SUMMARY. Also find a presentation of the third quarterly on the website of The research.
From HAL 9000 to Terminator via C-3PO, science fiction gives us machines with capacities that one could qualify as human, even superhuman. Today, in reality, we are dealing with artificial intelligence (AI) programs that beat chess or go champions, such as AlphaZero, created by researchers at Google’s DeepMind company. Or software that writes newspaper articles automatically. Thus, the GPT-3 language model, developed by the company OpenAI (co-created by Elon Musk) and opened to users in July 2020, is capable of producing written content having all of a human production. In all of these systems, processes that seem far removed from thought result in these actions, which are indistinguishable from those of a conscious human.
Machines that mimic thought?
This is what is at stake in the debate: do these machines mimic thought, without being conscious, or is our brain also functioning in this way, without our knowing it yet? The parallel between the still mysterious functioning of our brain organ and that of artificial intelligence raises questions. For AI, researchers train the model with a deluge of data, and the result is well-constructed sentences and texts. For our brain, it is continuously bathed in a smaller set of information, but on which we rely to make inferences from which we can act. Either way, we are still unaware of the intimate workings of these systems. The only thing of which we are certain is of the existence of our own consciousness, since the experience is eminently subjective (the “Cogito ergo sum” – “I think therefore I am” – of Descartes). But what about that of my neighbors? I only know what they tell me, but should I believe them? By taking into account a similar biological substrate – namely “1.5 kg of fat and protein, locked in a bone shell”, as the neuroscientist Victor Lamme jokes in our pages -, we can deduce without act faith that the men and women who inhabit the planet are aware.
Brain and artificial intelligence
With regard to other animals, the observation of meta-cognitive capacities and induction also allow us to conclude that there are conscious capacities. There remains the question of machines. The reasoning is more delicate and involves what we mean by this concept. For some scientists, only what we do matters, while, for others, only what we are, which prohibits de facto an artificial consciousness. We can bet in any case that the pioneering attempts to pierce the mysteries of consciousness, which have recently multiplied, will succeed. What if the answer required a deep understanding of artificial intelligence alongside that of the brain? When a robot recognizes itself in a mirror, we can start to be afraid.