Mar 5, 2025 07:02 PM
https://www.abc.net.au/news/science/2025.../104996484
EXCERPTS: Today, Dr Kagan and his team launched the product — called CL1 — at an international technology conference in Barcelona, promising companies access to the "first commercialised biological computer".
Filled with hundreds of thousands of lab-made neurons sized somewhere between an ant and cockroach brain, CL1 is primed to learn. But even Dr Kagan can't imagine exactly what the human neurons might be tasked with.
[...] their ultimate goal, and what they've pushed at this conference, is to use these tiny collections of neurons as a type of biological AI.
The idea behind CL1 is that since companies like Google and OpenAI are trying to create an AI that works like a brain, why not use the parts — neurons — to achieve the same goal? "The only thing that has 'generalised intelligence' ... are biological brains," Dr Kagan said.
[...] there are a few ways neurons intrinsically work that make them useful in certain situation, such as medical research. The first, is power consumption. The current generation of traditional AI models take extreme amounts of power to produce results.
CL1 on the other hand, uses just a few watts of power. "It doesn't all have to be systems that consume huge amounts of power," he said.
The second benefit, according to Dr Kagan, is how fast brains can learn. "What humans, mice, cats and birds can do [that AI can't] is infer from very small amounts of data and then make complex decisions," he said.
CL1 is not much larger than a shoebox. Most of it is built to house and keep neurons alive. Neurons are finicky, and they need to be kept in optimum conditions, including removing waste, feeding nutrients and keeping out unwanted microbes.
But the most important part is the chip — a small silicon device with hundreds of thousands of lab-grown human neurons attached to it and each other. The neurons are made in the lab in a process which turns blood cells back into stem cells — cells that can develop into many different types of cells — which are then turned into neurons.
"[They're] grown from induced stem cells that are generated from a small amount of blood from volunteers," Dr Kagan said. "Same as what a doctor might need for routine tests."
The neurons can be "taught" via small amounts of random or patterned information provided by the chip. The incorrect response receives random information, while the correct one receives patterned data. Eventually, the neurons start to learn what the correct response is. This is how Cortical Labs taught the system, then called "Dishbrain" to play Pong.
It's worth noting that DishBrain was not a pro Pong player, it hit only slightly more balls than it missed. But it was better than a system that had received stimulus but no feedback. Since then, the system has been updated, software and hardware has been created to house the neurons and improve their accuracy... (MORE - missing details)
EXCERPTS: Today, Dr Kagan and his team launched the product — called CL1 — at an international technology conference in Barcelona, promising companies access to the "first commercialised biological computer".
Filled with hundreds of thousands of lab-made neurons sized somewhere between an ant and cockroach brain, CL1 is primed to learn. But even Dr Kagan can't imagine exactly what the human neurons might be tasked with.
[...] their ultimate goal, and what they've pushed at this conference, is to use these tiny collections of neurons as a type of biological AI.
The idea behind CL1 is that since companies like Google and OpenAI are trying to create an AI that works like a brain, why not use the parts — neurons — to achieve the same goal? "The only thing that has 'generalised intelligence' ... are biological brains," Dr Kagan said.
[...] there are a few ways neurons intrinsically work that make them useful in certain situation, such as medical research. The first, is power consumption. The current generation of traditional AI models take extreme amounts of power to produce results.
CL1 on the other hand, uses just a few watts of power. "It doesn't all have to be systems that consume huge amounts of power," he said.
The second benefit, according to Dr Kagan, is how fast brains can learn. "What humans, mice, cats and birds can do [that AI can't] is infer from very small amounts of data and then make complex decisions," he said.
CL1 is not much larger than a shoebox. Most of it is built to house and keep neurons alive. Neurons are finicky, and they need to be kept in optimum conditions, including removing waste, feeding nutrients and keeping out unwanted microbes.
But the most important part is the chip — a small silicon device with hundreds of thousands of lab-grown human neurons attached to it and each other. The neurons are made in the lab in a process which turns blood cells back into stem cells — cells that can develop into many different types of cells — which are then turned into neurons.
"[They're] grown from induced stem cells that are generated from a small amount of blood from volunteers," Dr Kagan said. "Same as what a doctor might need for routine tests."
The neurons can be "taught" via small amounts of random or patterned information provided by the chip. The incorrect response receives random information, while the correct one receives patterned data. Eventually, the neurons start to learn what the correct response is. This is how Cortical Labs taught the system, then called "Dishbrain" to play Pong.
It's worth noting that DishBrain was not a pro Pong player, it hit only slightly more balls than it missed. But it was better than a system that had received stimulus but no feedback. Since then, the system has been updated, software and hardware has been created to house the neurons and improve their accuracy... (MORE - missing details)