A Worm's Mind In a Lego Body

mikejuk writes The nematode worm Caenorhabditis elegans (C. elegans) is tiny and only has 302 neurons. These have been completely mapped, and one of the founders of the OpenWorm project, Timothy Busbice, has taken the connectome and implemented an object oriented neuron program. The neurons communicate by sending UDP packets across the network. The software works with sensors and effectors provided by a simple LEGO robot. The sensors are sampled every 100ms. For example, the sonar sensor on the robot is wired as the worm's nose. If anything comes within 20cm of the 'nose' then UDP packets are sent to the sensory neurons in the network. The motor neurons are wired up to the left and right motors of the robot. It is claimed that the robot behaved in ways that are similar to observed C. elegans. Stimulation of the nose stopped forward motion. Touching the anterior and posterior touch sensors made the robot move forward and back accordingly. Stimulating the food sensor made the robot move forward. The key point is that there was no programming or learning involved to create the behaviors. The connectome of the worm was mapped and implemented as a software system and the behaviors emerge. Is the robot a C. elegans in a different body or is it something quite new? Is it alive? These are questions for philosophers, but it does suggest that the ghost in the machine is just the machine. The important question is does it scale?

Comments

Put the glasses on, stupid.

[ColdWetDog]

Initially read it as "A Woman's Mind in a Lego Body". Wasn't quite sure where to go from there so I squinted a little bit. Fortunately Timothy saved me from having to explain to my wife just what 'that stupid Slashdot article" is about.

Re:Put the glasses on, stupid.

[ColdWetDog]

Actually, that's pretty cool. The bot goes back and forth, kinda like a real worm. It would be interesting to scale this behavior up to several thousand 'neurons' (I'm sure somebody is going to try).

Re:Put the glasses on, stupid.

[whereiswaldo]

What else would be interesting: let this "worm" mate and see how it offspring adapt to their new body.

Re:Put the glasses on, stupid.

[Anonymous Coward]

ah but then it begins to eat harvesters.

Re:Put the glasses on, stupid.

[Anonymous Coward]

As the article mentions, this isn't too interesting to AI developers. We already know how neural networks work and some are turning complete so they can do anything. What we aren't good at is designing them. Add a connection here or there, set the weight to .000803 or .0040075, switching to pulsating, or whatever. We don't know. Instead we run thousands upon thousands of simulations that use other AI algorithms to make the networks for us.

We haven't scaled up to human levels because there's so much more to complex brains. There's some sort of cross talk with chemicals, other chemicals coating neurons to make them fire differently, neurons growing together or apart, cells dying, new cells emerging, etc... Now maybe all that's not needed, good enough is fine for evolution, but were not at that level yet.

There are human-level brain simulations being worked on, but I haven't been following them closely. I don't think they're implementing everything. Actually, I know they aren't because we keep discovering new things. Are they working off and standard model of the human brain or a specific person's brain?

It would be more ground breaking if someone did the reverse. Engineer a neural network then grow it into another animal. That would be new, but due to the nature of neural networks, we also already know it would work.

Re:Of course the ghost is just the machine

[Livius]

Superstition comes from the instinctive default assumption that unexplained things are animate things out to get you.

The false positives are a nuisance, but living on the savanna without modern science it was sometimes the safe assumption.

No programming?

[tomhath]

The key point is that there was no programming or learning involved to create the behaviors.

Yes, there was. The behaviors didn't just "emerge", they're coded into the robot.

Re:No programming?

[93 Escort Wagon]

Yeah, that silly statement in the summary stood out like a sore thumb.

(As does my bad metaphor)

Re:No programming?

[Kjella]

If you call copy-paste programming. They took an "executable", dumped it from the worm's brain, put it in a robot and found it acts like a worm. The behavior emerged through evolution and was encoded in the neurons by nature, not the researchers. If you could dump a human brain, put it in a robot and have it act like a human without ever "reverse engineering" it that would be most impressive.

Re:No programming?

[teslar]

The behavior emerged through evolution and was encoded in the neurons by nature

What has been implemented in this robot has nothing to do with biological neurons of C. elegans.

The robot uses integrate-and-fire neurons. The "signal" sent from pre- to postsynaptic neuron is an integer equal to the number of connection between the neurons in the real worm. If the sum of input exceeds a threshold, the neuron "fires" (sidenote: right here's a bit of programming: how did the threshold values get chosen?).

C. elegans neurons do not "fire" (they are not spiking neurons and lack Na+ channels) but use calcium-based analog signals.

The body does matter too. C elegans has muscles on either side that it contracts alternately to move in a sinusoidal fashion. Not wheels. C elegans locomotion does not work like wheeled locomotion.

So, yes, you are right, C elegans neurons encode behaviour appropriate for a C elegans body given the biology of the neurons available here. None of this, however, makes it into this robot. An abstraction of the connectome does (C elegans has both electrical and chemical synapses; that distinction seems to be lost here too) and that's it.

It is kinda cool that the connectome does seem to naturally implement some basic response patterns; but given that muscles have been replaced by wheels, I'm not sure how meaningful that actually is.

Re:No programming?

[ashshy]

If you call copy-paste programming. They took an "executable", dumped it from the worm's brain, put it in a robot and found it acts like a worm. The behavior emerged through evolution and was encoded in the neurons by nature, not the researchers. If you could dump a human brain, put it in a robot and have it act like a human without ever "reverse engineering" it that would be most impressive.

All of this is true, but the inputs and outputs still have to be mapped to the appropriate endpoints. Unless, of course, mapping them at random still produces the perfect Lego/worm beast after a little bit of real-world action. The article doesn't talk about this, so I'm assuming the sensors and effectors were hooked up to the proper Lego tools by hand.

Which, in my book, counts as programming.

Re:No programming?

[Tyr07]

Actually they did just emerge.

You see, these neurons were wired in the living worm to send signals to different parts of the body. A signal from its nose, in the real worm, would go to a neuron, which would fire other neurons, that caused it to stop moving. This happened by sending a signal to other neurons which then sent them down to parts of the body.

What they did, was use UDP packets to be the 'signal', and then sent electrical currents down to the robotic corresponding parts as the original living worm did, which resulted in the same behavior.

They didn't put a chip in where it goes if range 20cm stop. They mimicked the worm and said if range 20cm send signal to neuron C. Neuron C in the worm and robot goes if receive signal, send electrical pulse to wire C. Wire C causes it to stop motion.

So behaviors emerged, were not "programmed" so to speak.

Does it scale?

[Anonymous Coward]

Try Duplo.

Memory mapping?

[Kevin Fishburne]

Emulating the connectivity and functionality of neurons is pretty awesome, but it would seem the next logical step would be to map and interpret how memories are stored and processed, as well as organ feedback (skin, smell, glands). What's really interesting about this is that it shows, at least to some degree, that a simple brain can be reproduced using mathematical relationships (programming) and "run" with a I/O feedback loop. As far as the philosophical stuff, I think eventually we'll be forced to accept that life is a type of machine and that the "ghost" is an illusion emerging from its complexity. Other than better neuroscience, the main thing holding us back is pride.

Re:Memory mapping?

[Kjella]

Emulating the connectivity and functionality of neurons is pretty awesome, but it would seem the next logical step would be to map and interpret how memories are stored and processed,

We actually have a fairly good clue on how the brain stores information chemically, but that's all but useless without understanding the neurons because they're the ones that disperse a memory during storage and gather all the sensory clues to trigger semantic meaning like recognizing a person's voice as well as all the associations related to that person during retrieval. It's not like computers with a storage unit, all neurons can store information and it also modifies their behavior so the memory and path to the memory is integrated and extremely multi-path, you can read a person's name or see their photo or smell their perfume and it all triggers the same memory.

In particular it seems we have two very different kinds of associations, one that tries to join same with same like how one person looks similar to somebody else, the other hooking up disjoint information that this name belongs to this face and the former seems to go by brain centers so we get these nice macro maps of what happens where. I guess that's great for those trying to create machine vision or something like that, but for AI it's the links between the sights, sound, smells, tactile and semantic information that matter and you don't understand those without understanding the micro scale, what hooks those two particular pieces of information together.

Re:Memory mapping?

[Kevin Fishburne]

Call me when you show non-biological free will. Emulation of deterministic life processes is interesting, but it's free will that needs to be demonstrated in silicon.

Life is extremely efficient, from the micro to the macro scale. To attempt to recreate even a simple organism using current technology (including a purely logical recreation in silicon) would be like building a modern supercomputer out rocks and sticks. When you speak of "free will" being recreated, you've pretty much chosen the highest possible level of what we'd consider a property of advanced life. What excited me about the article is that it suggests instead of tackling the mountain it may be more fruitful to attack a single grain of sand first. Perhaps once we understand a grain of sand, we can start working our way up to the higher and more complex relationships and functionality.

For example, rather than trying to create AI using programming, try reverse-engineering a single-celled organism's molecular composition and chemical processes. If that can be understood completely it provides a starting point for how to reproduce and modify it. Being able to "run" a bacterium in a simulated environment, and later being able to create one physically, is the first step toward truly understand how life works as a machine. Until we have that kind of understanding, the idea of creating real intelligence or artificial life will be confined to cheap imitations which work nothing like the real thing. If we don't understand how a human works as a massive ongoing chemical reaction, we have zero chance of creating one out of gears and silicon.

Re:Memory mapping?

[WillAdams]

More importantly, the Bible tells those who believe in it that nothing is unknowable: Genesis 11:6

>The Lord said, “If as one people speaking the same language
>they have begun to do this, then nothing they plan to do
>will be impossible for them.

So it's blasphemous for Christians (or Jews or Muslims) to say that humanity can't understand such things (or anything).

Cylon worms ?

[Anonymous Coward]

So the creator of Battlestar Galactica dies, and we learn that people are building LEGO cylon worms. Interesting...

Accelerando IRL

[l0n3s0m3phr34k]

This is the first step to the "cat chasing a mouse" AI in the Charles Stross's book Accelerando. They programmed the AI to see the missile's target as a mouse so it would chase it. We're just a few steps away from this.

Despite Elon Musk's recent anti-AI ranting (which does have truth too it), we'll get our flying cars once we can implement a "bird-based" AI to fly it for us. The more we replicate nature in our tech the further we'll get. I predict we'll see "emergent features" such as social hierarchies, empathy, emotions, and such in our tech the more neurons we add without even really needing to program it on purpose.

Re:Accelerando IRL

[l0n3s0m3phr34k]

Forgot to mention the research on the stomatogastric nervous system heavily relate to this too...

They've made something that mimics C. elegans

[Beeftopia]

It's fascinating but it's not C. elegans. It doesn't reproduce. It doesn't die. It's not alive.

The sensors are implemented in large, electro-mechanical hardware. Not biochemical systems. It has no telomeres. No cells.

Humans have several subsystems: digestive, endocrine, pulmonary (pneumatic and hydraulic), muscular, skeletal, nervous. If they manage to create an electro-mechanical system to mimic the nervous subsystem, it's just that - mimicking the subsystem. It would be an amazing feat, and what's been done here is fascinating, but we're still quite some distance away from stating that a human - or C. elegans - is 2^n nand gates.

Is something that mimics a nervous subsystem via an electro-mechanical system equivalent to the nervous system? Be it the 302 neurons of the C. elegans or the approximately 100 billion of the H. sapiens? It might become very intelligent... more intelligent than us... and then we'd have a problem... Frankenstein didn't appreciate being locked in his form...

Would it really feel emotions? Pain, rage, joy, fear, ennui? Or is it just mimicking them?

Fascinating stuff.

This isn't about technological developments,

[Shadow of Eternity]

it's about moral ones. If we make a perfectly simulated animal brain and it works just like the real thing does that mean we've made an animal? Do we consider that animal to be alive? Does it have less "worth" than a flesh and blood creature? Better that we answer these questions now than when we have robots asking us if they have a soul.

Re:This isn't about technological developments,

[radtea]

If we make a perfectly simulated animal brain and it works just like the real thing does that mean we've made an animal?

Does it taste good? If not, you haven't made a real animal.

There is nothing deep or even particularly interesting about these questions, and just how stupid their breathless idiocy is can be seen by asking, "Does the newly created entity lack almost every interesting property of the entity some philosophy-addled idiot thinks we should 'wonder' if it is absolutely identical to in every respect?" The answer is always, trivially, "No."

So only an extremely stupid person or a shill trying to market something (fake wisdom?) would ask such an idiotic question.

There are more reasonable questions that people who are neither idiots nor philosophers (but I repeat myself) are reasonably well-equipped to answer. Like this: lacking anything remotely resembling neurochemistry, is it appropriate for us to impute to this model any of the effects of neurochemistry that may or may not be lumped into the neuron behaviour? Since we're pretty sure neurochemistry is independent of network architecture, it would be incredibly stupid to identify the entirety of the robot's responses with the neuronal architecture, rather than the neurochemical environment they behave in?

For example, if you starve a worm its behaviour changes because its neurochemistry changes. Hormone levels, cortisol levels (or their worm equivalents) change, and that changes behaviour, in some cases quite dramatically. So what happens to the robot when you starve it? And if you can't starve it, why do you think it is in any way identical to a worm, rather than just an interesting simulation of part of it?

And of course, simply because we can imagine a more complete model of a worm doesn't mean we can build one that is sufficiently similar to a worm in all respects to make any of these questions interesting. It would have to eat and excrete and so on. It would have to have environmental sensitivities. And imagining those things aren't important is stupid: what we imagine is not relevant to what is real. There is no basis for saying a mechanical worm is a "real" worm (what would an "unreal" worm be?) It is a "real" mechanical worm. You still can't eat it, so it isn't a worm. Saying certain properties "don't count" is pure magical thinking, unworthy of scientists.

Re:This isn't about technological developments,

[linuxrocks123]

Instead of calling everyone around you an idiot, why don't you read the question again and consider again what is being asked.

Unless you have absolutely no ethical qualms about what Dr. Mengele did to his experimental subjects, the ethical questions raised by emulating a complete human brain are in no way trivial and in no way unimportant. Right now, we reformat computers, turn them off, turn them on, and don't and don't have to care at all about what they "want" or about treating them with any kind of respect. If we successfully simulate a human brain to the point where it can "think" and has humanlike "emotions", deleting that neural net file might be fairly considered murder. No, really. If you can talk to the thing and it can talk back, and it looks, talks, and acts like a human ... it's a duck. Sorry, human.

Now, we are nowhere near having that capability. We don't have to worry about that question now. But it's a very interesting question to think about, because thinking about it can grant insights into what it means for something to be sentient or human in the first place.

Re:This isn't about technological developments,

[roman_mir]

What do you mean 'prove it'? Wrong, you have to prove that such a thing is even a remote possibility, I have to prove shit, absolutely nothing, nada, zilch. There is no soul. I don't have to prove anything because it is an extraordinary claim to make that there is a soul and so those who make extraordinary claims have to come up with all the proof in the world to back those up.

Re:This isn't about technological developments,

[narcc]

What do you mean 'prove it'?

You made a positive claim. See:

The answer is no, you don't have a soul, there is no such thing as a soul.

Remember: you're talking about knowledge here, not belief, after all. Learn the difference.

you have to prove that such a thing is even a remote possibility

The only claim I made was "we don't know" which is true. We don't know.

it is an extraordinary claim to make that there is a soul

Sure. Did you miss the part where I never made such a claim?

those who make extraordinary claims have to come up with all the proof in the world to back those up.

"All the proof in the world" What does that even mean?

Sigh... I really wish the science cheerleaders with no actual scientific background would go away. They're dangerous.

An interesting specimen

[Sarten-X]

I first learned about C. elegans while researching simple neural systems. There's a nice map of the neural connections available. Today, I stumbled across the name again, when Wikipedia informed me that Caenorhabditis elegans is the most primitive animal that sleeps. Now I find that there's a robot worm that I'd consider to be alive.

This guy's pretty awesome.

Re:An interesting specimen

[LeadSongDog]

Silly, everyone knows that the dietary fancy of C. elegans is E. coli strain OP50, so that should probably be "Do Caendroids Dream of Electric Germs?" (further reading at https://www.ncbi.nlm.nih.gov/p...)

Scaling

[abe ferlman]

Imagine a beowulf cluster of those.

Important question.

[fahrbot-bot]

The important question is does it scale?

No. The important question is does it run Linux? It's a given that it runs NetBSD - sure, my toaster does.

Not the Functionality of a Neuron

[raftpeople]

They did not emulate the functionality of a neuron. If you read up on the subject you will find that the neuron is a network all by itself with spikes moving forward and backward, local spikes on the dendrites, the dendritic tree performing multiple simultaneous linear and non-linear computations, etc. etc. etc. They used an extremely simplistic formula that completely skips over these computations that have been shown to be very important for the proper functioning of the neuron.

Re:Not the Functionality of a Neuron

[raftpeople]

Just one example of many of the need to properly model the dendritic computation is the hippocampal CA1 pyramidal neurons performing 2 different computations depending on the spatial distribution of the incoming signal (and level of synchrony). Either it performs input strength encoding or feature detection. You can't ignore this and end up computing the same thing, your end result will be different.

Re:Stupid Questions

[roman_mir]

Naming conventions are what they are based on historical precedent, nothing else. If we devise a machine that can do all the things that many other living creatures can do (probably procreate, grow, learn, feed to sustain itself) under normal circumstances (excluding edge cases that we can compare things to, like people in coma who are still alive but cannot do many things that normal people not in coma can do), then there is no difference between that machine and another living creature. However we kill living creatures on daily basis, hundreds of millions of them, most large ones are killed to eat, the invisible ones are killed because we don't care and we have to do what we have to do in life (sterilise stuff, burn stuff, whatever).

So the reality is that none of these questions matter, we are the ones in charge and as long as we can stay in charge such questions will only be a curiosity that our minds are capable of engaging into, but they won't stop us from using our inventions in whichever way we see fit.

Re:Of course it scales

[frank_adrian314159]

Everyone knows it's Lisp all the way down until you get the the atoms, then it's this weird probabilistic stuff.

Re:UDP packets?

[friesofdoom]

Ummm,

"The model is accurate in its connections and makes use of UDP packets to fire neurons. If two neurons have three synaptic connections then when the first neuron fires a UDP packet is sent to the second neuron with the payload "3". The neurons are addressed by IP and port number."

My initial comment is 100% valid in that context. The overhead for the UDP communication between neurons is just ridiculous, why would anyone in their right mind try and prototype a neural network in this way unless they have NO idea about what they are doing? This could be done in a thousand other ways (GPIO between simple micros if you want your neurons to be discrete) rather than getting some crazy network protocol working and slowing the whole simulation down...
OR
get this... you could get any 90's era calculator to simulate a 302 neuron network with out any problems.