Living human brain cells play DOOM on a CL1 [video]

The truth is, God really gave 11 commandments.

It's just "Thou shalt not grow a brain in a test tube and force it to play a 1993 shooter" didn't make any sense to Moses and therefore didn't make the editors cut.

It is creepy, I agree.

I saw this article over the weekend and felt similarly: https://theinnermostloop.substack.com/p/the-first-multi-beha...

> Watch the video closely. What you are seeing is not an animation. It is not a reinforcement learning policy mimicking biology. It is a copy of a biological brain, wired neuron-to-neuron from electron microscopy data, running in simulation, making a body move.

And the simulated world they put it in is a sort of purgatory-like environment.

Maybe you're a brain in a jar somewhere being forced to live this life you're living.

> Just feels like it's starting to get quite questionable

There's no way the technology to make and modify "life" including cloning humans hasn't been secretly used or attempted at least once ever since it was discovered.

How else are they going to train the pilot wetware for the AI robot army?

The thing should watch cats.

> it's creepy.

It's awesome.

People's ick around bodies, which are machines, have always held us back.

It wasn't until we started cutting them open that modern medicine was developed.

We might have brain uploads already had we not been so averse to sticking brains with electrodes.

I'll go further: had we not been so scared of cloning, we'd probably have cured cancer and every major ailment if we'd begun cloning monoclonal human bodies in labs. Engineered out the antigens and did whole head transplants. You could grow them without consciousness or deencephalize them, rapidly grow them in factories, and have new blood / tissue / organ / body donors for everyone.

New young bodies means no more cancer, no more cardiac or pulmonary age. It's just brain diseases left as the final frontier once we cross that gap. And if we have bodies as computers and labs, we'd probably make quick work on that too.

Too tired to lay out the case / refute, so past discussions:

https://hn.algolia.com/?dateRange=all&page=0&prefix=true&que...

>But perhaps something getting close to human intelligence

this isn't getting close to human intelligence. They're using about as many cells as a fruit fly has (of course not actually functioning like an animal brain) processing signals to play Doom. The treatment of a single farm chicken is about a few magnitudes more worrying than this.

I'm sorry to tell you that you're made out of human cells and I don't think you got consent from each brain cell before firing up the old boomer shooters.

What if the braincell-vibe JS libraries turn out pretty much identical to the legacy human JS libraries, aside from being better-commented. That might lead to an existential crisis for some folks.

Old news. Google "my dog vibecoded a game".

Do you feel like you have no mouth and you must scream?

> The neurons serve as a biological filter: the training system translates screen pixels and ray-cast distances into electrical zaps, the living cells fire spikes, and those counts feed straight into a PyTorch decoder that maps them to Doom actions. The PPO agent, CNN encoder and entire reward loop run on ordinary silicon elsewhere. Cole’s ablation modes make the split testable, set decoder output to random or zero and the game still plays. The CL1 hardware interface works exactly as advertised. What remains unproven is whether 200,000 human neurons can ever carry the policy instead of just riding along.

Yeah… That’s quite the smoking gun.

So it’s quite likely then that the neurons are just acting as a bad conductor. The electrodes read a noisy version of the signals that go into the neurons, and they just train a CNN with PPO to remove that noise, get the proper inputs, and learn a half-decent policy for playing the game.

If this worked as advertised they shouldn’t need a CNN decoder at all! The raw neuron readout should be interpreted as game inputs directly.

Besides, they are not streaming the video into the neurons at all. Just the horizontal position of the enemies and the distance, or some variant of that. In that sense it’s barely more than pong isn’t it? If enemy left, rotate left, if enemy right, rotate right, if enemy center shoot. At a stretch, if enemy far, go forward, if enemy close, go back. The rest of the time just move randomly. Indeed, the behavior in the video is essentially that…

While we are at it, the encoded input signal itself is already pretty close to a decent policy if mapped directly to the keys (how much enemy left, center, right), even without any CNN, PPO or neurons.

EDIT: It seems like the readme does address these concerns, and the described setup differs significantly from the description in the critical blogpost. Still not entirely convincing to me, a lot of weights being trained in silicon around the neurons, but it sounds better. I don’t have time right now to look deeper into it. They outline some interesting details though.

> Quote from: https://raw.githubusercontent.com/SeanCole02/doom-neuron/mai...

Isn't the decoder/PPO doing all the learning?

No, this is precisely why there are ablations. The footage you see in the video was taken using a 0-bias full linear readout decoder, meaning that the action selected is a linear function of the output spikes from the CL1; the CL1 is doing the learning. There is a noticeable difference when using the ablation (both random and 0 spikes result in zero learning) versus actual CL1 spikes.

Isn't the encoder/PPO doing all the learning?

This question largely assumes that the cells are static, which is incorrect; it is not a memory-less feed X in get Y machine. Both the policy and the cells are dynamical systems; biological neurons have an internal state (membrane potential, synaptic weights, adaptation currents). The same stimulation delivered at different points in training will produce different spike patterns, because the neurons have been conditioned by prior feedback. During testing, we froze encoder weights and still observed improvements in the reward.

How is DOOM converted to electrical signals?

We train an encoder in our PPO policy that dictates the stimulation pattern (frequency, amplitude, pulses, and even which channels to stimulate). Because the CL1 spikes are non-differentiable, the encoder is trained through PPO policy gradients using the log-likelihood trick (REINFORCE-style), i.e., by including the encoder’s sampled stimulation log-probs in the PPO objective rather than backpropagating through spikes.

Counterpoint: a major use case for this technology would be to experiment on human brain structures to research and hopefully cure neurological diseases like Alzheimer's. If you want to cure Alzheimer's in humans, you might as well use human brain cells from the start.

But yes, I agree that they're likely using human brain cells mainly because it's attention-getting.

Whoever thought people would become Dr Frankenstein for the karma.

[dead]

I looked into it. They're not feeding the framebuffer to the neurons, but have a "signal" when an enemy is on screen to some of the tissue's inputs, and how to locate it in the x/y axis, and have outputs for the character to turn right or left or fire.

It's "see this input signal, send these output signals", which seems consistent with the title.

It seems they grow the neural tissue on a chip the neurons can interface with and send out / receive electrical impulses. They let the neurons self assemble, and "train" via reward or punishment signals (unclear to me what those are).

Either way this makes me nauseous in a way I haven't experienced much with tech. The telling thing for me is, all these people are so excited to explain, but not once, ever, in the video speak of ethics or try to mitigate concerns.

We know this is only 200,000 neurons. Dogs have 500 million. Humans have billions. But where is the line for sentience, awareness? Have we defined it? Can we, if we don't understand it ourselves? What are the plans to scale up?

It's legitimately horrifying to me.

Previously it played pong. Rather poorly. Then they added a "python programming layer." Now it "plays" doom. I agree with your suspicions.

These are lab-grown biological neurons. Why are they any more problematic than Nvidia's silicon neurons?

100%

We know that neurons can produce subjective experience.

This is the first time in my life that I've felt a scientific avenue of research should shut down.

To be a fly on the wall in that ethics committee meeting...

I have no mouth and I must scream.

I’m confused by this statement. A neuron is a machine. A silicon chip computer is a machine. All they have done is interfaced two machines.

I imagine the point is not replacing hardware with neurons, but improving our ability to understand in vivo brains.

ahem

If you're in the US, you can buy human neurons online at sciencellonline.com/en/human-neurons/

There's a number of "immortal" human cell lines dating back to as far as the 1950s (you may have heard of Henrietta Lacks? [1] and the immortal HeLa cell line).

Today there are several immortalized neuron cell lines used in research to model neuronal function, like HeLa but of neuron type obviously, that are also typically derived from tumours (e.g., SH-SY5Y, PC12) or immortalized via genetic modification (e.g., v-myc) like CTX0E03 [2] which was designed to allow for continuous growth in the presence of particular reagents.

1. https://en.wikipedia.org/wiki/Henrietta_Lacks

2. https://www.prnewswire.com/news-releases/reneuron-announces-...

I think the Thought Emporium youtube channel has some explanatory [videos](https://www.youtube.com/watch?v=bEXefdbQDjw) of the whole process. I couldn't wrap my head around the thing tho.

They are lab grown.

You don't need to understand how neurons work in detail to be able to use them to do something. In the past, we were able to use electricity for various purposes without knowing about electrons.

I haven’t looked into it deeply either.

To my knowledge, we understand how an individual neuron works quite well. We just don’t really understand macro effects in large networks of neurons.

The video seems buzz wordy. Without looking into this too deeply, it seems like they’re using neurons individually or in small groups rather than creating a true “brain”. I would guess they’re using neurons or small groups of them sort of like transistors that do a single basic thing rather than a full “brain” that they just feed images to.

From their video it just comes across as they stimulate different left/right neurons depending on where the enemy is on screen and then listen to some output that also says left/right. Shooting looks completely random, to be frank.

If you connected electrodes to two different fish, shocked them and interpreted twitching as intelligent output, fish could also play Doom. The interface is doing all the work.

It doesn't sound like the neurons have any concept of the game other than "left input means left output", which is a rather trivial result... It's effectively no different than the pong example.

They don't say anything on how much training is required for this to happen, or if there's any "learning" going on at all. The learning part is "next".

When they answer back to us in personal pronouns, we will always be wondering if it is like LLM just putting most probable words together or something really sentient.

When someone makes a virtual girlfriend of it, is it really a disembodied person or just a smart answering machine?

A whole lot of ethical and psychological issues are to open up here.

Playing the devil's advocate: Why not use human neurons? Are they different to animal neurons and if they are, wouldn't that make it even more interesting?

The androids will dream of electric sheep.

I’m reminded of the brain in a jar robots from Fallout.

These are all largely unrelated technologies; how would they help us upload consciousness? We can't even map the human brain with complete accuracy.

Speaking of earlier concepts of The Matrix, there's an old 1973 German movie/mini series World on a Wire that's really good.

Wasn't the matrix using humans as some sort of power source/batteries? I may be misremembering though, as that seems pretty silly in retrospect ..