News World's first bioprocessor uses 16 human brain organoids for ‘a million times less power’ consumption than a digital chip

[helper800]

I am one of the co-founder of this startup, and I will be happy to give some insights.


Indeed, Steve Potter was absolutely visionary with his work and is actually now one of our advisors. However, as @bit_user mentionned, we are not targetting any robotic application at the moment.

I fully agree to this.

Indeed, when one to look at it is if we consider that our brain consumes the equivalent of 20W while a digital simulation of 100 billion neurons will be several order of magnitude higher. Some projections give 1'000'000 times more energy efficiency for biological systems.
Now, we feel that we are a bit like quantum computers 20 years ago, but believe we will have much faster progresses: after all, we are using a proven technology

Actually the organoids are kept at 37C, we have thousands of them stored like this, waiting to go on the electrodes.

At the beginning we indeed had a lifespan of a few hours... currently we have about 3 months.

What sort of timelines are you guys expecting to have a reasonable amount of usable performance, and then scaling production for a product? It can be very vague, just wondering what amount of time in magnitudes it would be to see some sort of product. 1, 3, 10, 25+ years?

 

[FredJordan]

What sort of timelines are you guys expecting to have a reasonable amount of usable performance, and then scaling production for a product? It can be very veg, just wondering what amount of time is magnitudes it would be to see some sort of product. 1, 3, 10, 25+ years?

If we stay as we are (6 people), I would say 5 years to start having some stable prototypes doing some basic tasks. But if we raise money, progresses could be way faster...

 

[TerryLaze]

Actually the organoids are kept at 37C, we have thousands of them stored like this, waiting to go on the electrodes.

So you actually have to keep them warm instead of cooling, still an added cost I guess but shouldn't be much.

 

[fireaza]

Though I thoroughly enjoyed the original movie, it was written by people with a poor grasp on science. In their vision, the machines were using humans as a biological power source and the point of The Matrix was basically to keep us from dying of boredom... or something like that. For such a cool movie, you'd think they could've done a little more work on the premise, but times were different and most movies made very little effort to getting the basics of science & technology right. Anyway, I somehow managed to watch the movie without even picking up on that detail, so it didn't ruin it for me.

From what I've heard, the original concept for the movie was the machines would use the brains of the captive humans as a sorta distributed computation network, but the studio vetoed it because it would be too complicated for 90s era movie-goers to understand.

 

[Alicepizza]

I am one of the co-founder of this startup, and I will be happy to give some insights.


Indeed, Steve Potter was absolutely visionary with his work and is actually now one of our advisors. However, as @bit_user mentionned, we are not targetting any robotic application at the moment.

I fully agree to this.

Indeed, when one to look at it is if we consider that our brain consumes the equivalent of 20W while a digital simulation of 100 billion neurons will be several order of magnitude higher. Some projections give 1'000'000 times more energy efficiency for biological systems.
Now, we feel that we are a bit like quantum computers 20 years ago, but believe we will have much faster progresses: after all, we are using a proven technology

Actually the organoids are kept at 37C, we have thousands of them stored like this, waiting to go on the electrodes.

At the begining we indeed had a lifespan of a few hours... currently we have about 3 months.

How about if you design those organoid electrode like a hotswap dish which similar to computer server RAID harddrive (e.g.RAID 5) ? After that, you may insert multiple eletrode dishes running at the same time and keep the whole AI system running for a longer period?

Then, lifespan is no longer a very big issue before you can find a new way to make them run for years?

 

[bit_user]

How about if you design those organoid electrode like a hotswap dish which similar to computer server RAID harddrive (e.g.RAID 5) ? After that, you may insert multiple eletrode dishes running at the same time and keep the whole AI system running for a longer period?

Then, lifespan is no longer a very big issue before you can find a new way to make them run for years?

Unless someone has cracked the problem of how to duplicate the training weights in "wet" neural networks, they're not hot-swappable!

That's one of the key advantages that artificial neural networks have over biology. In a wet brain, everything you know is locked up in your brain cells. In a digital equivalent, the weights can be copied and multiple clones can be created. This underlies the capability of training once and deploying at scale.

 

[Alicepizza]

Unless someone has cracked the problem of how to duplicate the training weights in "wet" neural networks, they're not hot-swappable!

That's one of the key advantages that artificial neural networks have over biology. In a wet brain, everything you know is locked up in your brain cells. In a digital equivalent, the weights can be copied and multiple clones can be created. This underlies the capability of training once and deploying at scale.

Um..... from the experiment.. I can see that they are using something like headstage MEA digital board to connect the brain eletrode to the Intan controller for signalling transfer.

How about if they give the headstage MEA vendor some advice... so they can create a specific model headstage MEA digital board that can connect to multiple wet brain sample? Will this solve the limitation for training weight?

Since the wet brain eletrode are connecting to the same digital MEA board in this way.. I guess it will solve the problem?

 

[FredJordan]

Um..... from the experiment.. I can see that they are using something like headstage MEA digital board to connect the brain eletrode to the Intan controller for signalling transfer.

How about if they give the headstage MEA vendor some advice... so they can create a specific model headstage MEA digital board that can connect to multiple wet brain sample? Will this solve the limitation for training weight?

Since the wet brain eletrode are connecting to the same digital MEA board in this way.. I guess it will solve the problem?

The problem is not to connect to several organoids (we already have 16), but to duplicate the synaptic weights. Consider about 10'000 synaptic connections per neuron.