‘Biocomputers’ made of human brain cells now available on rent

FinalSpark, a Swiss technology firm, recently launched a program that allows scientists to rent cloud access to “biocomputers”.

These biocomputers are made of human brain cells and cost $500 a month.

The company claims that its biocomputers or organoids (a technology that offers low-energy alternatives to AI models) can last up to 100 days, which is indeed short. However, AI systems need to be trained on organoids, which might need to be periodically transferred to new clusters or face degradation.

Training the organoids is a complex procedure, which involves two setups. The first one is positive dopamine rewards and negative electrical stimulation. By using these methods organoids have formed new pathways, just like a living human brain would. As of now, these organoids are being watched carefully, so that they can also operate better.

Limited access granted to FinalSpark’s biocomputers for top universities

As of now, around 34 universities have asked for FinalSpark’s biocomputers to be used. It (the company) has, however, asked that only 9 institutions get to use this. This includes the University of Michigan, the Free University of Berlin, and the Lancaster University in Germany.

As of the latest reports, the University of Michigan is likely to carry on with its research on the electrical and chemical prompts. They are most likely to carry out research on chemical and electrical prompts which would help alter organoid activity. Along with that, the researchers are also looking forward to fitting the organoids into various models of AI learning.

They are also working to help them run on a larger scale and lay a hand at silicon-based AI systems.

Previously, FinalSpark had made the news for launching Neuroplatform, which is the world’s first bioprocessing platform. In this case, the human brain organoids specialize in computational tasks instead of silicon chips.

According to the company, Neuroplatform could process around 16 brain organoids. FinalSparks believes this is a lot less power being used as compared to their other counterparts.

Exploring new techniques for training AI models with brain-cell biocomputers

Training AI models is a tricky task, the traditional AI models would include something called “reinforcement learning”. This means that an AI model is being given a reward for completing a task.

In most cases, the reward would mostly be a numeric value that can be used to gauge how close the model’s behavior was. This would be in comparison with the intended outcome. Even though this technique would work in most cases, and whatever the effect — positive or negative, scientists discover techniques that work on brain cells.

One of the ways to do this is by feeding neurons dopamine , a molecule with a reward, a positive reinforcement. And, an electric signal which qualifies as a negative reinforcement.

As of now, the Swedish organization is intensifying its neuro platform facility, and generating around 2000 to 3000 biocomputers. However, despite all the innovation, it might have attracted some competition for itself.

Some researchers from Spain are working on new cellular computers which can respond to environmental stimuli. Other researchers from England are working on advanced research on fungal networks to be used for reservoir computing and patterns.