A just published study has presented a new neurocomputational model of the human brain, which could shed light on how we develop complex cognitive skills, and advance research on a artificial intelligence neural.
The study was conducted by a French-Canadian team: French scientists from the Institut Pasteur and Sorbonne University, Canadian scientists from the Quebec Artificial Intelligence Institute and the University of Montreal.
The model was featured on the cover of Proceedings of the National Academy of Sciences of the United States of America (PNAS), I link it here.
Three levels of development…
In essence, the model describes neural development across three hierarchical levels of information processing.
The first level of information processing, the sensorimotor one, explores how the internal activity of the brain learns different patterns from perception and associates them with actions.
Then the second levelThe cognitive one examines how the brain contextually combines these patterns.
The conscious levelFinally, consider how the brain dissociates itself from the outside world and manipulates learned patterns (through memory) that are no longer accessible to perception.
… And two kinds of learning
The study of the new neurocomputational model of the human brain also places emphasis on two fundamental types of learning.
One is Hebbian learning (by neuropsychologist Donald Hebb), associated with statistical regularity, i.e. repetition. The other is reinforcement learning, associated with reward and the neurotransmitter dopamine.
The interaction of the two types of learning with the various levels of information processing could give us new awareness of the fundamental mechanisms underlying cognition.
How does the human brain model presented in the study work?
The model solves three tasks of increasing complexity, from visual recognition to the cognitive manipulation of conscious perceptions. Each time, the team has introduced a new core mechanism to allow him to progress.
The results highlight two fundamental mechanisms for the multilevel development of cognitive abilities in biological neural networks:
- Synaptic epigenesis, with local-scale Hebbian learning and global-scale reinforcement learning;
- Self-organized dynamics, through the spontaneous activity and the balanced excitatory/inhibitory relationship of neurons.
It's like having established how a cybernetic brain is turned on and operated. “Our model demonstrates how AI proceeds with biological mechanisms and cognitive architectures that could lead to artificial consciousness,” says the team member William Dumas, assistant professor of computational psychiatry at the University of Montreal.
New computational model of the brain: where will it take us?
Can a model like the one studied bring out a conscience in an artificial intelligence?
“Reaching this milestone may require the integration of other factors,” says Dumas, “such as the social dimension of cognition.” And that's what researchers are now trying to do: the next experiments aim to make two "simulated cybernetic brains" interact, to see what comes out.
In short (Italian only), the team believes that anchoring future computational models in biological and social realities will not only help us understand the fundamental mechanisms underpinning cognition.
It will also help us provide these mechanisms to an artificial intelligence, which one day (unlike today) will have some form of self-awareness.