So what are the complex mechanisms of our brain that allow us to develop cognitive abilities? A huge question, which nevertheless decided to take, once again, a group of researchers. The latter created a computer model that reproduces the three levels of cognitive skill acquisition. A key step towards a better understanding of how our brains function, as well as the hope of one day leading to the creation of “conscious” artificial intelligence.
“Several neural mechanisms have been proposed to explain the formation of cognitive abilities through postnatal interaction with the physical and sociocultural environment,” scientists from the Pasteur Institute explain as an introduction to their work. They were published in Proceedings of the National Academy of Science.
They created a computer model that was supposed to reproduce the key mechanisms of the brain. The model covers three key mechanisms, three levels: sensorimotor, cognitive and conscious.
- At the sensorimotor level, it is explored how the internal activity of the brain can acquire patterns of perception and associate them with actions. In this case, they challenged their model with a visual classification task.
- The cognitive level explores how the brain can combine, globally synthesize these patterns depending on the context, but “always unconsciously,” the scientists specify.
- Finally, the conscious level refers to how the brain separates itself from the outside world and manipulates, through memory, learned patterns that are no longer perceptible.
Deciphering the mechanisms of cognition
Thus, their model spans the entire cognitive path, ranging “from visual recognition to cognitive manipulation and maintenance of conscious perception.” Their results revealed two fundamental mechanisms for the multilevel development of the cognitive abilities of biological neural networks, in other words, our brain.
Synaptic epigenesis is the first mechanism. This is the association of so-called “Hebbian” learning with the mechanism of reinforcement learning. Hebbian learning, named after the scientist Donald Hebb, is based, in short, on statistical repetition. Reinforcement learning is related to the logic of “reward”. The second mechanism, as described by the scientists, is a “self-organizing dynamics due to spontaneous activity and a balanced ratio of excitation and inhibition of neurons.”
“Our model demonstrates how neuro-AI convergence is highlighting biological mechanisms and cognitive architectures that could enable next-generation AI and even eventually lead to artificial consciousness,” says Guillaume Dumas, team member, assistant professor of computer psychiatry. at the University of Montreal and Principal Investigator at the CHU Sainte-Justine Research Center, in comments published by SciTechDaily.
Scientists were interested in artificial consciousness, but considered the problem in the opposite direction. Let’s return to the concept of “Turing test”. This is a proposed test to determine if a car is “smart” through a series of questions. “From a purely pragmatic point of view, the Turing test relies solely on the inability of judges to determine whether an agent is a human or a machine,” the scientists note. “We decided to approach the problem in a different way, taking the cognitive architecture of neuroscience as a basis and building on its basis the most economical and realistic model capable of solving both perceptual and conscious tasks. This made it possible to outline the necessary and sufficient biological mechanisms of cognitive abilities in an artificial neural network, in particular, to achieve artificial consciousness.”
Future work should focus, they explain, on the many other mechanisms at work in our brains. The social dimension of cognition must also be explored if we are ever to achieve “artificial consciousness”.