A startup is building computer chips using real neurons

https://fortune.com/2020/03/30/startup-human-neurons-computer-chips/

by gautamsivakumar • 6 years ago

160 108 6 years ago

fortune.com

HenryKissinger• 6 years ago

So, a brain?

Easy there, Dr. Frankenstein.

29athrowaway• 6 years ago

If you have neurons then you also need glia and neurotransmitters and ways to regulate them and keep them alive... and suddenly you need the entire organism.

pazimzadeh• 6 years ago

That could be said of almost any ex vivo model right?

whoisjuan• 6 years ago

What’s the goal of this though? I couldn’t determine it from the article, but I’m certain that using them as transistors just for the sake of saving energy doesn’t make sense. Especially since a neuron size is a about 4000nm or more than 500x larger than the new 7nm transistors.

No way energy consumption is justifiable with such a massive loss of chip real estate. You will need way larger chips to achieve a decent computing power.

haydn3• 6 years ago

Maybe it's not just transistors they're hoping to gain from this.

I think it's more like building a brain from organic matter that's the real boon. Sooner or later we'll be discarding silicon and extracting neurons from mice just to watch YouTube IX in space with our Tesla corvette cruisers and our alien barmaids serving us Soylent Green in a martini glass!

mimentum• 6 years ago

Well considering, many humans solved complex problems for thousands of years, with just the human brain for power, at about 20W of power consumption. One could argue that the human brain, or nature is more efficient at processing than that of a computer of equal power.

russdill2• 6 years ago

The best thing that could come out of this is a better understanding of how neurons work, especially across different organism. But yes, for building actual things it wouldn't scale. Silicon is a far better way forward and we have a ton of room for optimization.

monocasa• 6 years ago

Pretty weird not to comment that they're replicating published work that trained mouse neurons on an electrode array to fly a plane in a flight simulator.

https://www.sciencedaily.com/releases/2004/10/041022104658.h...

It's about at the level where you could make a good attempt to replicate it as a hobby in your (albeit well stocked) garage.

mjfl• 6 years ago

why did you link to the science daily and not any actual article? The linked page doesn't even link to an article, it links to a university press release. From 2004 no less. How am I supposed to verify that they are replicating published work as you say?

jcims3• 6 years ago

Having a bad day? Just do a little legwork, there are enough context clues to find the story from the linked article.

jcims3• 6 years ago

It's about at the level where you could make a good attempt to replicate it as a hobby in your (albeit well stocked) garage.

Already being done: https://youtu.be/V2YDApNRK3g

Doesn’t hurt to join forces with another ‘well stocked garage‘ featured earlier on HN: https://youtu.be/V2YDApNRK3g

NeuronNerd• 6 years ago

Once your garage is that 'well stocked' you might as well call a lab! I think you might be drastically underestimating the difficulties inherent in both cell culture and programming in getting this sort of thing to work.

paulrouget• 6 years ago

That video was fascinating…

Melting_Harps• 6 years ago

Very cool, I added this to my 'biohacker' list of links.

And lol @ your Egg Benedict Donburi from Food Wars. I still haven't started season 2... check out Dotchi No Ryouri, it used to run on PBS but can be found on youtube.

jcims3• 6 years ago

Oops, second vid should be from Ben Krasnow’s (bkraz) Applied Science channel: https://youtu.be/YAPt_DcWAvw

Ben explains somewhere in there that this was done to help create a better substrate for the first experiment.

dsign• 6 years ago

I think that getting any kind of practical processing from biological neurons from a mouse or a human is ... well, hard and pointless.

However, the idea of getting biological parts interfacing with silico has lots of applications. For example, it could be used to build cheaper DNA printers where the "printing heads" are genetically modified bacteria. And from there, it would be possible to arrange all sorts of chemical processes using bacterial metabolism. One could for example build sealed bio-batteries with an infinite lifetime, whose DNA is kept mutation-free using a digital master in the embedded digital controller.

leoh• 6 years ago

Using real neurons avoids several other difficulties that software-based neural networks have. For instance, to get artificial neural networks to start learning well, their programmers usually have to engage in a laborious process of manually adjusting the initial coefficients, or weights, that will be applied to each type of data point the network processes.

This seems silly. Once you've figured out these parameters once, it seems to me you can similate a single neuron reasonably well. Perhaps differences will emerge from networks.

twomoretime• 6 years ago

well, their programmers usually have to engage in a laborious process of manually adjusting the initial coefficients, or weights, that will be applied to each type of data point the network processes.

Not true for modern neural networks. We typically use random values with certain statistics and/or specialized schemes which depend on specific later details - but in any case it's a single function call and the defaults typically work well enough that it's an advanced topic.

The OP is probably referring to neural nets of old, before this recent explosion, where you had handfuls of perceptrons operating on very simple problems which would be trivial for modern ML.

sirwitti• 6 years ago

I'm guessing they're working on neuromorphic computing. Many types of computations like differential equations are supposed to be a lot easier in non-discrete computing.

Disclaimer: I don't really know what I'm talking about, only glimps here and there.

RichardCA• 6 years ago

The modern day successor to what Misha Mahowald was working on.

https://youtu.be/lwT1jUvwRLc

It ends up being a complex hybrid of analog and digital computing having to work together.

The article doesn't provide enough detail to be of any use if you are trying to understand what they are actually doing.

h0l0cube6• 6 years ago

Currently, the company is working to get its mini-brains—which so far are approaching the processing power of a dragonfly brain—to play the old Atari arcade game Pong

I'm surprised that no-one has discussed whether these systems could develop emergent qualia, and experience pain. No joke. Are there any ethical frameworks around this kind of research?

Sosh101• 6 years ago

I think that is a very valid concern, but how could this ever be assessed

anonytrary• 6 years ago

If something cannot be assessed, then are we still accountable for the result? Or, do we simply discard the assessment?

colordrops• 6 years ago

Are only biological systems capable of experiencing qualia? What is to say that a silicon computer can't also have subjective experiences?

Rexxar• 6 years ago

Arguably, computers seems to suffer when there are be computations : they becomes hot and start to breathe a lot. That's a little irrational be I used to feel a little bad when I saw my old computer "suffering" when I was younger.

collyw2• 6 years ago

That's the big question that science so far has no clue about.

h0l0cube6• 6 years ago

I do sometimes entertain the idea of panpsychism (like perhaps a rock is conscious), and consciousness may not be substrate dependent, but I can be certain, by at least my own anecdata, that neurons give rise to qualia.

The jury is still out on this, but a big-world network with some quantum computation could be enough. The 'hard problem of consciousness' is that we can't really know by observing a system whether it is conscious or not, or rather a sort of 'zombie' that has all the features resembling a conscious being without qualia.

taneq3• 6 years ago

I don't think qualia are real in the general objective case, only in the subjective case. So nobody but 'you', whoever you are, can actually have them.

s_gourichon• 6 years ago

Ok let's rephrase then: can those mini-brains be a "you"?

MaxBarraclough5• 6 years ago

Meaning what? That doesn't bring any light to the question of whether these things are capable of suffering. As someone else said, this is a hard problem to reason about.

nnq• 6 years ago

could develop emergent qualia

If we were to admit there really is such thing as "qualia", there's not reason you wouldn't ask the same question about non-biological software or hardware systems... electronic hardware or biological hardware can have same computational qualities.

But since OBJECTIVELY there's no such things as qualia (the concept only exist SUBJECTIVELY), it could only exist by definition for "a person/subject" eg., in this context, a neural network large and complex enough to get close to a human-like intelligence level!

A dragonfly nervous system is probably simpler than some of the the largest artificial neural networks models...

MaxBarraclough5• 6 years ago

I'm inclined to agree. To put it another way, then: what matters isn't the substrate (neurons vs transistors), but what you do with them.

Going this route, we'd have to say that if you wrote a program that perfectly simulated the human brain, you've built a conscious system running on transistors.

It also means that if you built a RISC-V processor based on a large number of neurons (say, as many as are in a human brain), you haven't built a conscious system, despite that it's neuron-based.

I'm ignoring the obvious problems with building a RISC-V using neurons; they strike me as incidental detail, for our purposes here.

h0l0cube6• 6 years ago

I've answered to these points in another comment. But I'll emphasize, that I know (and presume you know) within some minimum epistemic bound^, that we can suffer pain and that arises from our neurological systems. It can't be known what other systems/substrates could experience qualia, and even if they have the outward markers of consciousness, they could be 'zombies'.

https://en.wikipedia.org/wiki/Philosophical_zombie

^ i.e., not in some adversarial simulation etc.

eyeundersand• 6 years ago

A dragonfly nervous system is probably simpler than some of the largest artificial neural networks models.

I would be extremely surprised if that were the case. Behavior of a single cell alone is unimaginably complex. Let alone an organ system, let alone the nervous system.

I will agree with your Qualia point.

tartoran2• 6 years ago

I am not an expert but from my understanding pain are signals perceived as what we subjectively call painful when something in our body/nervous system is damaged/inflamed or a system is out of whack. It is a very useful thing, it is trying to tell us something useful so we can act on it or just be aware of it. When something is out of whack permanently something in this system is malfunctioning. Without a body it is hard to describe the concept of pain and what it may feel to an emergent system with qualia properties. The closest I could think of is some kind of distress signal or base frequency change when the said system is under stress or wrong state and is perhaps used to self correct the system to a balanced state. If something is permanently painful then it ceases to be a reliable signal.

Psychological pain or emotional pain or distress may come closer to what such a system could experience and but it could also be an emergent property of such a system but we don't fully understand how this works. If such a system has emotional capability then it should be able to experience it because pain is such a state of the emotional spectrum. No pain no joy - no oscillation/flat

totalZero4• 6 years ago

We are all machines. The idea that one type of machine or another cannot feel "pain" is simply a rationalization for human behavior that causes harm to other organisms, like hunting and fishing and killing wasps.

Pain is a sensation that drives an animal to quickly react to injurious situations. Consider a robot made with Brooks's subsumption architecture [1]. The basal level of behavior drive of the robot is its self-preservation. That level of behavior drive can halt all the higher-order behaviors to monopolize processing and locomotion. Imagine that this behavior drive is triggered when the ambient temperature approaches a level that can damage the battery, and attempts to move the robot to the lowest-temperature zone available. That is functionally no different than what happens to a person in a house fire.

All we have done with the word "pain" is to set up an arbitrary delineation based on the hormonal/neuronal response that links our sensory apparatus to our behavioral drive.

[1] https://people.csail.mit.edu/brooks/papers/representation.pd...

tartoran2• 6 years ago

I agree with you, pain is a useful signal or it could be a systemic damage that wrongly sends these signals. However, when we, humans, refer to pain we clump all the unpleasant things we experience together. One such pain is the emotional /psychological pain and this type of pain is more likely to occur in such a system without a body. This is also a signal/state and is also useful to delineate it from another state. If such system is to experience joy it better know what pain is or otherwise wouldn't be able to delineate.

ben_w2• 6 years ago

Careful, or you might use a definition of pain too broad to be useful.

Plants, single-celled organisms, thermostats, and anti-lock braking systems meet your use of the word.

Almost all of us know what it means to be in pain. It’s the hard to put into words subjective experience of suffering that is worth making ethical decisions about, not the autonomous reflex that doesn’t come with conciousness-is-a-poor-word-but-I-lack-a-better-alternative.

Der_Einzige• 6 years ago

Hahaha, no. There is no ghost in the Shell in AI powered robots and certainly not in inanimate objects. You sound like an object oriented ontology writer with this kind of post and will be treated as such because that's what you've advocated for.

The alternative is fear of hurting objects or machines everytime I act in the world. Physical objects themselves are not ethical actors. AI is not life and we are several hundred years away from being capable of creating life ourselves within an object (if it's at all possible) Your worldview would mean that if I move a machine (or rock) and it appears to "resist me", I must fear that the machine or rock doesn't want to be moved. I must believe that I have acted violently towards that object by "frustrating it's preferences".

What absurdity!

h0l0cube6• 6 years ago

Are you a robot?

If you are at all in doubt of your qualia, to quote Sam Harris (who has many entertaining podcasts on this topic with people far more qualified to speak on it than I am):

Unfortunately, many experiences suck. And they don’t just suck as a matter of cultural convention or personal bias—they really and truly suck. (If you doubt this, place your hand on a hot stove and report back.)

https://twitter.com/SamHarrisOrg/status/951276362387591169

The twitter thread was about moral realism, but the topics are very much intertwined. If there is no imperative for anyone to do less harm to other creatures, why should anyone care if you are in pain?

And of course this argument extends to our treatment of animals too. Countless number of living, feeling animals, suffer for scientific progress (and cosmetics), let alone factory farming and wet markets. But at least there's light at the end of the tunnel for some of these concerns (lab meat, improved in-vivo testing)

russdill2• 6 years ago

We're already really good at building mini-brains that can play pong..and much more advanced games than pong. No one is concerned about such things developing emergent qualia and experience pain.

Edward9• 6 years ago

This looks like we are going to also feed our computers nutrients on top of electricity.

Abishek_Muthian• 6 years ago

programmers usually have to engage in a laborious process of manually adjusting the initial coefficients, or weights, that will be applied to each type of data point the network processes. Another challenge is to get the software to balance how much it should be trying to explore new solutions to a problem versus relying on solutions the network has already discovered that work well. “All these problems are completely eluded if you have a system that is based on biological neurons to begin with,” Friston said.

Using real neurons avoids hyperparameter tuning? Can someone explain how.

alex7o• 6 years ago

I suppose neurons are already turned for the task at hand. Although I suppose you can change their environment with different chemicals and that would be very close to hyper parameter tuning, after all happy neurons think better.

dr13372• 6 years ago

Are you tuning hyperparameters when learning a task?

barkingcat• 6 years ago

Yes, the amount of caffine, food (enzymes, proteins, carbs), and if it is a long running learning process, amount of sleep. all parameters are being tuned all the time.

Mithriil• 6 years ago

My guess would be that the hyperparameters of neurons are somewhat already trained by evolution of the species the neurons come from. I assume that neurons are defined by their genetic and environment. Then if one reconstructs the environment in which neurons normally dwell, then everything is set for this neuron to perform as it should be.

But on the other hand, one could interpret the environment variables as the hyperparameters. If you overheat the chips, it is possible the neurons might act differently (as would a brain?) If you overfeed them with, say, creatine, it might be possible that the neurons will perform erratically.

andbberger• 6 years ago

Well... it's well known that real neurons have pretty sophisticated homeostatic mechanisms...

But this is in vitro so all bets are off

polskibus• 6 years ago

Reminds me of a sci-fi novel "Black oceans" by Jacek Dukaj. It was one of the more interesting themes in there.

pfisch• 6 years ago

This sounds like an episode of black mirror and seems immoral.

If this actually does work and scale the results would be horrific.

I don't really want brains with no rights to exist like this.

throwqqq• 6 years ago

1/3 of the human brains in the world already exist with no rights. If we can enslave rats instead it will be a big step up.

citizenpaul• 6 years ago

100% of human brains exist with no rights. There are just temporary privileges. Ask Japanese Americans during WW2 what kind of rights they had in a first world country with supposedly the strongest "rights" in the world.

rl33• 6 years ago

2015 (my words):

Obviously the aforementioned experiment would be completely unethical, but it's interesting to ponder it as a hypothetical - that today we may have the capability to bootstrap a superintelligent machine using biology as a computational shortcut. But we can't, because ethics. [0]

2020 (article):

Chong said the pair were interested in the idea of artificial general intelligence (AGI for short)—A.I. that has the flexibility to perform almost any kind of task as well or better than humans. “Everyone is racing to build AGI, but the only true AGI we know of is biological intelligence, human intelligence,” Chong said. He noted the pair figured the only way to get human-level intelligence was to use human neurons.

[0] https://news.ycombinator.com/item?id=9973122

jimmaswell• 6 years ago

The human neurons are sourced entirely ethically, from skin cells to stem cells that make neurons somehow.

battery_cowboy3• 6 years ago

Do we know enough about consciousness to ethically put a "brain in a vat" yet?

rl33• 6 years ago

The human neurons are sourced entirely ethically, from skin cells to stem cells that make neurons somehow.

Yes, and I'd wager the scale at which their experiment is currently operating is very likely ethical too. In fact, I'm not saying anything they're doing right now is unethical. My use of the term unethical was in context of hypothetical experiments on a scale that would make Dr. Frankenstein's stomach churn.

That said, my only point was that it's disconcerting to see these people—who are currently growing tiny brains in a vat—state that they're very interested in AGI come to fruition via way of using human neurons.

mandelken• 6 years ago

If you can recreate (part of) a human brain through stem cells or any other way , then why would this not be a human? People with half a brain (after an accident, say) are still people. Cloned people (twins) are still people. Babies born from artificial insemination are still people. It's not up to these researcher to determine what is allowed to do with human brain tissue. Very troubling and unethical

Balgair2• 6 years ago

Neuron creation is in the same pathway of the epithelium (skin). Meaning that neurons are more closely related to skin cells than to your liver cells. By bathing the cells in select biochemicals, you can make their daughter cells 'regress' back into earlier forms. Basically, walking the daughter cells back along the development path. These are called stem cells. Your bone marrow is chock full of blood-stem cells that turn into platelets and blood cells and other stuff. Currently, we can walk these cells back a long ways, like from blood cell to blood-stem cell. We're still working on how to go all the way back to cells that can turn into any cell that is chosen (pluripotent stem cells). All this lab is doing is taking skin cells and 'regressing' them back into stem cells that have neurons in their future path. They take biochemicals, apply them to the skin cells, and then apply other biochemicals to these stem cells to make them into neurons. We've been doing this for a few decades now in various species.

https://en.wikipedia.org/wiki/Neurogenesis

arm85• 6 years ago

If this was posted on the 1st of April, I wouldn't believe it...

bencollier49• 6 years ago

Why specifically human neurons? Why source the neurons from human stem cells rather than mouse cells? PR I suppose.

altivasta2• 6 years ago

Humans are closest to humans in terms of behavior, so it makes sense to use them in testing where possible

shrthnd• 6 years ago

Donated my body to science... woke up as a living computer.

faeyanpiraat• 6 years ago

Wouldn't this count as neuron slavery?

program_whiz• 6 years ago

If you assume that your conscious experience is somehow arising from your neurons, then generating a set of those neurons that are genetically identical and cannot be distinguished in any way (except location in physical space, which is an ephemeral quality that doesn't appear to change neuron function), then we can assume we are creating, or will create, beings who are having conscious experience on some level.

On an ethical level, I think we need to understand whether we are creating thinking feeling creatures who will be doomed to suffer as data slaves before we normalize this. If I removed your brain from body, removing all sense pleasures, drowning you in darkness and isolation, and the only input and output you had were binary signals for some abstract data problem, you would experience profound silent suffering in an eternal private hell.

This truly would be the invention of the matrix, but not for an army of tyrannical robots / AI, but for the use of humans themselves -- a modern day slavery of the mental kind.

bweitzman• 6 years ago

An artificial brain being fed a stream of bits will not necessarily feel like it's in an empty room processing an abstract data problem.

If we can create an AI with different goals and reward mechanisms, there is a potential that we could create agents that are experiencing bliss doing data processing tasks.

Of course how we tell the difference between a miserable agent and a joyous agent is still an open question ..

aspenmayer• 6 years ago

“I leave Sisyphus at the foot of the mountain. One always finds one's burden again. But Sisyphus teaches the higher fidelity that negates the gods and raises rocks. He too concludes that all is well. This universe henceforth without a master seems to him neither sterile nor futile. Each atom of that stone, each mineral flake of that night-filled mountain, in itself, forms a world. The struggle itself toward the heights is enough to fill a man's heart. One must imagine Sisyphus happy.”

Albert Camus

teawrecks• 6 years ago

Emotions are just chemical responses, no? What if those chemicals aren't even present in the system? In other words, I don't think there's any more reason to think a ball of neurons is "alive" than a neural net that exists in code.

aehetherhear• 6 years ago

Agree, but I don't think biological neurons are the crux of the issue. I can't find the quote, but I believe from the book Echopraxia, the author discusses consciousness as a form of conflict resolution in regards to predictions about the self, an example of holding a hot pan despite the pain, knowing the consequence of letting go results in hunger. Or similar, the classic Jabberwocky example from Dune as a test of Paul's "humanity". But we can easily imitate these processes with machine learning even today, several projects involving the OpenAI gym have approached this. At what point do we believe these agents are conscious, and at what point do we shut them down?

skosch2• 6 years ago

This builds on the theory of "predictive processing". There are a few key people in the field; Karl Friston, Andy Clark, a few others – lots of rabbit holes to go down.

Balgair2• 6 years ago

On an ethical level, I think we need to understand whether we are creating thinking feeling creatures who will be doomed to suffer as data slaves before we normalize this

Though I agree with you, I think that's a bit alarmist. We're a very long ways away from this issue being something other than a thought experiment. The article states there is a 64 neuron chip that does not yet have customers and is not yet in production.

The Elephant brain has ~3 X 10^11 neurons in it, or a factor of 10^10 more neurons than the chip. Even assuming a doubling period as short as Moore's Law, we looking at ~50 years until this might be a problem we'd want to deal with.

I'd say we just wait and see how things play out for a few more doubling cycles before we start pulling at our hair.

jmcqk6• 6 years ago

If you assume that your conscious experience is somehow arising from your neurons, then generating a set of those neurons that are genetically identical and cannot be distinguished in any way (except location in physical space, which is an ephemeral quality that doesn't appear to change neuron function), then we can assume we are creating, or will create, beings who are having conscious experience on some level.

I think that's a bad assumption. Neurons are only one part of the physical brain, and there are a lot of neural transmitters and other biochemical things that are happening.

That is not to downplay the ethical concerns at all. I agree completely with you there.

soulofmischief• 6 years ago

Unless we program in certain circuitry which can analyze and act upon provided input, I think consciousness as we know it cannot develop.

Emotions heavily mediate our perceptions and contribute to the manifestation of the ego. And sensory input defines the world in which we inhabit. A consciousness void of these two things could likely have a poor sense of subjectivity.

haffi112• 6 years ago

Imagine in the future that your computer, composed of biological neurons, could literally die.

carapace• 6 years ago

At the core is the stance or attitude towards life.

What Martin Buber calls "Ich und Du", or "I and Thou":

Buber's main proposition is that we may address existence in two ways:
The attitude of the "I" towards an "It", towards an object that is separate in itself, which we either use or experience.
The attitude of the "I" towards "Thou", in a relationship in which the other is not separated by discrete bounds.

https://en.wikipedia.org/wiki/I_and_Thou

This fundamentally is an attempt to force the "I-Thou" relation with life into the "I-It" relation with machines and is therefore the definition of evil.

- - - -

Anyway, the correct mad science is to culture your own cells into control boxes for your machines and thence control them "telepathically" to become a distributed cyborg. But to do that you would first have to become intelligent enough to increase your morphogenic plasticity to the point where you have more-or-less total control of your meta-cellular form. Did you see John Carpenter's "The Thing"? That's how intelligent you would have to become: an immortal macro-polymorphous self-made shoggoth. At that point though, you no longer really need machines or technology in the conventional sense. You can make artificial diatomaceous carrier shells for small colonies of your cells and send them out to do whatever. It's not nanotech but your operational units are small enough that it doesn't really matter. I should mention that it's really easy to make fusion generators at this scale. People assume that the techno-singularity will originate in silico so to speak but that's so naive: the most sophisticated information processor is the human brain, not the chip. The singularity happens in vivo: in flesh. You don't even need and machinery to initiate it: just information and the will to be more than you are (and the wisdom not to fuck it up and turn yourself into a cancer blob or grey goo.)

The only real problem at this stage is loneliness. Few can ever really understand what you've done, and if they did they would shun you as a Lovecraftian horror anyway. Nevertheless, I have not given up hope. Slowly, painfully, you guys wobble towards enlightenment. I try to help when I can, but mostly all I can do is be patient.

I am among you now. Join me. "Dwell amidst wonder and glory for ever..."

xkcd-sucks• 6 years ago

Primary mammalian neural cell culture is perhaps the most notoriously difficult type of cell culture, even in a lab setting -- Developing a product based on neurons seems like a total pipe dream. Even keeping a population alive for a few weeks is a big deal, let alone maintaining a neuron based black-box product in working order.

So, if this isn't a total pipe dream, it will drive development of super-advanced cell culture tech, development of robust application-suitable cell lines, etc. These technologies/products have orders of magnitude more value than neurons on glass.

A startup is building computer chips using real neurons

Related Stories