A dish of neurons may have taught itself to play Pong (badly)

Given control of a paddle and feedback, the neurons figured out what to do.

Read the whole story

 

[euzeka]

This feels both amazing and scary. Irrationally so in both cases, as there's not much there yet. It's a very interesting experiment regardless.

 

[r0twhylr]

Absolutely fascinating research!

I would love to see a clip of them playing, even if most of it was just random movement.

[jk]
Of course, in 15 years it will probably be able to shut itself in its room for 20 hour video game marathons, will have learned how to camp a spawn point, will laugh at the n00b neuron cluster petri dishes and insult their mom, and will only come out for more Monster energy drinks.
[/jk]

 

[dennis.kext]

I wonder if the could they use "brain drugs" as the reward? You hit the paddle and get a rush of endorphins?

Conversely can we find out what chemical signals are being sent?

 

[everythingallatonce]

This is just really cool. I wonder if research like this could help find ways to improve artificial neural networks, especially when it comes to power efficiency and training time.

 

[Jeff S]

While it's likely not the case that this dish of neurons has enough intelligence for this to potentially qualify as torture, I do worry about the ethics of this sort of experiment.

Presumably, the trend would be to grow ever larger and more complex DishBrain's, and at what point are you creating a living intelligence that you are experimenting on and possibly causing pain, anxiety, and distress?

 

[EvolutionaryBiochemist]

What seems the coolest about this from my perspective is that it provides a way to test how neurons do their stuff. The current paper provides a (preliminary, weak) test of the hypothesis that neurons learn using mismatches between predicted and real system state. It's very hard to distinguish various learning models in the complex context of a working brain. But, if developed further, this in vitro method seems like a very clever way overcome such intractable problems. By discovering which stimuli train the neurons in a dish, researchers can begin to unravel what stimuli train neurons in the brain.

 

[jbmadscience]

I wonder if the could they use "brain drugs" as the reward? You hit the paddle and get a rush of endorphins?

Conversely can we find out what chemical signals are being sent?

I think hormones and neurotransmitters operate on a higher level than this little dish of neurons. If individual neurons "wanted" endorphins, the brain would just be stimulated to produce lots of them rather than engaging any kind of conscious behavior.

I believe endorphins and other neurotransmitters like dopamine or serotonin, operate on the level of major functional circuits or structures in the brain. On a lower level we have GABA and glutamate, and a ton of other less well-known communication chemicals like TAAR1 ligands (trace amine associated receptor) which modulate how excitable nerve cells are, how they form connections, and other things.

I presume we could eventually learn enough to use these chemicals to guide the neural network's behavior, so your point isn't wrong at all. It's just that it's more of a mechanical process at this level, rather than something goal-drected.

I am not an expert so I could be wrong about any of the above, but it is fascinating if self-organizing neural networks try to minimize "surprise" signals!

edit: added missing word

 

[Fearknot]

Giving no feedback when a game ends with the ball crossing the end line produced performance that was intermediate between trained systems and control systems.

I'm not sure I understand this: how does the control system differ from a system that doesn't receive feedback?

 

[Dr. Jay]

Giving no feedback when a game ends with the ball crossing the end line produced performance that was intermediate between trained systems and control systems.

I'm not sure I understand this: how does the control system differ from a system that doesn't receive feedback?

It receives positive feedback while successfully playing, but not negative feedback when failing. That places it in between the no-feedback and positive + negative feedback options.

 

[SirMrManGuy]

Absolutely fascinating research!

I would love to see a clip of them playing, even if most of it was just random movement.

[jk]
Of course, in 15 years it will probably be able to shut itself in its room for 20 hour video game marathons, will have learned how to camp a spawn point, will laugh at the n00b neuron cluster petri dishes and insult their mom, and will only come out for more Monster energy drinks.
[/jk]

Be careful washing it's smelly goop filled socks, they might be sentient.

 

[VidasDuday]

I wonder if the could they use "brain drugs" as the reward? You hit the paddle and get a rush of endorphins?

Conversely can we find out what chemical signals are being sent?

I think hormones and neurotransmitters operate on a higher level than this little dish of neurons. If individual neurons "wanted" endorphins, the brain would just be stimulated to produce lots of them rather than engaging any kind of conscious behavior.

I believe endorphins and other neurotransmitters like dopamine or serotonin, operate on the level of major functional circuits or structures in the brain. On a lower level we have GABA and glutamate, and a ton of other less well-known communication chemicals like TAAR1 ligands (trace amine associated receptor) which modulate how excitable nerve cells are, how they form connections, and other things.

I presume we could eventually learn enough to use these chemicals to guide the neural network's behavior, so your point isn't wrong at all. It's just that it's more of a mechanical process at this level, rather than something goal-drected.

I am not an expert so I could be wrong about any of the above, but it is fascinating if self-organizing neural networks try to minimize "surprise" signals!

edit: added missing word

It would be interesting (to me, at least) to see those layers in an OSI 7-layer model type arrangement.

Neurons as the hardware layer I think.

Where would consciousness be?

 

[lewax00]

This sort of undermines the argument people make that playing games requires some skill.

How exactly did you make that Evel Knievel-scale leap?

 

[Sabrewings]

This sort of undermines the argument people make that playing games requires some skill.

How exactly did you make that Evel Knievel-scale leap?

The same way you, and about 10 other people made a superman level leap of logic about what I was trying to say, because you got it exactly backwards.

Surely all of us have heard someone say how playing a game doesn't require any skill. Maybe you say how it helps improve hand-eye coordination, and they scoff saying how it does so at such a low level it doesn't count. The specific topic isn't important. What is, is that now those people can say that a disembodied bit of brain matter can play games, so how much skill can it take for the rest of us? Sure, it's a strawman, and not even a particularly good strawman, but you know someone's going to make it, and it'll be someone like a parent or sibling. That means it will be in person and all the piss and vinegar of your typical keyboard warrior will be completely neutralized.

Wouldn't how well your brain adapts to new input/output responses be a part of "skill?" How else do you define it?

 

[jbmadscience]

I wonder if the could they use "brain drugs" as the reward? You hit the paddle and get a rush of endorphins?

Conversely can we find out what chemical signals are being sent?

I think hormones and neurotransmitters operate on a higher level than this little dish of neurons. If individual neurons "wanted" endorphins, the brain would just be stimulated to produce lots of them rather than engaging any kind of conscious behavior.

I believe endorphins and other neurotransmitters like dopamine or serotonin, operate on the level of major functional circuits or structures in the brain. On a lower level we have GABA and glutamate, and a ton of other less well-known communication chemicals like TAAR1 ligands (trace amine associated receptor) which modulate how excitable nerve cells are, how they form connections, and other things.

I presume we could eventually learn enough to use these chemicals to guide the neural network's behavior, so your point isn't wrong at all. It's just that it's more of a mechanical process at this level, rather than something goal-drected.

I am not an expert so I could be wrong about any of the above, but it is fascinating if self-organizing neural networks try to minimize "surprise" signals!

edit: added missing word

It would be interesting (to me, at least) to see those layers in an OSI 7-layer model type arrangement.

Neurons as the hardware layer I think.

Where would consciousness be?

That question gets deeper the longer I think about it.

I wonder how many layers we could even begin to describe? I think it shows how far we are from understanding the brain and consciousness. If we could model it in an OSI style, I think that would imply understanding at what point consciousness emerges from the layers below. And that would imply understanding what consciousness even is.

 

[Batmanuel]

I have no mouth, but I must play Pong.

 

[Person_Man]

I for one welcome our future new petri dish neuron overlords

 

[lewax00]

This sort of undermines the argument people make that playing games requires some skill.

How exactly did you make that Evel Knievel-scale leap?

The same way you, and about 10 other people made a superman level leap of logic about what I was trying to say, because you got it exactly backwards.

Surely all of us have heard someone say how playing a game doesn't require any skill. Maybe you say how it helps improve hand-eye coordination, and they scoff saying how it does so at such a low level it doesn't count. The specific topic isn't important. What is, is that now those people can say that a disembodied bit of brain matter can play games, so how much skill can it take for the rest of us? Sure, it's a strawman, and not even a particularly good strawman, but you know someone's going to make it, and it'll be someone like a parent or sibling. That means it will be in person and all the piss and vinegar of your typical keyboard warrior will be completely neutralized.

Wouldn't how well your brain adapts to new input/output responses be a part of "skill?" How else do you define it?

Yes, but you're trying to approach this with logic when the scenario involves people who are making a derisive comment from a position of supposed superiority.

Well if we're just randomly stringing together arguments for strawmen, no rational thought required, then obviously this study proves that brains exist for the purpose of playing video games.

 

[Ralf The Dog]

The thing that bothers me is, the punishment is too harsh. If you effectively erase the training for missing one shot, training can't last longer than a single ball. It would be cool if the level of randomization could be based on how long the ball was in play. Do quite a bit if it was a single ace. Do very little if it was a 256 bounce volley.

 

[l8gravely]

What seems the coolest about this from my perspective is that it provides a way to test how neurons do their stuff. The current paper provides a (preliminary, weak) test of the hypothesis that neurons learn using mismatches between predicted and real system state. It's very hard to distinguish various learning models in the complex context of a working brain. But, if developed further, this in vitro method seems like a very clever way overcome such intractable problems. By discovering which stimuli train the neurons in a dish, researchers can begin to unravel what stimuli train neurons in the brain.

And it might also lead to a deeper understanding of *how* neurons actually work, and what the second and third order effects are, which would lead back into simulations of neurons getting better. And then we'll have neuron compilers, optimizers and linkers and "hello world" in neuron programming.

But seriously, this is really neat. Just think about how things would change if they used even finer electrodes for even higher spacial distribution. Just think of how many neurons we have in our brains, and the sheer amount of data they get from our eyes. Or skin. Or taste. And how they then filter and process it all.

This is just super interesting stuff.

 

[Ralf The Dog]

What seems the coolest about this from my perspective is that it provides a way to test how neurons do their stuff. The current paper provides a (preliminary, weak) test of the hypothesis that neurons learn using mismatches between predicted and real system state. It's very hard to distinguish various learning models in the complex context of a working brain. But, if developed further, this in vitro method seems like a very clever way overcome such intractable problems. By discovering which stimuli train the neurons in a dish, researchers can begin to unravel what stimuli train neurons in the brain.

And it might also lead to a deeper understanding of *how* neurons actually work, and what the second and third order effects are, which would lead back into simulations of neurons getting better. And then we'll have neuron compilers, optimizers and linkers and "hello world" in neuron programming.

But seriously, this is really neat. Just think about how things would change if they used even finer electrodes for even higher spacial distribution. Just think of how many neurons we have in our brains, and the sheer amount of data they get from our eyes. Or skin. Or taste. And how they then filter and process it all.

This is just super interesting stuff.

I suspect we have a decent understanding of how isolated neurons work. The interesting question is what kind of emergent behavior happens when you let them congregate and form neuron gangs?

 

[mdrejhon]

The art of improving AI…
…colliding with the the art of simplifying down neurons to a petri dish.

In the endeavour of understanding “What is Intelligence?” down to a basic nanoscopic level.

It helps humans slowly better understand how human brains came to be.

Amazing science to see this convergence.

 

[Veritas super omens]

Played pong (badly)? So more of an Abbie something brain, and not Hans Delbruck's.

 

[Veritas super omens]

In an alternate timeline this leads eventually to gel-pack processors in starships like Voyager. In this one, it leads Musk to put it in charge of Twitter...and, it doesn't end well....

 

[aerogems]

Played pong (badly)? So more of an Abbie something brain, and not Hans Delbruck's.

But what was the density of the neurons relative to say a human brain at different ages? The article makes it sound like it's your standard petri dish, so like maybe 6 inches in diameter and a half inch or so thick. Even assuming the thing were completely filled with neurons, that's like maybe the amount of brain matter for a raccoon or something.

And of course you know what this means... we were all missing out on prime gaming time in the womb! We all need to have a serious discussion with our mothers about depriving us of this opportunity!

 

[aerogems]

In an alternate timeline this leads eventually to gel-pack processors in starships like Voyager. In this one, it leads Musk to put it in charge of Twitter...and, it doesn't end well....

Can't it be both? Musk fucks off to Mars, takes Twitter with him, and then like Voyager gets blasted to some far flung part of the galaxy and is never heard from again. Or he crashes on Mars and is never heard from again. The important part is he's never heard from again.

 

[Baumi]

This sort of undermines the argument people make that playing games requires some skill.

How exactly did you make that Evel Knievel-scale leap?

The same way you, and about 10 other people made a superman level leap of logic about what I was trying to say, because you got it exactly backwards.

Surely all of us have heard someone say how playing a game doesn't require any skill. Maybe you say how it helps improve hand-eye coordination, and they scoff saying how it does so at such a low level it doesn't count. The specific topic isn't important. What is, is that now those people can say that a disembodied bit of brain matter can play games, so how much skill can it take for the rest of us? Sure, it's a strawman, and not even a particularly good strawman, but you know someone's going to make it, and it'll be someone like a parent or sibling. That means it will be in person and all the piss and vinegar of your typical keyboard warrior will be completely neutralized.

First of all, if the overwhelming majority of people doesn't get the intended meaning of a comment, it might well be the case that it just wasn't stated very well.

Aside from that, the whole thing seems like a bit of a strange exercise to me: You've made up a strawman ("someone like a parent or sibling") that makes a flawed argument that you (and most likely everyone else in here) disagree with. But when someone tries to point out specific flaws in that flawed argument, you rebuff them by saying that the imaginary person making it isn't receptive to logic.

If you don't mind the question: What's the point?

 

[aerogems]

This sort of undermines the argument people make that playing games requires some skill.

How exactly did you make that Evel Knievel-scale leap?

The same way you, and about 10 other people made a superman level leap of logic about what I was trying to say, because you got it exactly backwards.

Surely all of us have heard someone say how playing a game doesn't require any skill. Maybe you say how it helps improve hand-eye coordination, and they scoff saying how it does so at such a low level it doesn't count. The specific topic isn't important. What is, is that now those people can say that a disembodied bit of brain matter can play games, so how much skill can it take for the rest of us? Sure, it's a strawman, and not even a particularly good strawman, but you know someone's going to make it, and it'll be someone like a parent or sibling. That means it will be in person and all the piss and vinegar of your typical keyboard warrior will be completely neutralized.

First of all, if the overwhelming majority of people doesn't get the intended meaning of a comment, it might well be the case that it just wasn't stated very well.

Aside from that, the whole thing seems like a bit of a strange exercise to me: You've made up a strawman ("someone like a parent or sibling") that makes a flawed argument that you (and most likely everyone else in here) disagree with. But when someone tries to point out specific flaws in that flawed argument, you rebuff them by saying that the imaginary person making it isn't receptive to logic.

If you don't mind the question: What's the point?

First part's fair. I wrote it in a hurry because I needed to get back to work. My 90 degree commute is brutal!

The point is, it was a joke. I've had that exact conversation, and if I've had that conversation odds are everyone here has had some version of the same conversation. My delivery may have been lacking, such are the perils of a dry sense of humor and written mediums, but I guess I just assumed people would see an absurd comment and think, "Oh, that must be a joke! Even if I didn't think it was particularly funny, the relative absurdity suggests that it was a comment made in jest and I should treat it as such until I have reason to believe the OP was serious."

 

[Baumi]

My delivery may have been lacking, such are the perils of a dry sense of humor and written mediums, but I guess I just assumed people would see an absurd comment and think, "Oh, that must be a joke! Even if I didn't think it was particularly funny, the relative absurdity suggests that it was a comment made in jest and I should treat it as such until I have reason to believe the OP was serious."

Fair enough.

Unfortunately, the general insanity of the past few years likely only served to strengthen Poe's Law.

 

[GFKBill]

This sort of undermines the argument people make that playing games requires some skill.

How exactly did you make that Evel Knievel-scale leap?

The same way you, and about 10 other people made a superman level leap of logic about what I was trying to say, because you got it exactly backwards.

Surely all of us have heard someone say how playing a game doesn't require any skill. Maybe you say how it helps improve hand-eye coordination, and they scoff saying how it does so at such a low level it doesn't count. The specific topic isn't important. What is, is that now those people can say that a disembodied bit of brain matter can play games, so how much skill can it take for the rest of us? Sure, it's a strawman, and not even a particularly good strawman, but you know someone's going to make it, and it'll be someone like a parent or sibling. That means it will be in person and all the piss and vinegar of your typical keyboard warrior will be completely neutralized.

First of all, if the overwhelming majority of people doesn't get the intended meaning of a comment, it might well be the case that it just wasn't stated very well.

Aside from that, the whole thing seems like a bit of a strange exercise to me: You've made up a strawman ("someone like a parent or sibling") that makes a flawed argument that you (and most likely everyone else in here) disagree with. But when someone tries to point out specific flaws in that flawed argument, you rebuff them by saying that the imaginary person making it isn't receptive to logic.

If you don't mind the question: What's the point?

First part's fair. I wrote it in a hurry because I needed to get back to work. My 90 degree commute is brutal!

The point is, it was a joke. I've had that exact conversation, and if I've had that conversation odds are everyone here has had some version of the same conversation. My delivery may have been lacking, such are the perils of a dry sense of humor and written mediums, but I guess I just assumed people would see an absurd comment and think, "Oh, that must be a joke! Even if I didn't think it was particularly funny, the relative absurdity suggests that it was a comment made in jest and I should treat it as such until I have reason to believe the OP was serious."

I've had it asserted that game playing is a waste of time, but don't think I've ever had anyone claim it doesn't require any skill.

 

[azazel1024]

While it's likely not the case that this dish of neurons has enough intelligence for this to potentially qualify as torture, I do worry about the ethics of this sort of experiment.

Presumably, the trend would be to grow ever larger and more complex DishBrain's, and at what point are you creating a living intelligence that you are experimenting on and possibly causing pain, anxiety, and distress?

In the other direction, when might an artificial intelligence qualify for the same. That isn’t to diminish what you are concerned over. I agree with your concern for organic and inorganic intelligence research (and yes I realize even the most universal artificial intelligence engines are not remotely intelligent in a living thing way. Yet).

 

[acefsw]

In an alternate timeline this leads eventually to gel-pack processors in starships like Voyager. In this one, it leads Musk to put it in charge of Twitter...and, it doesn't end well....

While reading the article the bio-neural gel packs of VOY were one of the first things that came to my mind. I always thought they were an interesting concept.

Anyway, in the real world, this is absolutely fascinating and hopefully further experiments will help us to better understand the human brain and possibly contribute to the field of artificial neural networks for computing.

 

[DavidEmami]

How many quatloos were wagered?

 

[YetAnotherBoris]

In an alternate timeline this leads eventually to gel-pack processors in starships like Voyager. In this one, it leads Musk to put it in charge of Twitter...and, it doesn't end well....

More likely, Musk puts them into self-driving cars, which then in the fullness of time figure out how to end their own misery, by smashing into things at high speed... or else, we end up with real-life enactments of that classic horror movie trope where pedestrians are being hunted by serial-killing homicidal sentient vehicles.

 

[Longmile149]

*cut to X years in the future as an Atlas sits telling ghost stories to a group of Spots around a camp fire*

“Some of them even dumped living brains into bowls and made them play Pong!”

 

[fcrary]

In an alternate timeline this leads eventually to gel-pack processors in starships like Voyager. In this one, it leads Musk to put it in charge of Twitter...and, it doesn't end well....

More likely, Musk puts them into self-driving cars, which then in the fullness of time figure out how to end their own misery, by smashing into things at high speed... or else, we end up with real-life enactments of that classic horror movie trope where pedestrians are being hunted by serial-killing homicidal sentient vehicles.

Could we try to stay vaguely on topic? I know people have strong feelings about Mr. Musk, but he's basically irrelevant to this article and we're still on the first page of comments.