The Spaniard in Korea who 'copies' the brain to improve artificial intelligence

Spread the love

The Spanish in Korea who 'copy' the brain to improve artificial intelligence

Suddenly… chachas! One sound and your brain kicks in: identify a snap, close behind the head, possibly fingers snapping together. It parses everything practically instantly, without having to set bits, ones and zeros or anything like that. How? That's what you're doing. studying Miguel Sánchez-Valpuesta, a 33-year-old Spanish man who works at the Korean Brain Research Institute (KBRI) with the aim of applying brain processes towards improving the architecture of artificial intelligence (AI). ) and make it faster, more efficient and sustainable by 'copying', as far as possible, the innovation that the human being hides inside his head.

“The goal is not to copy nature as it is, but to learn from something that is already beautiful and try to develop new things. In general we know little about how our brains work, or even how we think, but it is fascinating to try to discover it. How, only with the vibration of two membranes in our ears, can we locate where something we are not seeing or touching is in three dimensions? Through calculations and predictions that occur instantaneously in our heads and that they are the result of infinitely more advanced processing than current computers”, explains the young man from Barcelona to El Periódico de España -a newspaper that belongs to the same publishing group as El Periódico de Catalunya- from the South Korean city of Daegu, where it has been working for two years.

The challenge for those who, like him, intend to discover the secrets of the brain, is precisely to understand how procedures developed over millions of years of evolution are articulated and how they can be applied to improve the current technology.

During the next few years, he explains, various models of artificial intelligence will appear, from those that will bet on a traditional computer system to those that they will aspire to copy the model through so-called neuromorphic computing.

Miguel, who is one of the international experts at the Hermes Institute and has a master's degree in biomedicine from the University of Barcelona, ​​did a doctorate for six years in Japan to study the mechanisms and neural circuits of language learning. ;stico. From there He made the leap to Korea, where this study of the way in which the auditory circuits interact with the motor brain could improve an artificial intelligence that has been uncovered in 2022 as one of the technological advances that will mark the next decade.

“Just as we see with the brain, since the image it presents to us is not the one our eyes receive, we also listen with it. The brain 'invents' itself and 'fills in' , so to speak, much of what we perceiveIt is based on predictions and previous experience, and continually maintains a representation of the external world that is selectively 'updated' by sensory information. That is why there are also pathologies of phantom sounds that the ear is not aware of. perceiving but what if which are in our head. This type of knowledge allows us to understand how sound travels through our neurons, and, therefore, a little more about how that processing is”, explains Miguel.

The human brain, however, is not limited to 'filling in', it has extraordinary adaptability when it comes to sound. For example, if we receive a sound at 60 decibels instead of 30, it sounds a little louder in our head, but, as the young man from Barcelona explains, in reality it should sound 1,000 times louder. be strong. “The same thing happens with our own voice or when we exercise. You don't hear the sounds you make as loud because your brain muffles them, it's an involuntary survival mechanism,” he says.

If these processes could be synthesized, for example, the ability of autonomous vehicles to adapt to new situations and unexpected stimuli could be improved. It is just an example, because they have already improved, but before, a car could be prepared to identify a child, an adult or a dog and not run over it, but, what about that? What would happen if a wild boar came across? or a skateboard? The AI ​​needed to be able to process that new item it was not prepared for and  judge its relevance instantly, without time for traditional processing.

“Until now artificial intelligence works, so to speak, with classifiers. For A, B or C it gives more or less standardized answers, and that was what we thought was happening with the brain, but it is not true, most of what he does is not based on stimulus and response.80% of brain activity, for example, consists of maintaining an internal representation of what is happening around us,” he notes.


However, even if it were possible to discover the way to emulate all this through a conventional computer, it would be practically impossible and impractical. “Even supercomputers perform algorithmic operations step by step, which slows down processes and wastes an enormous amount of energy. Quantum computers, on the other hand, don't do algorithmic operations, and neither does our brain, although they have other limitations today”, explains the engineer.

In this way, he says, right now computing is carried out “little by little”, that is, “even when we simulate neural circuits, we do it as if it were software on a CPU following algorithmic rules that are completely foreign to the computer.” functioning of the brain”. This is where neuromorphic computing comes in, a field that aspires to replicate, both in chips and in processes, the way in which human thought is articulated. p>

Right now, only South Korea, where Miguel works, and Taiwan are capable of manufacturing the most advanced chips in this field, with what that implies for the race for technological development. This is a logical link between America, Europe and Asia.

In addition, it is estimated that today 3% of all the electricity used in the world is consumed in data centers, and In 2030, this percentage could reach 13%. Optimizing computational models, as the brain has done over millions of years of evolution, would make artificial intelligence a much more sustainable technology.