Growing mammalian cell spheroids in vitro was one of the highlights for me and my scientific career. So rewarding and interesting, so much more useful than monolayer cell cultures.
The article is about neurons, but they are fascinating in cancer research also. You can grow fibroblasts together with cancer cells and they start making something like a real tumor and start exhibiting responses more like in vivo tumors. My favorite way to grow them was just seed the cells in special plates that the cells don't want to attach to and they just start attaching to one another and start making a colony of self-organizing cells floating in the media.
>any model that was trained using a quantity of computing power greater than 10^26 integer or floating-point operations, or using primarily biological sequence data and using a quantity of computing power greater than 10^23 integer or floating-point operations."
>using primarily biological sequence data
I was wondering about that bit.
Those who are straying from gods light the fastest are firmly in the shadows.
Can anybody point me to a better introduction into organoid intelligence? This all looks important and interesting, but this article seems to be all over the place. Turing test and CAPTCHA are not very convincing arguments brains beat AI. Brains store 2,500 (terabytes)? I'd like to see a source for that. Most brains I meet don't seem to know more about the world than a 7B LLM
Most brains don't treat themselves to an abundance of training data, the ones who do seem pretty knowledgeable to me.
Witty comment aside, the human brain is pretty efficient in terms of energy use considering it's taking in a ton of data while it's conscious. Two each audio and video streams, olfactory, gustatory, touch, vestibular, and all the interoception. Inference and training in real time. All for the low price of 125 watts, a quarter that if you're just measuring the brain and not the whole body.
I have a suspicion that organoids will be to artificial intelligence what ornithopters are to flight. Technically possible, but reinventing it from scratch instead of following biology strictly will give us dramatic performance improvements.
All for the low price of 125 watts, but you need to sleep one fourth of the time, during which it doesn't do much, and on the other three thirds "Most brains don't treat themselves to an abundance of training data".
I'm not sure I'd call this machine efficient at using its computational power.
Also, that is the net energy expenditure of the body, but not the input required. We use a relatively antiquated form of the internal combustion engine to turn food into energy, which seems to be around 25% efficient, at best. It's also built on chains of other fairly inefficient biological entities, so it's hard to say what the overall efficiency is, but suffice to say that we're prevented from using the most efficient energy sources.
It's basically a cleaning cycle for the brain. If we don't sleep we die pretty soon. Meaning it's necessary daily scheduled maintenance downtime, which should count directly to efficiency calculations!
A completely unquantized fp16 model weight 7B LLM is about 15GB on a disk. You need closer to 24GB of memory for inference with a decently sized context.
Quantization is black magic of the software variety that seems to be able to significantly reduce that without a commensurate loss in quality, though the results are a little subjective. Some well reviewed quantizations of 7B models can get them below 9GB.
I wonder if any of the people working in this field write SF that ponders the many thorny ethical knots this involves. There's a paragraph late in the article that makes it clear they are aware of them.
On the one hand, we have tiny, artifical lumps of vat-grown neurons, that lack any self-awareness, emotions or even basic pain reception (because, you know, you kinda need a whole sensory apparatus for this).
On the other hand, we have living, breathing, feeling beings with complex social and emotional lives, capable of experiencing joy, sadness, horror, panic, friendship and attachment. Oh, and they absolutely can feel pain. Also known as pigs, cows, chickens and other vertebrate farm animals.
Now, humanity kills about 80 billion such farm animals per year. [1]
So how do the "many thorny ethical knots" compare to that if I may ask?
> Now, humanity kills about 80 billion such farm animals per year. [1]
Ostensibly, those animals are killed for sustenance, where lab intelligences will be created, poked , prodded, altered and killed merely for fun or curiosity. Even if we disagree on the ethics of eating meat, hopefully we at least agree that there's a meaningful difference there, especially since we don't know where the boundary of "basic pain reception", or indeed any kind of experience, truly lies.
Except eating meat is unnecessary and ultimately unsustainable for the planet.
So really people like me who eat meat are guilty two sins -- eating intelligent creatures that don't need to be eaten and destroying the planet by doing so.
also, don't underplay the utility of this kind of research, people don't just do it for funsies, they do it because it could change the world.
With the many alternative food sources we have available to meet our needs for sustenance (which are also less wasteful and more climate-friendly) it is hard to argue that these animals die for anything but to keep up our eating habits or simply for the pleasure of eating them.
The big problem with this thinking is the assumption that change can happen overnight. Large complicated systems to not change overnight unless they are burning down and even in the smoking ruin, the incentive structure that lead to the previous outcome might still be there.
As sad as it may be, we don't live in a world where each and every human on earth today can go and eat a serving of tofu. There isn't enough soybean grown and distributed around the world for that to happen. Stop eating meat tomorrow and a huge portion of the world doesn't just eat soybean that doesn't yet exist, they starve, or they eat meat that currently exists and is produced by large systems that individuals are powerless to shape. Overtime we can strive for change, but to expect immediate action and blame selfish pleasure and not the complexity of reorienting the entire worlds food supply is a bit naive.
Given that meat production accounts for >15% of global greenhouse gas emissions, this is literally what is happening.
> There isn't enough soybean grown and distributed around the world for that to happen.
Excuse me, where do you think all the calories we feed livestock comes from? Despite what greenwashing pictures on meat-packets try to tell people, we don't grow 80 Billion pieces of livestock anually by letting them graze peacefully on fresh green pastures in wide open plains.
Instead we waste truly vast areas of land growing maize, soybean and grain, aka. foodstuffs that are completely fit for human consumption, which then get sold as feedstock, to go into one of the most nonsensical wastes of resources in history. Just as an example of how absurd this is: The energy efficiency for beef production is 25:1, meaning you need 25 calories of feedstock, to make 1 calorie worth of beef. And again, these are 25 calories of things humans could eat.
For comparisons sake: The energy differential between LED and incandescant bulbs is merely 1:10. No one argues whether we should use LED or incadescant bulbs any more, but somehow for meat and it's much worse ratio this is still somehow up for discussion?
So yes, we do absolutely produce enough calories from growing plants alone, to feed every single person on this planet. In fact we could feed many more people, and have way less problems, if we didn't waste mountains of maize, soybeans, grain and other valueable foodstuffs, by force-feeding them to animals.
I don't think you'd be happy eating out of a bag of feed corn. they eat a little soybean for balancing the diet but they don't each much soybean as it affects their gut if they have too much. Still, that soybean has a long way to go before its a brick of tofu wrapped in plastic sitting on your grocery store shelf. There is a certain amount of manufacturing capacity that would need to be built out. You might not even want to put it where you have the existing feedlots and meatpacking industries because these are sited with the idea of moving livestock by rail from grazing lands to feed lots near marketplaces in mind.
> I don't think you'd be happy eating out of a bag of feed corn
I am however, more than happy to eat any number of products that contain corn in various other forms. Because, you know, "feed corn" is just, well, corn, and instead of feeding it to animals, we can give it to humans instead.
> Still, that soybean has a long way to go before its a brick of tofu wrapped in plastic sitting on your grocery store shelf.
That cow has a long way to go before its a bag of meat on a grocery shelf as well. And that way wasted ALOT more energy and resources than the way my Tofu block does.
You are not wrong about the energy equation. But the facts are still we have an industrialized food supply that depends in no small part of meat consumption. Replacing that with non meat alternatives is not always possible for certain things but also expensive to retool and reoutfit. Supply chains take a long time to establish.
> that depends in no small part of meat consumption.
No, it really doesn't. It's a conscious decision of society that we produce, and consume, a lot of meat.
As I have repeatedly demonstrated in this thread, a huge amount of the calories we put into the production of meat, come from human consumable foodstuffs. We have the production facilities to turn these grains and beans into flour, bread, tofu, canned beans, whatever. We do it every day, on an industrial scale.
We could do so any time to these feedstocks as well. The technology exists, the facilities exist, the infrastructure exists, the supply chain exists.
It cannot happen overnight, but it could happen in a few years. Already now for each gram of beef for example, you have to grow 8-10 grams of maize or feed corn to raise the animal. (Almost no animals are gras fed.) The same area used to grow that maize could be used to grow plants for humans directly instead of "refining" the plants into meat/beef/dairy as it is sometimes called.
Processes have to be adapted,for sure and that will take time. But that is not, I think, the limiting factor at the moment. The limiting factor is our appetite for animals.
A few years is also a rapid change to replace all the products on the market that are animal derived. They use the entire animal you know, not just the parts in the supermarket. Usually these analysis don't consider that they just consider the caloric benefit. Imagine telling cardiologists they can no longer source pig hearts for open heart surgery.
Study after study has demonstrated that people have issues with vegetarian diets, not just with limited food availability, but health issues and cost. Healthy vegetarian diets are still the domain of the economically and educationally privileged.
You can make an argument that we don't have to eat as many animals as we do, but you simply can't make the argument that we shouldn't eat animals at all.
Feel free to browse through the "similar articles" section under the linked study.
> Healthy vegetarian diets are still the domain of the economically and educationally privileged.
For most people in the world, especially outside of comparatively rich western societies, meat is actually a costly alternative, and plant based foodstuffs are the norm.
The research could help billions of future people avoid certain brain/mind diseases, and could lead to the sort of breakthroughs that will allow quadrillions of minds to live within the solar system or beyond.
Even if we accepted that this somehow makes the ethical conundrum of killing beings that can feel all the pain, horror, fear, terror and dispair thus inflicted, despite having more than enough alternative food sources available to us, magically disappear...
...different studies indicate that western households waste ~1/5th of the food they buy. So, simplifying of course, this would mean that ~16 Billion creatures, who feel pain just as much as humans do, are killed every year ... just so we can throw their rotten carcasses in the garbage.
> hopefully we at least agree that there's a meaningful difference there
Oh, absolutely, only I guess not in the sense that you think we do.
Because these vat-grown lumps of neurons don't feel. They are not conscious. They don't experience pain. Even simple E.Coli bacteria grown on an Agar-Medium have more resemblance to life than they do, because the bacteria at least are beings fully capable of existing outside of a lab-provided growth medium.
So yes, there is a difference: Probing around in lab grown lumps is research, and killing billions of conscious beings so they can then rot in our garbage bins is morally indefensible.
> Because these vat-grown lumps of neurons don't feel. They are not conscious.
Being conscious is not a prerequisite for them feeling something as far as we know, and you simply don't know that it doesn't feel anything.
I agree that food waste is an issue that requires better education, as does nutrition in general. Then again, nutrition research has been abysmal until only very recently, so people have been getting confusing messages for decades now.
> and you simply don't know that it doesn't feel anything.
Yes, I do know that, because I know that pain reception requires nociceptors [1].
In fact all bodily sensory information requires a periphery of sensory neurons. The central nervous system itself, is simply incapable to collect sensory information on its own.
And our complex emotional lives are the result of a developmental process that involves pathway selection by usage patterns.
None of these things are present in a lump of cells in a growth medium.
Note that The Time Machine explores the ethical knot of "humans eat other vertebrates" from one direction, while Animal Farm explores it from the other.
This is literally it. In fact I'd posit that when we start being able to easily translate animal communication we'll probably stop being able to stomach eating them.
What is the limit? I imagine fully sentient super-humans. Somewhere in between the Kings from Foundation and the androids from Bladerunner. Or perhaps both - the Kings from Foundation ruling over Androids from Bladerunner in a world ever so decreasingly "natural".
> While ChatGPT can efficiently collect information on the internet, it can’t react to a change in temperature like a cultured cellular system can, he wrote.
Is there a qualitative difference between an LLM looking up the temperature on the internet (or getting it from an attached thermometer) and an animal brain receiving it from a thermoreceptor? The latter may be more integrated, but it sounds functionally identical to me.
In particular the human brain is an 80-billion core processor with each core running at 300Hz. Its architecture is not so good at FLOPs and other things we typically use computers for, but overall human brains dramatically outperform anything in 2024 except the world's fastest supercomputer.
I think the correct term is “outclasses”, not “outperforms”. Our performance is pretty bad, especially on average and compared to other species (when comparable).
A Neuron on its own is capable of exactly one thing, and that's signal integration from it's dendrites. That's it. Wait until the excitation potential reaches the point where the Potassium channels switch and boum: Signal cascade goes down the axon.
The only moderation a single neuron is capable of, is varying signal resistance to filter out repeating noise.
There are any number of systems that stubbornly refuse to be capable supercomputers, despite being incredibly hard to simulate.
Accurately simulating every aspect of the liquid flow through an acorn leaf, with it's billions of micro and nanometer channels, would easily be one of the most difficult tasks in the history of computing, and yet, acorn leaves are not even capable of the simplest thought, and the entire modus operandi of water flow through an entire tree can be accurately summarized in a simple 1-page diagram.
Every single muscle fiber with its trillions of moving cytoskeleton components, has a much higher complexity than a neuron, that doesn't mean muscles are supercomputers.
I agree that complexity doesn’t imply capability. However, in general they are different architectures which cannot be readily compared. Saying a neurone can only do signal integration from dendrites is like saying a computer can only add 0 to 1. The reality is those dendrites are spatially arranged and may perform computations before reaching an action potential.
Also, neurones are still cells and can change structure/position as well as gene expression, which may (or may not) impact on processing over a longer time interval.
No, I mean in terms of raw computational power: 300Hz * 80 billion cores. Again the architecture does not work well for churning through a giant array of floating point arithmetic since 300Hz is a limiting factor for that specific type of computation. But that's not the kind of problem the human brain was designed to solve.
Humans can't eat just anything though, and there's a tremendous amount of energy embodied in human food, including growing, storing, transporting and cookihg it.
And is also true for computers in different ways. Human food comes included in the natural
Environment, electricity requires some degree of commercial mining.
If you want to go with imaginary algorithms that don’t exist yet (or possibly ever) sure such an imaginary computer or system could do it more efficiently.
My understanding is that neurons are simultaneously electrically active and chemically active, and are organized into larger systems that are dynamically plastic.
Somewhere in there there's some efficiencies being achieved that come from being functionally active in ways that it don't make for an easy one to one comparison to transistors.
Right, lots of things can map onto models of computation described in typical theory of computation textbook, and that gives you your one to one comparison to transistors. You can also find equivalences between board games, lego, conways game of life etc. etc.
But what I'm saying is, there is a layer (or layers) of organic function that is abstracted away in machine learning/AI. While the field does borrow ideas liberally from brain architecture (as it well should), you nevertheless have neurons doing things like chemical signalling, metabolic regulation, unique site-specific structural organization, and processes for dynamic reorganization of structure as needed.
That is, not just in the informational representation, which can be sufficiently abstract to model practically anything, but in the hardware itself upon which the information rests. It's there that I suspect there are efficiencies gained by brains over transistors.
The article is about neurons, but they are fascinating in cancer research also. You can grow fibroblasts together with cancer cells and they start making something like a real tumor and start exhibiting responses more like in vivo tumors. My favorite way to grow them was just seed the cells in special plates that the cells don't want to attach to and they just start attaching to one another and start making a colony of self-organizing cells floating in the media.
A good overview here-
https://cancerci.biomedcentral.com/articles/10.1186/s12935-0...