Skip links
Published on: Press

Doing a lot with little

There is no need to introduce Charles Darwin, biologist and father of the theory of evolution. He argues that living things can adapt to their environment throughout their lives and pass on these adaptations to their descendants. He adds that those better suited to their environment pass on their characteristics better than those less suited to their environment.

Born in the context of biology, this theory can also be seen from the perspective of information processing. Let’s put it another way: is evolution a form of learning? If so, what information is learned and transmitted?

From the first cells to today’s living beings, evolution has produced a multitude of organized forms of life. And naturalists work to classify themselves as genealogists of life. Thus, millions of years of evolution have produced organisms adapted to their ecological niche: blue whales, finches, moles, crabs, platypuses, tardigrades or sloths…

Under the umbrella of information processing, the product of evolution is a set of condensed experiments that function in each environment. What does not work disappears, and what no longer works makes paleontologists happy. The mark of success is staying alive. Simply. But staying alive is not accurate enough. The purpose of life is to be transmitted.  Therefore, natural selection does not operate on individuals but on what they transmit. What we call genes. Richard Dawkins, who defends this position, said “we are robots blindly programmed to preserve the selfish molecules known as genes.” (Dawkins, 2003)

The magnificent trees of evolution can therefore be replaced by genetic trees that are certainly much less beautiful to look at but probably more at the heart of the phenomenon of life.

But are we really sure? Genes are not immortal compounds. They are physical supports that code information. In humans, our 23 pairs of chromosomes contain 3 billion pairs of letters A, T, G, and C. How much information does that represent? Following Stanislas Dehaene, each letter is a code for 2 bits as four possible letters and one human genome is reduced to 750 MB of information

The human plan therefore holds in a CD.

However, it is the content of this CD that is passed down from generation to generation. Genes are only the medium for this information, for these messages transmitted. That is what survives. Natural selection is therefore a mechanism that promotes the transmission of messages adapted to a given environment and at any given time.

This original reading is proposed to us by the authors of the Thread of Life (Dessalles, Gaucherel, Gouyon, 2016) who summarize it with a lapidary formula: information would be the real “thread of life”.

Since the dawn of time, the living has made a myriad of attempts to convey their messages. The algorithm of natural selection without any intention, keeps what works and keeps out what does not. The messages are therefore traces of the past that are not mere bottles discarded over time, containing old withered parchment. The messages contain an algorithm.

The 750MB CD of stored information brings together everything life has learned during its selection process.

But what in fact did it learn? Is it the finished plan of the different organisms where all you would need is so follow it to the letter? If this was the case, genes would carry a message producing mature organisms, ready for use. We all know that this is not the case. All living organisms are born in the making. From the earthworm to the hippopotamus, all are born with the ability to learn. They are not pre-wired.

Natural selection is perfect for adapting to an ecological niche but the algorithm is slow. Terribly slow. This learning can take hundreds of years. Though it is sometimes much faster, such as for the birch moth which adapted to coal pollution in the 1850s in England. This butterfly changed its color from white to black… then from black to white when the coal mines disappeared. But generations must disappear for learning to succeed.

Life has therefore found the key to a good compromise. What is stable and regular is inscribed in the body:  the upper, the lower by gravity, the heat by radiation, the sounds by different pressures. No need to create eyes to see or arms to move. This is done by the learning of the species. This is what we call innate.

The rest must be acquired. What do we see in the environment with our eyes? Horizontals, verticals, what colors, what speeds? How do I communicate with my fellow people? How do I integrate into my society and acquire its rules?

And now the big question of learning comes to the heart of our thinking. The CD of life does not transmit a definitive plan of the organism. It transmits the plan of a system that is pre-equipped to develop in its environment, but which will have to learn to flourish by itself. The CD therefore contains an algorithm capable of building a system that will learn: a meta-algorithm.

In our previous articles, we dissected the self-organization that maneuvers the cerebral cortex and discovered that it shapes perception from stimulation alone. This self-organization stems from the structure of the cortical columns, which are themselves the fruit of evolution. These cortical columns are pre-wired but they will subtly modify themselves to bring out forms of knowledge: recognize objects, count them, recognize faces or smells. Knowledge that is not taught by language or transmitted by parents or teachers.

This idea of self-organization is at the heart of our approach and completely out of step with the current of deep-learning. A child clearly does not need millions of data to recognize a cat or its mother’s smell. A few spoonfuls of vegetables and a few examples would suffice. Energetic frugality and data parsimony.

Our first realization, InspireMe, is in accordance with these principles, which we will soon present to you to convince you that we can do a lot with little.

References

Dawkins, R. (2003). The selfish gene. Paris: Odile Jacob.

Dehaene, S. (2018). Learn! The talents of the brain, the challenge of machines. Paris: Odile Jacob.

Dessalles, J.-L., Gaucherel, C., And Gouyon, P.-H. (2016). The Thread of Life. The immaterial face of the living. Paris: Odile Jacob.