2023-11-21 18:18:25
November 21, 2023 Today at 7:18 p.m.
Use a layer of biological neurons to exploit artificial intelligence. The idea seems to come out of a science fiction film. However, it is being seriously studied at FinalSpark, a Swiss company.
Thanks to Sam Altman and all his (ex-)friends in the industry, artificial intelligence is slowly beginning to reveal its incredible potential. But if evolution is a dream, it still has more than one aspect to improve. Particularly in terms of its energy consumption. Unsurprisingly, when trying to artificially approximate the skills of a brain, it requires a nice amount of energy. If an AI one day manages to achieve the performance of a human brain, it would need the equivalent of a small nuclear power plant to run. For its part, our brain is satisfied with around twenty watts.
For three years, FinalSpark, a Swiss company, decided to seriouslytackle the problem of this energy consumption particularly important. The company is interested in organoid intelligence (IO), an innovation that might well be the future of artificial intelligence (AI). The basic idea is the same. IO, like AI, relies on a neural network. Via a long learning process, it should be able to interact effectively with ever more complex human requests.
5.000
Neurones
Currently, FinalSpark has managed to create organoids half a millimeter long composed of 5,000 to 10,000 neurons.
The Swiss start-up, however, wishes to replace, at least partially, the artificial neurons currently used with real biological neurons. By (partially) reproducing the structure of a human brain, the technology will benefit from its incredible energy efficiency. “We can hope to need a million times less energy than for AI”, immediately explains Doctor Fred Jordan, one of the co-founders of FinalSpark.
Today, the company’s research has already made it possible to obtain initial advances. “Currently, we have managed to create brain organoids with human neurons and make them interact via electrodes. (They don’t talk regarding brains, because neurons have no consciousness, Editor’s note.) Our networks are half a millimeter with a structure made up of 5,000 to 10,000 neurons”, specifies Doctor Martin Kutter, the other co-founder of FinalSpark.
These were generated via IPS cells, these human stem cells created from scratch from other skin or blood cells. Discovered in 2012, this protocol earned its initiator the Nobel Prize in Medicine. “This thus avoids all the ethical questions linked to the use of primary stem cells from the fetus or umbilical cord,” slips Doctor Martin Kutter.
The neuron lifespan challenge
When working with biological neurons, the first challenge for FinalSpark was to keep them alive long enough. “At first, their lifespan did not exceed a few hours. We now manage to keep them alive for several months, which is sufficient. The idea now is to make them work in a network and carry out “backups” between each organoid, so as not to start from scratch each time an organoid dies,” explains Fred Jordan.
“We are at the same level of knowledge as we knew regarding quantum computers fifteen years ago.”
Doctor Fred Jordan
Co-founder of FinalSpark
With the framework more or less established, the challenge now is to embark on the very long process of learning intelligence but also to increase the size of the organoids. Today, their IO is capable of storing the equivalent of 1 bit of information. In other words, almost nothing. “At the research level, we are therefore still at the fundamental aspect. We are at the same level of knowledge as what we knew regarding quantum computers fifteen years ago,” confides Fred Jordan.
The challenge of expanding their biological neural network will not be simple. “Once cultured, the neurons interact directly. But beyond a certain number of neurons, vascularization of the system is necessary, as in a human brain. Without this, the neurons located in the center necrotize too quickly,” explains Dr. Martin Kutter.
Read also
Altman and OpenAI, a mess with stakes
Working with a living aspect is necessarily particularly complex. “At any time, our organoid can die without us understanding why. We are also dependent on viruses and bacteria, because our network has no immunity.” With such fragile technology, there is little chance of seeing it one day in all smartphones. “But it’s not really a problem. Today, a good part of what telephones offer is handled remotely.”
Create the cake before sharing it
Before arriving at a first concrete offer, research will still take years. To the point that they are still very few people are interested in IO. Currently, only two other companies in the world are looking into the subject. “For the moment, the market is not a cake that we share but that we prepare,” smiles Fred Jordan.
No AI giant has started to study the question. “Almost no patents have yet been filed,” confirms one of the co-founders. “This is the difficulty of fundamental research. As we cannot expect income for several years, investors are still very few in number.”
“The ambition is to raise up to 50 million euros with a minimum of 10 million euros.”
Doctor Fred Jordan
Co-founder of FinalSpark
Until now, FinalSpark has been able to count on a budget of approximately two million euros, directly contributed by its two founders. “In parallel with this project, we are the owners of AlpVision, a company specializing in automatic counterfeit detection, via smartphone software. As the company works well, it has allowed us to finance FinalSpark so far,” explains Martin Kutter.
To continue the adventure, however, the company needs to move up a gear. Its two founders are working on a first fundraising round. “The ambition is to obtain up to 50 million euros with a minimum of 10 million euros. There are six of us currently on the project. The idea is to quickly reach around twenty researchers to make significant progress”, say the entrepreneurs. They are giving themselves one year to complete the roundtable. “We are thinking of investors in deep tech and institutions specializing in fundamental research. The Gates Foundation, for example,” says Fred Jordan.
1700592326
#future #human #neurons