⁇ [VIDÉO] You may also like this partner content (after the ad)
Bacteria play tag, it seems impossible. However, scientists have made this possible by modifying the bacteria Escherichia coli. Thus they are “tuned” to act as electronic components and act as a neural network.Artificial intelligence Can learn.
“ We propose an evolutionary strategy to design genetic circuits capable of autonomously learning to make decisions in complex contexts. ”, Summarize the researchers in a single sentence in their pre-published research report bioRxiv. For now, the questionable “decision” lies in making the right choice in a game: which box? This test is not so strange as it is often used to test artificial intelligence systems.
It all started in 2019 with the work of another research team. The strain of a particular bacterium, Escherichia coli, is genetically modified to sense 12 different chemicals and reacts accordingly by altering the function of certain genes. This strain got the nickname “toy”.
The research team took this work and applied other modifications. They combined several copies of two circular pieces of DNA called “plasmids”. Each of these plasmids “codes” a different fluorescent protein. One is red, the other is green. The ratio of these two plasmids in bacteria is therefore not determined in advance by their final color: it is affected by 12 different chemicals, but also by some antibiotics. If no such change is applied, the ratio will remain the same. In a sense, the composition of DNA creates a kind of “memory” that stays frozen until the next change.
Conversely, when a new “data” is entered, with chemicals or antibiotics, the ratio (hence the color) is formed, but the previous configuration is taken into account. This means that some memory is retained, so some kind of learning is possible. This behavior is similar to that of electronic components called “memory”. Hence this new biological “component” is named “memregulons” by scientists.
A memory, in fact, is a transistor-like component that can not only process but also store data in its internal memory. It is a component used specifically to create so-called “neural networks”. In fact, in many cases today, the so-called “artificial intelligence” is a system “fed” with a large amount of data to extract “learned” and logical connections for a given purpose. Recognize faces, texts … or play tic tac toe … These learning methods are inspired by the activity of biological neurons. Today, they are closer to statistical methods. So we’re talking about the “artificial neural network”. The transmitted data is transmitted in an artificial “phase” of neurons, usually virtual. They are actually points in the network, connected together by computer code. So this network receives incoming information, learning data and outgoing information.
From the moment the bacteria are able to pick up the incoming data (chemical product or antibiotic), it can send “results” (DNA and hence changing color) while retaining previous data in memory so that one can understand the comparison made. In the same way, scientists created a “neural network” with these bacteria.
The result was a system that could learn from its past behavior and modify itself by rearranging itself – like a learning AI, in a nutshell. In the example of crab lice, here is how the scientists worked: just as they “trained” the bacteria, they used the learning through “reinforcement,” that is, the reward and punishment system depending on whether the action was taken. Adequate strategy or not.
These bacteria were grown in boxes associated with a crab stage. Every time a man plays, the bacterium is “reported” by the addition of a chemical to the box, each product being associated with a place. First, if the bacterium “played” its next move roughly, it eventually learned by “punishment” through the method of reinforcement, by adding antibiotics when game activity was bad. However, it was not so easy: it took several days to play a game.
The demonstration is about a crab, which is harmless. However, according to scientists, the possibilities for their research are far-reaching. ” Modified genetic circuits can give cells decision-making abilities “, They say. They can” rearrange “themselves in more complex decision-making situations.