A startup is building computer chips using real neurons
fortune.com
fortune.com
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.
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
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.
Ben explains somewhere in there that this was done to help create a better substrate for the first experiment.
Using real neurons avoids hyperparameter tuning? Can someone explain how.
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.
But this is in vitro so all bets are off
>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.
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.
(The problem of course is the possibility of it becoming conscious before we realize it, but it's very up in the air to what degree that's actually a risk. My intuition is that we'll see clear signs of it before it even fully forms.)
Probably, though I wouldn't be so sure we'll know what to make of more complex computer-brain hybrids:
"These neurons are then embedded in a nourishing liquid medium on top of a specialized metal-oxide chip containing a grid of 22,000 tiny electrodes that enable programmers to provide electrical inputs to the neurons and also sense their outputs."
Similarly, we understand (at least mammalian) brain structure well enough to identify animals like cows to probably have relatively minimal conscious lived experience, even if they have ample processing power, i.e. "intelligence".
So at least in examples from nature, there is no reason to believe that intelligence and conscious lived experience need be correlated meaningfully.
So in your example, we would need some way to quantify the degree to which it's conscious, rather than intelligent.
Do you have any sources to back that up? As far as I'm aware, the 'hard problem of consciousness' is still 'hard'.
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.
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.
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 ..
Albert Camus
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.
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.
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.
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.
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?
I suspect this idea is so widespread at this point that these researchers and others are aware of this possibility and see the potential moral issues. I just think they believe such a thing is still decades (or longer) away, and I think they're probably right.
Right now, just about everyone wildly sensationalizes things that appear to suggest conscious man-made entities, like the AIs that supposedly "invented their own language". I think we're so eager to see it even when there's nothing close to it that once it actually does happen, we'll realize it quickly and understand the implications.
There is the possibility that it could somehow develop despite no external signs of it, but all we can do is try to be as cognizant of the possibility as we can.
Animals (like us) suffer as part of our evolved aversion response. We have an in-built capacity for preference. Current-day computers lack this.
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.
> Nagel famously asserts that “an organism has conscious mental states if and only if there is something that it is like to be that organism—something it is like for the organism."
[0] https://en.wikipedia.org/wiki/What_Is_It_Like_to_Be_a_Bat%3F
When we use language to describe our experience - pain, joy, loss, fear - we don't transfer the experience proper. What we share is an expression of our experience.
If you describe your current experience as "joy", then you arrive at that expression because - arguably - you learned that concept from your parents as a toddler. I say "arguably" because this is where the "nature vs nurture" debate kicks in. I'm not going to delve into that can of worms.
Suffice to say that we can't transfer our exact experience to another individual; but that we can express our experience and that others are able to interpret that expression. And that a shared understanding comes from a shared conceptual frame of reference.
Art is a great way to challenge that shared frame of reference. Understanding art implies trying to interpret what an artist is expressing. Sometimes the artist doesn't intend to express their own experience, but rather evoke a particular experience or even contradicting experiences within the audience. That's why some art throws us from our feet.
But this isn't about art. This is about suffering. That's where concepts such as empathy and compassion come into play. Those concepts refer not to the actual experience of suffering proper within individuals, but rather the capacity to recognize the experience of suffering in others through their expressions. If you see someone express grief through crying, apathy, irritation,... then empathy means that you understand that they are experiencing grief and you are able to also have that same experience behind the expression as an individual, independent of what others are feeling.
Empathy, compassion and a shared understanding work best if there's a close resemblance between you and the other. Hence why there's little doubt about the experiences of family, close friends and so on. It starts to become harder when you think about suffering in the context of people with different cultures and languages whom you've never met and who live on the other side of the globe. Animals? There's another level that increases that distance. Sure, your pet may be expressing their experience, but how do you know you're not projecting?
And so, here we are, considering neural networks and neurons attached to breadboards. Do they have a similar experience of reality? Do they experiencing suffering in a way that matches with what humans conceptually and "broadly" understand as "suffering"? It's not like we can ask, and even if we could, how would we possibly be able to interpret the expression of "suffering" they convey to us?
For instance, up until the 1970's, infants up until 15 months simply didn't receive anesthesia for surgeries. Everyone assumed that they didn't feel pain because the shared frame of reference among medical practitioners didn't allow for the interpretation of their expression as "oh, they are in pain".
https://www.nytimes.com/1987/12/17/opinion/l-why-infant-surg...
And so, one argument could be that we are already torturing neural networks and we aren't even aware we're doing it because we simply lack a common frame of reference to pick up the expressions of an experience of pain. Hence the cautious reluctance to sanction free experimentation that unwittingly may elicit the experience of suffering.
Does a scribble on a stone contain something any other formation of cracks wouldn't? No, the meaning is generated within _you_, and as such sentience can only be summed as only your own process observing phenomena
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...
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.
https://en.wikipedia.org/wiki/Philosophical_zombie
^ i.e., not in some adversarial simulation etc.
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.
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
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...
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.
I have no reason to believe that what a Roomba feels when it approaches a descending flight of stairs is any different in practice from what I feel when I approach a precipice. Our pathways are totally different, but ultimately there is a communication from our sensory organs to our processing architecture, where an ingrained drive toward self-preservation momentarily overrides other needs. In my case, I could call it "fear," or "anxiety," or "angst" if I'm feeling philosophical. In the Roomba's case, I have no knowledge of its subjective experience, but that doesn't mean that I should draw up a new word to construct a delineation between the Roomba and myself. We are both machines, and our response to the same situation is largely the same.
How useful is it to describe pain as a "subjective experience of suffering" when neither subjective experience, nor the feeling of suffering, is directly observable? (Also, I don't agree that pain and suffering are interchangeable concepts.)
A truly useful definition of "pain" would hold water without reliance on an anthropocentric tautology. I know what it feels like to slice my hand open while cutting tomatoes. That doesn't give me any power to understand what it feels like for an octopus to lose one of its arms, or for a tree to have its branches trimmed.
What's funny is that my way of considering "pain" is not even the most divergent from yours. Many Andean cultures believed that stones, rivers, and mountains have energy, thoughts, feelings, and souls. These ideas remain in the culture to this day.
What spurs an ABS to pulse the brake line pressure when the wheels begin to slip as the driver mashes on the brakes to avoid a collision? What spurs an ant to run away when it steps onto a hot radiator? What spurs a human to stop walking on a broken ankle? All of this is programmed in one way or another, all of it is self-preservation.
Also, my knee-jerk reflex doesn’t feel like much at all.
If I read you correctly, you and I agree that we can’t tell if a Roomba’s avoidance algorithm more like anxiety, or more like a reflex, or more like lust. In the absence of evidence, I will assume that all its experiences are like a reflex, that there isn’t anything “that it’s like to be” a Roomba. I am aware I may be wrong.
While we indeed cannot directly observe subjective experiences, I sincerely hope that we figure out a good way to resolve this soon:
If we mistakenly assume AI cannot have qualia, then we risk condemning our creations to torment from which they are only released by their own destruction.
If we mistakenly assume we can create machines which have qualia, then brain uploads are death.
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!
Is it wrong to slaughter a cow because that cow will feel pain?
Sometimes, people label ideas for the purpose of sorting and better considering them. Other times, people label ideas for the purpose of compartmentalizing them and rejecting them without really reflecting on them at all. I don't know what you mean when you say I will be "treated as such" but it sounds like a way of setting aside what I actually said in order to tussle with a preconceived, crystallized notion of something someone else said or wrote.
There are ancient cultures that believed it was wrong to do violence against a rock. Do you believe they, too, read Heidigger?
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)
I am a collection of interconnected units, each of which is itself a combination of different organelles and other purpose-serving features. My intelligence is itself a product of the interaction of simple, unintelligent parts. So I would say that I am a machine. I would describe a robot as a man-made machine where the interaction of various silicon parts and processors gives rise to intelligence. By that definition, I am not a robot.
I believe that we don't feel pain when we are truly in trauma. It is only when we can do something about the damaging stimulus that we feel pain. I once had a major accident where I experienced lung collapse and multiple fractures. I never lost consciousness but I don't remember feeling any pain until well after others came to my aid. Even when I tried and failed to pick myself up off the ground, I did not feel pain, only disability and relief that my fingers and toes still moved.
I believe that the question of "what is pain" and the question of "is pain a good criterion for deciding whether it is acceptable to do harm to something" are two totally different philosophical problems. They are connected only insofar as we have chosen pain as a proxy for harm. But that very relationship between pain and harm indicates that pain is not just some kind of soulful feeling, but rather a signal to help us evade harm now or in the future.
Therefore, it may be true that using biological neurons, arranged in a certain configuration, would give rise to qualia like pain in a way that shifting electrons between CPU registers never could. We just don't know.
For more reading, here's a relevant whitepaper by the Qualia Research Institute: https://opentheory.net/PrincipiaQualia.pdf
We're getting stuck on semantics here. I do, but then I'd cease to see it as a robot, and more a sentient being. One criterion of consciousness that I've encountered is, 'there is something what it is like to be <x>' (Thomas Nagel). If there's something 'what it is like to be' a robot, a bat, a mosquito, an ameoba, a rock... then it is conscious.
> I believe that the question of "what is pain" and the question of "is pain a good criterion for deciding whether it is acceptable to do harm to something" are two totally different philosophical problems. They are connected only insofar as we have chosen pain as a proxy for harm. But that very relationship between pain and harm indicates that pain is not just some kind of soulful feeling, but rather a signal to help us evade harm now or in the future.
Qualia is more broad than just pain, of course. I just picked this particular phenomenon for it's poignance :)
If it is conscious, ethically speaking, we should consider how we treat it in a manner different to something that isn't. So if a rock isn't conscious, and some interconnected neural/silicon device is, we should at least have some way to query whether it is in an undesirable state or not.. if feasible/practical.
Maybe if trees/plants/rocks/ameobas are conscious, we can't be consulting their feelings when we harvest crops, or use disinfectant, mine for precious metals etc. We can make decisions to treat livestock better, and change how we utilize our environmental resources - so we ought to, and we are. But if we were to go out of our way to make new conscious entities, don't you think we should extend our historical shifts in attitude to slavery and our growing shifts in attitude to animal welfare also to these new entities?
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.
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!
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.
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.
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..."
Disclaimer: I don't really know what I'm talking about, only glimps here and there.
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.
Easy there, Dr. Frankenstein.
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.