Swiss technology firm FinalSpark recently launched a program that allows scientists to rent cloud access to “biocomputers” made out of human brain cells for a mere $500 a month.
The purpose of these biocomputers, according to FinalSpark, is to develop a highly efficient, low-energy solution to the ballooning costs associated with developing artificial intelligence models. The company says it could be as much as 100,000 times more efficient to use computers made of organic material to train AI than it is to use traditional silicon-based technology.
According to FinalSpark founder Fred Jordan, Neurospark is the only company to offer access to computers made out of clumps of human brian cells (called organoids). Their technology can be viewed live online.
Training brain cells
Traditional methods for training AI models often included a technique called “reinforcement learning” where an AI model is given a reward for completing a task. Typically this so-called reward is merely a numerical value used to help calibrate how close the behavior was to the intended outcome.
However, in order to provide positive and negative reinforcement when training AI models made of the neurons inside organoids, scientists have to use techniques that work on brain cells. This involves feeding neurons dopamine — a molecule associated with reward — as positive reinforcement and an electrical signal as negative reinforcement.
Temporary living AI
FinalSpark says its organoids can “live” for up to 100 days. Theoretically, this means an AI system could be trained and operated for a few months before it would either have to be shifted to another cluster or face being lost forever.
At longer scales, perhaps it could become possible to create AI systems with lifespans similar to our own as a method for aligning the artificial intelligence experience with humanity’s.
It’s also hypothetically possible that, one day, we could build blockchains running on computers made of organic material.
Related: Coinbase CEO Brian Armstrong: AI ‘should have crypto wallets’