AeroSpace News Pharma Research Solutions

What does AI gain from being inspired by the nervous system?

AI Inspired by the nervous system

If I tell you: vision, touch, smell, taste and hearing, you know immediately that I am talking about your 5 senses. But if I ask you to tell me about a sixth sense, you would probably think of a hidden, perhaps esoteric sense. And yet we all have a sixth sense: proprioception. It is our ability to locate every part of our body. It allows us to feel our own body. 

Author: François Blayo, chief scientist officer NeoInstinct

SwissCognitive, AI, Artificial Intelligence, Bots, CDO, CIO, CI, Cognitive Computing, Deep Learning, IoT, Machine Learning, NLP, Robot, Virtual reality, learningTo experience it, sit down and just close your eyes in silence. Then, still with your eyes closed, move one arm from left to right. Slowly, then quickly. You undeniably have a feeling and even an “idea” of where your arm is. You can even tell if you are moving it quickly or slowly. You can feel the position of your arm and the speed of movement without seeing it. 

The proprioceptive sense allows the brain to determine, for all our limbs, their position, their speed and their direction. Widely unrecognized, the proprioceptive sense is nevertheless indispensable for the simplest acts of everyday life. There would be no walking, no painting or piano, no sport without an effective proprioception.

This sense relies on sensors capable of measuring the position of a joint and for others the speed of movement of that same joint. It’s not surprising that our brain is able to feel the speed and position of our arm as we have just experienced it. But, in nature, nothing happens as one would expect.

It turns out that this information, speed and position, are not transmitted separately by nerve fibers. They are actually mixed and this mixture is transmitted by nerve fibers to the central nervous system. However, we do not feel a mix of our arm’s speed and position. We feel them both separately. Therefore, there exists a mechanism in the central nervous system capable of separating the two mixed signals.

Why would nature do this? Probably to ensure some redundancy in case of nerve fiber damage. All information would not be lost.

We could therefore imagine that the separation of the signals “speed” and “position” is innate or pre-wired. But this is incompatible with the effects of aging which influence the mixture. 

The other hypothesis is that our central nervous system is able to separate signals without knowing how they were mixed. By learning. Impossible?

This is a situation that is frequently encountered. At a party, with a drink in hand, we converse with a friend while others do the same. Yet we clearly separate the sound signals from our interlocutor from the ambient noise, which is itself made up of other conversations. Several sources, an unknown mix, and a successful separation.

It is actually a very general principle that is at work in the nervous system. Separate independent sources without knowing how they were mixed.

Two researchers, J. Hérault and C. Jutten (Hérault- Jutten, 1986), were interested in the mechanism that allows the nervous system to achieve this feat. By formulating the hypothesis that the signals are independent, they showed that a very simple neural network was perfectly capable of achieving this separation through continuous learning. And “very simply” are 2 neurons to separate a mixture of 2 independent signals.

This extraordinarily fruitful work has defined a concept, analysis into independent components, hundreds of research projects and just as many applications. Examples include medical imaging, intracranial electroencephalography, recording work meetings, recognition in smartphones, transmitting satellite signals or astrophysics, and improving airport guidance.

But just as interesting is the approach taken. In his 1988 publication (Jutten, 1988), Christian Jutten points out that “the existence of this problem of source separation in biological systems, particularly in the nervous system, naturally led us to design a neuromimetic solution (digital imitation of living neurons). It may be objected that the model finally proposed, no longer has much neuromimetic, such as it is simple with respect to biological systems. The neuromimetic aspect is in fact due to a state of mind where the sources of inspiration for the proposed solution is not only mathematical physics, signal processing or other theories. Our approach constantly consults the “meta-source of inspiration” which constitutes natural systems. In particular, this leads to a discussion and questioning of the methods used and the results obtained. However, since our goal is not to model the nervous system, this source of inspiration will never be a constraint: for example, to develop a powerful application dedicated to a particular problem, we will not attach great importance to the neuromimetic aspect. Such an approach is not unique in signal processing: the work on bat sonar systems is similar in approach; the trick and performance of biological solutions no longer needs to be proven.”

This is exactly the state of mind in which we work to develop a frugal and fast that continues the definition attributed to Jean Piaget: “intelligence is not what we know, but what we do when we don’t know.”


About the author:

François Blayo is an Artificial Intelligence pioneer, expert, author, speaker and popularizer, currently leading  research at NeoInstinct.


Works cited

Hérault, J., Jutten, C. (1986). Space or time adaptive signal processing by neural network models. International Conference of Neural Networks for Computing. It’s Snowbird.

Jutten, C. (1988). A digital solution to the problem of source separation. Signal processing, 5(6).

Image Source:
– clipart-library.com

0 Comments

Leave a Reply

Your email address will not be published. Required fields are marked *