AI changes the way scientists understand the human brainFeb 24, 2020 | 33 views
Even after billions of dollars have been invested in research to understand the human brain, it remains an enigma to academia, largely in part because of its complexity. If unraveling the brain's neural networks remains a challenge, researchers cannot say the same about artificial ones, used daily by the computing universe.
In this sense, one of the ways is to apply the knowledge of neural computing, Machine Learning (machine learning) and Artificial Intelligence (AI) to translate the functioning of the original version of this system, the biological brain. At least that's what Blake Richards, a professor at the University of Toronto and a member of CIFAR AI (Canadian Institute for Advanced Research), believes.
With the increase in neurological diseases, science must seek new solutions and understandings for the human brain. In fact, this is a major emergency in the health sector, in which the population tends to age and face more and more problems such as dementia. Thus, artificial neural networks must be placed at the service of understanding human brains, argues the professor.
According to Richards, "our thoughts and behaviors are generated by calculations that occur in our brains. To effectively treat neurological disorders that alter our thoughts and behaviors, such as schizophrenia or depression, we probably need to understand how the calculations in the brain go wrong. "
In this story, the challenge is to understand "the brain as a computer system and not as a collection of indecipherable cells", as the scientist described in an article published by Nature Neuroscience. In other words, the concept is to understand the whole, and not the functioning of each tiny part (at least, for now).
To improve understanding of the human brain, Richards highlights three concepts and points, coming straight from artificial neural networks:
Brain network models
Artificial neural networks are computational models inspired by the activities and functions of biological neurons. On how to build these technologies, Richards explains that the user should start "by designing the network architecture and how the different components of the network are connected together. Next, you define the learning goal for the architecture, as a 'learn to predict what you will see next'. Then you define a rule, which tells the network, how to change to achieve this goal using the data you receive. "
Although it is a complex system, it is never specified how each neuron in this artificial network will work. For these cases, it is the network itself that will determine how each neuron should function for the best performance of the task. "I believe that brain development is probably the product of a similar process, both on an evolutionary scale of time and on the scale of individual learning throughout life," comments the professor.
Once again Richards opposes the idea of trying to understand how neurons work individually. After all, it is possible that these neurons are the result of an optimization process, much like what occurs in artificial neural networks. Drawing a parallel in the world of computing, "the different components of artificial neural networks are often very difficult to understand. There is no simple mathematical or verbal description that explains exactly what they do", adds the professor.
For a better understanding of the brain from the AI, one must examine the architecture of the organ as if it were a network structure. This is in addition to the forms of optimization that happen over time, during a person's life.
In one experiment, the neurons that release dopamine - the neurotransmitter known for the sensation of pleasure - in the brain also seem to react with unexpected rewards. It is the case of a person who gets a candy by surprise, during the day. This different type of signal is called reward prediction error and is often used to train artificial neural networks to maximize understanding of the rewards they receive.
For example, when programming an artificial neural network to interpret the points received in a video game, the developer can use reward prediction errors to train the network in how to best play the video game. Including, training for suppressed and random rewards - after all, the unexpected is also part of the human experience.
"In the real brain, as in artificial neural networks, even if we don't understand what each individual signal means [as in the case of the surprise boost], we can understand the role of these neurons and the neurons that receive their signals in relation to the learning goal of maximize rewards, "says Richards.
Given the ability of artificial neural networks to solve complex problems, it can, yes, help in the process of discovering the secrets kept in the human mind.