What if computers became organic?

A tangle of wires, printed circuits, transistors… and full cell clusters! These could be the computers of the future: machines where information flows through the neural network of a structure made up of living cells. The first prototypes of this computer 2.0 are already there, notably Brainoware, developed by researchers at Indiana University in Bloomington (United States), one of the first of its kind. And according to the results published in late 2023 by the team, this organic machine not only works, but, even more, it proves to be effective.

Comparable functions

Why try to combine computer science and biology? First, because it seems likely: their function is comparable. “We can think of the cell as having computational capacity: it is a closed object that stores information and can react to the signals it receives from its environment.”, describes Guillaume Gines, CNRS researcher at the Ecole Supérieure des Physiques et de Chemis in Paris. An object with memory and that reacts to signals… The computer analogy works. The binary calculations of machines, sequences of 1s and 0s, simply take the form of chemical reactions between cells in living things.

However, this is not the only reason that motivates this search for the organic computer: “classic” computer hardware imposes certain limits, especially when it comes to working in artificial intelligence. “In a computer, calculations are done sequentially, that is, one after the other. Brain function is highly parallelised.' explains Guillaume Gines.

As a result, when we compare the efficiency of a human brain to that of a supercomputer, the differences are abysmal. A supercomputer would require 10 MW (megawatts) of power to achieve performance comparable to that of our royal organ, which requires only 20 W to operate. However, this surplus of IT energy comes at a cost: when a team develops a new language model – like the GPT-4 on which ChatGPT is based – it has to come up with millions of euros.

A hybrid machine

Under these conditions, imitating the efficiency of the human brain necessarily appeals to scientists and manufacturers… Hence this idea at the beginning of Brainoware: replacing traditional silicon chips with a brain organoid, a cultured cellular structure in laboratory from stem cells and organized to mimic certain brain functions.

Of course, the researchers did not manage to free themselves completely from our good old computers: Brainoware is a hybrid machine, on the border between classical and biological. Because before we can pass information through an organoid, hoping it integrates and analyzes it, we still have to figure out how to transmit it to it – quite a challenge! The scientists then decided to convert the computer data—a recording, for example—into a series of electrical pulses, which they then delivered to the organoid via electrodes. (see diagram).

And after ? “The stimulus is propagated between certain cells according to its intensity. Thus, the signal sent will be transformed”, describes Sarah Berkemer, researcher at the Computer Science Laboratory of the École Polytechnique (LIX). In short, the cluster of cells receives the information and processes it.

How, in what way, for what purpose… No one can say for now: the organoid behaves like a real black box. Still, within it, the signal passes and is modified. The response is just waiting to be captured and then decoded.

To collect this output signal – or “production” -, the researchers once again used electrodes to translate, this time, the electrical impulses into computer data. They then provided them to machine learning software responsible for deciphering them.

Well, the information is transmitted, processed, recaptured… The device is ready. But to what? It could actually be used for many tasks: researchers, for example, have tested it in an exercise to distinguish human voices. Recordings of vowels spoken by eight Japanese speakers were first presented to the organoid, which responded differently depending on who was speaking. Complete proposals were then submitted to him. By comparing the signals received by the cluster of cells, the researchers then tried to determine who the voice belonged to. Conclusion ? The organic computer was correct 78% of the time – a result certainly lower than the performance of current algorithms, but encouraging, especially since the machine seems to have improved over the exercises.

This is perhaps the most interesting aspect of Brainoware: it evolved naturally while doing the work it was responsible for! “The designers showed that this mini-brain was reacting to input signals, that it was an actor. However, they also noted certain characteristics of plasticityreveals Thomas Hartung, a professor at Johns Hopkins University, in the United States, and an expert on organoids. This is the first time that a brain organoid has replicated so many functions and architectural elements of the brain. It has all the necessary machinery for long-term learning.”

Cells that need to be fed

In fact, some connections between cells in the cluster appear to be strengthened, others weakened, and new ones formed – just like in a learning brain. Brainoware could thus improve further over time and perhaps even learn new tasks thanks to its ability to reorganize when current, highly specialized computer models often cannot.

Should we then expect the emergence of an organic computer more powerful than any computer and more practical to build and use than quantum ones? There is still a way to go… Various technical limitations remain, esp. Unlike a silicon chip, the organelle's cells are alive: they must be constantly supplied with oxygen and nutrients. However, this throughput increases with the size of the organoid, which currently prevents the largest of them – and certainly the most powerful for that matter – from exceeding one millimeter… Receiving the output signal is also a challenge: information is lost . “These engineering challenges are not insurmountable, however, everything takes its course”assures Thomas Hartung.

An unknown factor remains: from an ethical point of view, what should we think of these works? Could such a cellular aggregate feel pain in the face of electrical impulses from electrodes, boredom in performing the same repetitive task, and why not even fall prey to bursts of consciousness? “States should look into the matter to enact laws, says Sarah Berkemer. But that will take time. ” Who knows, maybe an organic computer will one day be able to answer these questions – that would be a shame.

Leave a Reply

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