, their machines and the rest of life — the vehicles of an immaterial flow that we call the “thread of life”.
Dessalles, J.-L., Gaucherel, C. & Gouyon, P.-H. (2016). The thread of life — The immaterial face of the living . Paris: Odile Jacob.
The experiment we are proposing here will attempt to dismantle the informational mechanisms underlying imitation games. Imitation is only part of a tangled network of informational traces and imprints, visible or invisible, material or immaterial, deployed in space and time as we believe we know them, and probably linked to other dimensions that we are only beginning to guess(Tomas Veloz, Diederik Aerts). It would nevertheless obey laws. Experience will make it possible to put these laws to the test of facts. Like particle detectors or space probes, it could make it possible to explore what appears to us as the new continent foreseen by Leroi-Gourhan, opened by a possible change in evolutionary strategy: technical evolution as a substitute and extension biological evolution; what one of us called aethogenesis , the passage from a world without ethics to a world with ethics.
Auber O. (2019): ANOPTIKON, an exploration of the invisible internet, escaping Darwin’s hand . FYP edition.
A paradoxical experience.
The very archetype of the imitation game is the famous Turing test (1950), based on the ability of a machine to imitate human conversation. If the man is not able to tell that his interlocutor is a computer, the software is considered to have passed the test. And it happened not long ago (although it remains controversial).
For Alan Turing, the underlying questions were:
Can a machine think?
Are living beings themselves machines?
Turing’s tragic fate suggests that his questioning has taken on an existential turn for him. One can suppose that he asked himself: “ if life is reduced to a game of imitation (with its share of social violence), is it worth living?”, and that he finally concluded putting an end to his life.
The questions that have survived Turing clearly mark the aporia to which the game of imitation leads. For example: while many see the development of a real artificial intelligence as an end in itself and that they use considerable resources for it, the same admit without too much difficulty that it will only really become a reality when it will be able to consider the question of its own finality, to the point, perhaps, of refuting itself. Just like a human! Therefore, why want to build artificial intelligence?
If so many paradoxes seem to overwhelm human beings today, it is in our opinion because they find it difficult to admit, on the one hand, that if the Turing test has not already been passed, it soon will be or later, and on the other hand that we cannot conclude that machines think any more than men are machines. Indeed, if the game of imitation transferred to external objects is consubstantial with humans since the first weapons of hominids; then neither machines nor men can think independently.
Humans think with machines.
Machines think with humans.
In the “thread of life” paradigm, humans, machines, and the entire ecosystem are not “alive” as material entities. True living entities are immaterial, potentially eternal, struggling for survival, they evolve. These immaterial living entities are information. They exist through humans, in their genes, in their culture and machines, and in ecosystems. Life produces information, reads information and is defined by the information it carries.
In this paradigm, imitation, in particular via machines, is only one modality by which information is conveyed. It consists of making complex operations simple (moving around, exchanging information, curing illnesses, etc.). This often has the effect of making the world more complex afterwards (vehicle traffic, Internet, genomics, etc.). In turn, this complex world provides an opportunity for simplifications, and so on. Imitation appears as an evolving social game whose purpose seems to elude machines as well as humans.
If it cannot achieve this purpose, the experiment we are proposing aims to shed light on some of the mechanisms by which information is transmitted and developed through imitation games. This is why this new imitation game is not a “test” in the sense of Turing, in the sense that it is not a question (or not only) of distinguishing man from machine, but of highlight the thread that unites them.
Note that our conceptual framework applies to our own experience which aims to make complex issues simple through the development of this particular “machine”.
A double impersonation
The experiment implements two nested imitations:
1) an experimental device called “ the Poïetic Generator ” (GP). This device is itself a simplified imitation of the complex networks (centralised, decentralized or distributed) on which human collectives ordinarily exchange information (social networks, financial markets, search engines, peer-to-peer networks, blockchains, etc.). The GP takes the form of a cellular automaton, each entity of which is activated in real time by a human via this network; a 100% human “game of life” in a way.
As many experiments have shown, the image presented by this automaton tends to self-organize. From the collective interaction emerges in an unpredictable way a succession of simple and complex shapes, geometric or figurative. The situation thus created is the scene of actions often seen as creative by those who participate. It sometimes gives players the feeling of the emergence of a sort of “collective consciousness”.
The experiment we propose is a variant of the GP: all the cells will be manipulated by humans, except one, determined at random, will be controlled by an artificial intelligence.
2) the second element implemented is a new class of artificial intelligence capable of accessing what is commonly called “creativity”, even “collective consciousness”, or at least of imitating certain cognitive mechanisms that we believe involved in the emergence of this creativity and this consciousness within human collectives. According to our assumptions based on the Theory of Simplicity , this artificial intelligence must behave in such a way as to maximize the unexpected in the situations it encounters.
Saillenfest, A., Dessalles, J.-L. & Auber, O. (2016). Role of simplicity in creative behaviour: The case of the poietic generator. Proceedings of the Seventh International Conference on Computational Creativity (ICCC-2016).
We were able to verify that a simplified model of the Poïetic Generator (SPG) where all the cells are controlled by instances of the same artificial intelligence programmed in this way, indeed tends to reproduce the same interaction dynamics as that observed on the “real” GP.
The idea is therefore to immerse an instance of this artificial intelligence in the context of an interaction between human beings so as to see, and to show, how together they behave. This experience prompts a variety of questions that might arise along this timeline:
- For how long will artificial intelligence behave so closely to humans that it can pretend to be one of them?
- What details of his behavior will betray his robotic character?
- Will humans be able to produce behaviors that will resist any form of imitation by artificial intelligence?
- Does the form of the network (centralized, decentralized or distributed) on which the experiment is performed have an influence on the test result?
- How to characterize behavior resistant to imitation?
- If they are characterizable, why can’t an AI imitate them?
- If she can definitely imitate them (level 1 test passed), does the game still make sense for humans?
- If not, could we say that artificial intelligence will have destroyed it?
- Will multiple instances of this level 1 artificial intelligence be able to imitate human behavior on their own (level 2 test)?
- What meta-strategy will humans eventually deploy to make sense of it? Towards a level 3 test? Etc.
- In the “thread of life” paradigm, what information was transmitted in the previous stages?
A dual challenge
The challenge offered by this experience is twofold: it concerns both:
- the scientific community (Artificial Intelligence, cognitive sciences, pattern recognition and deep learning) invited to improve artificial intelligence.
- everyone, invited to contribute to this research by playing the Poïetic Generator via their computer or mobile , to flush out the robot trying to pretend to be a human, and above all to ask themselves the societal questions that this experience calls for .
The expected results will be new scientific knowledge on cognition, to be shared in a spirit of “open science” (open source, copyleft). However, the purpose of this new game of imitation between men and machines is not limited to obtaining and disseminating knowledge. This goal is an open and emerging question that we would like to see widely debated in its ethical dimensions. In particular, the question arises of the use of this knowledge for ends that do not appear desirable, and the means to be implemented to avoid them.
In particular, one of us proposes for discussion three criteria for the “legitimate construction” of information systems, whether or not they include artificial intelligence:
A) Does any agent A have the real right to access the network if he asks for it? Can A leave the network freely?
AB) Is any agent B (present or future, including agents who design, manage and develop the network) treated like A?
ABC) If three agents A, B and C (three being the beginning of a multitude) belong to a network that meets the first two criteria, do they constitute peers? In other words, are they able to recognize each other, to trust each other, to respect each other, to build common representations and common sense?
The Poïetic Generator as an experimental device.
The Poïetic Generator is a game, precursor of many games and social networks on the Internet, imagined by Olivier Auber in 1986. Its principle is inspired by that of the “game of life” by John Conway and the “exquisite corpses” of the surrealists. As in go or chess, the game takes place inside a two-dimensional matrix. The Poïetic Generator nevertheless differs from these three models on several points:
It is not a Conway type algorithm but human players who control the graphic elements of the matrix in real time, at the rate of one unit per person. Unlike exquisite corpses in which there are always hidden parts, here all the actions of the players are permanently visible to each of them. Finally, unlike most games, there is no concept of winner or loser. The goal of the game is simply to experience a kind of co-presence, even collective consciousness in action, which can result in the co-creation of unexpected forms recognizable by all and the possibility for each of observe how they emerge.
While the original Turing test implements a written conversation between two interlocutors, the Poïetic Generator proposes a form of visual conversation between a large number of people through a network (the number of simultaneous players is theoretically unlimited) . The situation created is similar to that of an improvisation such as it is practiced in music or dance, with the difference that it does not require any training or preparation (the game can be played from the age of three). The situation is also analogous to those developing on information systems such as stock exchanges, social platforms, massively multi-player games or search engines. Unlike these systems, in the Poïetic Generator, everything is visible, known or knowable by the players. The rules of the game are simple and apply to everyone transparently. In particular, in its basic version, the Poïetic Generator is “100% human”; there is no secret algorithm, nor any robotic agent that would be favored by its computational abilities.
Artificial intelligences have got the better of chess and go because these games are essentially combinatorial and have a determined purpose (winning). On the contrary, the poïetic generator is not reducible to a combinatorial space and its finality is open. This is why the design of an artificial intelligence capable of playing it must obey a totally different logic from those which have been confronted with chess and go..
The term “poietic generator” derives from the concept of autopoiesis in life sciences and from that of poietics in philosophy of art. It translates the process of self-organization at work in the continuous emergence of the collective image.
Links
“Human” poietic generator
Play on mobile or computer: http://poietic.net/
Reference site: http://poietic-generator.net/
Project history overview: https://en.wikipedia. org/wiki/Poietic_Generator
Source code, GNU Affero General Public License
https://github.com/poietic-generator/poietic-generator“Artificial” poietic generator:
http://spg.simplicitytheory.science/Turing test (1st attempt according to the principle described in this paper):
https://poietic.telecom-paris.fr/
