Further research, using neural-network models, researchers attempted to simulate the functioning of humans and baboons with artificial intelligence inspired by basic mathematical ideas of what neurons do and how neurons are connected. These models – statistical systems operated by high-dimensional vectors, matrices multiplying layers by number of layers – successfully match the performance of the baboon but not with humans; They failed to reproduce the regularity effect. However, when researchers created a soup-up model with symbolic elements – the model was given a list of properties of geometric regularity, such as perpendicular, parallel lines – it closely mimics human influence.
These results, in turn, set a challenge for artificial intelligence. “I like advances in AI,” said Dr. Dehen said. “It’s very impressive. But I think there’s a deeper aspect missing, which is the symbolic process” – that is, the ability of the human brain to manipulate symbols and abstract concepts, as the human brain does. How We Learn: Why Brains Learn Better Than Any Machine … For Now. “
Joshua Bengio, a computer scientist at the University of Montreal, agrees that the current AI lacks anything related to symbols or abstract logic. Dr. Dehen’s work, he said, “is that the human brain uses abilities that present evidence that we do not yet see in advanced machine learning.”
That, in particular, he said, when we combine symbols when composing and recomposing pieces of knowledge, which helps us generalize. This distance may explain the limitations of AI – for example, self-driving cars – and the clumsiness of the system when confronted with different environments or scenarios from the training base. And that’s a sign, Dr. Where does AI research need to go, Bengio said.
Dr. Bengio noted that symbolic-processing strategies dominated “good old-fashioned AI” from the 1950s to the 1980s, but that these approaches were less motivated by a desire to mimic the capabilities of the human brain than by logic-based reasoning (e.g., proof of verification theorem). Then came the statistical AI and neural-network revolution, which began in the 1990s and gained traction in the 2010s. Dr. Benjio was the pioneer of this deep-learning method, which was directly inspired by the network of neurons in the human brain.
Their latest research proposes to expand the capabilities of the neural-network by training them to create or visualize icons and other representations.
It’s not impossible to make abstract reasoning with neural networks, he said, “it’s just that we don’t know how to do it yet.” Dr. Dr. Benjio has. Is a large project connected with. Dehen (and other neuroscientists) use it to investigate how human conscious process forces can inspire and promote next-generation AI. That is our understanding, “said Dr. Bengio.