A growing field of research known as organoid intelligence is trying to reproduce the human brain to capture AI and more.
As generative artificial intelligence (AI) research continues to spread rapidly, some scientists around the world are already working on the next big thing: a field that envisions computers with real brains, known as biocomputing. is known in
Advertisement
Current AI models use networks of a few hundred million neurons with extremely simplified neurons, and they require significant amounts of energy.
Meanwhile a human brain uses very little power to make connections between its approximately 90 billion neurons.
According to experts, if existing artificial intelligence companies want to replicate the number of connections in the human brain, they will need a nuclear power plant.
This is because generative AI models are synthetic and require neurons to be powered by electricity in order to communicate with each other.
Biocomputing proposes a fundamental paradigm shift by using real, biological neurons.
“We are at the beginning of a revolution,” Dr. Fred Jordan, CEO and co-founder of Final Spark, told Euronews Next.
In 2014, he and his colleague Dr. Martin Kutter formed one of the world’s first biocomputing companies. Today it is one of three corporations working in this area, along with Cortical Labs in Australia and Koniku in the US.
‘Building a thinking computer’
Biocomputers are machines using living neurons that can reason like humans and form ideas beyond their experience. They differ from AI programs such as ChatGPT, which can only provide answers from the knowledge it holds in its database.
“Since I was a teenager, my dream has been to build a thinking computer,” said Jordan, who three years ago decided that artificial intelligence and neuroscience — “areas that don’t usually mix” — A combination of is the way to reach that goal.
“The processing of information by the brain is incredibly complex, and today’s digital computers are not well suited for this task,” he said, “so we thought, since hardware is not sufficient, let’s replace it with living neurons or wetware.” “.
Jordan and his team work with neurons obtained through a method developed 15 years ago that transforms human skin cells into stem cells and then into neurons.
Advertisement
But so far no one has managed to build a biocomputer that passes the Turing test, which evaluates whether a machine is intelligent and can fool a user into thinking it is human.
How far is biocomputing research ahead?
Final Spark works with thousands of neurospheres (3D structures of living neurons that are biocomputer prototypes, with fewer neurons and stability) with 10,000 neurons surviving for 100 days – a time period during which Jordan and his team try to understand How to train those neurons.
The aim is to stimulate the neurons via electrodes to make the neurosphere achieve ‘useful functions’, such as learning and remembering information (this is also called neuroplasticity).
But this is no easy feat because each neurosphere is different.
For now, Final Spark’s neural networks can only store 1 bit of information – “like a quantum computer from 15 years ago”. The company’s biggest competitor recently made headlines for teaching living neurons to play pong.
Advertisement
So while biocomputing isn’t spreading across the globe just yet, Jordan expects research to accelerate.
“All of our work is open data, because we believe the biggest risk is not our competition, but not finding the right solution for biocomputing,” he said.
In the coming months, Final Spark will be working with universities around the world to allow students to remotely conduct their own electrode stimulation tests and attempt to contribute to research on neuroplasticity.
“I’m hopeful that next year we’ll be able to master some aspects of learning,” Jordan said.
“At the moment, we are moving in interesting and innovative directions”.
Advertisement
What can biocomputing do?
The most obvious use of biocomputing at the moment is to replace the synthetic processors used by AI companies to reduce energy consumption “1 million to 10 billion times,” Jordan said, citing data from Johns Hopkins professor Thomas Hartung. , who are working on it. Biocomputing with a community of scientists of which Final Spark is a member.
AI enterprises need to scale up their processors for each new model, and their carbon footprint must follow. On the other hand, neurons and biocomputers can be easily multiplied, and the AI field can be siphoned off for much of its emanation.
Jordan is already in contact with dozens of companies in the tech industry.
“Some people understand what we are trying to achieve, but most do not. What we’re doing sounds like science fiction to them,” he explained.
Nonetheless, Frontiers, one of the world’s most cited research journals, recently launched an “Organoid Intelligence” section.
“That recognition was really important to me, because there really wasn’t anything in the research that acknowledged our activity,” Jordan said.
Beyond reducing the energy consumption of some AI ventures, what biocomputing would be able to do is “unimaginable”, he said, “because neurons are self-programming”.
“We don’t know exactly what biocomputers will be able to do”.
Could they have taken over humanity then?
Jordan said, “Cars drive faster than humans and computers calculate faster than humans, but none of these have taken over humans.”
