Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
Whole-brain connectome of the fruit fly (flywire.ai)
190 points by jacobedawson on July 4, 2023 | hide | past | favorite | 94 comments


Related:

* Fruit Fly Brain Observatory https://www.fruitflybrain.org/#/posts/explore_ffbo which lets you visualize and explore datasets

* Neurokernel http://neurokernel.github.io/ - open source software which "aims to build an open software platform for the emulation of the entire brain of the fruit fly Drosophila melanogaster on multiple Graphics Processing Units (GPUs)." See also: http://www.bionet.ee.columbia.edu/projects/neurokernel

The recent work being published on this is pretty wild:

* https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7614541/ (summaries: https://www.spectrumnews.org/news/wiring-map-reveals-how-lar... and https://www.nih.gov/news-events/nih-research-matters/complet...)

Oh, and just in case people haven't read this, I'll just drop this short story from a few years ago Lena (MMacevedo): https://qntm.org/mmacevedo - like with AI, I think science fact will be rapidly catching up w/ fiction soon.


That paper is incredible. It’s crazy that they found 41% of all DN connections receive recurrent feedback from a downstream neuron.

Some things that may be commonplace understanding in the neuroscience community but that I found interesting:

- speculation that deep recurrence in the learning center is a mechanism for working memory, and allows for multiple high-level cognitive processes to occur simultaneously

- the description of how a variety of neurons in the learning center categorize stimuli, a different group controls the learned value of inputs (valence?), and then another group integrates the valences of both the learned and innate neuron groups for the given stimulus category

Oh, and that most of the neurons were engaged in multi-modal activity


Is it now possible to accurately model the behavior of a C. elegans from its neural connectome?


Nowhere near, and its connectome has been known for decades.


Very interesting!

A question from a layman: what is a “signal” in the context of a neuron? Is it discrete binary ON/OFF, or is it discrete with multiple levels, or is it fully analog? Do we know the “encoding” of the signals? What do they represent?


My understanding was that it is an analog signal of current. The voltage of which spikes when the Neuron fires. The signal travels along the dendrites and may boost or deactivate other neurons.


Visual processing requires an astounding amount of brain power. Over 50% of a fruit flies brain cells are dedicated to their rudimentary visual system.

Comparatively, the human visual processing system is truly astonishing and I am grateful for it.


Our visual system is basically a 3D game engine beyond anything we can currently program.

Takes raw input and builds up a 3D world of objects and color. We're not looking through a window into reality, but building up a version of it inside our heads.


As I age, it feels like the level of detail algorithm and optimization code are due for some updates.

Skimming a text varies from high to moderate fidelity some days.


They’re doing a more complex job than a self driving car, with a pinhead sized brain, miniature sensors, and less power than a set of AirPods.


I can't help but think about a fly buzzing against a window as I read this.

I don't think the complexity of the tasks are very comparable.


If you put a self driving car in an alien place (like roads with thick glass panels mounted on them) it wouldn't get very far either.

In fact, put a self driving car inside an actual jungle and it would do worse than a fruit-fly inside one of our "concrete" jungles.


this is because fly optimized to rely on its sense of smell first. but it can't use this to find way leave.


Battery dies in the same amount of time so I wouldn’t consider this a technical feat.


Can I pair this with adversarial deep learning to develop a way to trap these in my kitchen? I live in the Midwest and they are like a plague of locusts but tiny and annoying and strangely aware of countermeasures.

Or maybe I can apply fruit fly Roko’s Basilisk and threaten to simulate fruit fly brains in agony if the flies don’t leave me alone on the theory that since I am able to simulate a fruit fly brain I am fly basilisk?


Unfortunately, there's a good chance you have fungus gnats and not fruit flies. Do they live in houseplants? In my experience fruit flies go away if you rinse cans/bottles, and keep the sinks/drain clean, and use a lid for your compost.


I had some sort of gnat living in the soil of my plants. After I subjectively tried every chemical known to mankind, I got rid of them by ordering a package of egg- or larvae-eating nematodes over the net, mixing them with water and sprinkling them in. Instant success.


The nematodes have helped me, but not eliminated them. I have a lot of plants.


Nah it’s definitely fruit flies. Keeping things clean helps but we have kids and it’s hard to keep things clean enough to keep them from coming in. They’re not breeding indoors but they get in frequently and then annoy you.


It’s all fun and games until Flyezer Yudkowsky gives podcast interviews calling for missile strikes on your GPU cluster


Didn't quite catch, is this just the first mapping, or are they able to 'execute it'. Like the worm project a few years ago. Where they mapped worm "C. elegans" brain, and was able to put in a simulated environment and watch reactions, and eventually into a robot body.

This seems like next step up in complexity, so wondering how far they are along having the 'brain map' execute.


Whole Brain Emulation: No Progress on C. elegans After 10 Years

https://www.lesswrong.com/posts/mHqQxwKuzZS69CXX5/whole-brai...


I'd highlight my comment there on recent progress since that was posted back in 2021: https://www.lesswrong.com/posts/mHqQxwKuzZS69CXX5/whole-brai...


Wow very interesting.

They say the holdup is that they lack a way to read the weights and thresholds of the neurons.

So they have the network structure, but not the weights. It says that there isn’t enough interest in funding research on a method to read the weights either. That’s really surprising. This method could unlock some crazy cool science.


It seems like with the current fire around AI, that making a worm or fly brain function in a simulated environment would be a hot bed of research. I wonder if the research is hot, but private, like at OpenAI, or Neuralink.


They started publishing again this year! They’ve isolated the neural net and published it as a “liquid neural network”. Check it out, it outperforms CNNs for niche tasks with a fraction of the resources.


Sources?



Awesome! Are there any simple downloads that have just annotations and connectivity? Like a sparse adjacency matrix with annotations of the node and edge features (neuron type, etc). I didn't see any immediately. A condensed representation of the 3d model should be pretty great for researchers as well.


Yes, if you sign up in the Downloads sections (found in the header menu), there are individual downloads for neurons and connections (and other parts that make up the whole thing).


I think logical and incremental next step is the whole brain connectome of business analysts at bank of america technology division.


What an awful, roundabout way to say “drosophila.” Article says it’s already done.


Would instantiating this network and feeding it random inputs be akin to animal cruelty?

Would you run tests on a cat connectome? A human’s? Yours?


Hardly so. A bunch of matrix multipliers is a fully deterministic system, a mechanism in other words. There is nothing in it that can experience pain. Using a truly random RNG device will simply "shake" this mechanism randomly. However with a feedback loop from the mechanism back to the RNG device things become murky, for now we can say that the radioactive rock behind the RNG might experience something, it may even feel like a fruit fly.


What? Why would randomness matter to this?

If, hypothetically, it were possible to do a QFT simulation of an entire human, but where no true randomness was used in the simulation program, only a cryptographically secure (relative to an adversary with even more compute than the beyond-astronomical amount of compute used by the simulation) random number generator (and therefore fully deterministic),

you think that that would have less of a claim to having internal experience, than a simulation of a fruit fly brain connected to a true RNG?

This seems implausible to me.


The real answer is that the scientific method is based on observation so it is incapable of examining what lies behind those observations, consciousness. We do not have the scientific tools to know if a simulated brain experiences things. We dont know if protons experience things, or rocks, or monkeys, or even individual people.


That's right, it would have no internal experience. A giant deterministic calculator is just a mechanism that gives predictable outputs. It doesn't gain experience simply because it's superficially similar to a fruit fly or because it has so many transistors that we cannot count them.

If accurate simulation of a quatum system was possible within a deterministic 01 machine, we would have to declare that this is a dead world: a giant calculator that's moving gears in a deterministic manner.

A fruit fly does have internal experience, but its brains are based on a continuous system. With a deterministic machine we can approximate that continuous system with arbitrary, but finite, accuracy, and that approximation will be a dead calculator. Our computing paradigm - a deterministic turing machine - is simply unfit for replicating a continuous system, even though it can approximate it with sufficient (for science) precision. And yes, I believe that a quantum computer will be able to replicate a fruit fly well enough that we'll have to call that rwplica alive.

Edit. And let me address another dogma. A fruit fly, or a 10 neuron worm, doesn't exist in isolation like it's pictured in imagination of many computer scientists. The wave function is one for all particles, they all evolve as one intermingled system, so the moment you somehow isolate a fruit fly to make simulation tractable, it loses a good deal of what makes it a true fruit fly.


I don't see any reason why determinism should have anything to do with conscious experience.

Can't you conceive of a deterministic universe in which you have internal experience just like you do now? What specifically makes those two things mutually exclusive in your opinion?


That's the crux of the matter. Conscious experience cannot exist in determinstic worlds. In finite sized worlds, such as calculators, any deterministic algorithm is equivalent to a giant lookup table: (input, memory) -> (output, memory), all four entities have a finite fixed size. For infinitely expanding deyerministic worlds it's more complicated, but it's obvious that such worlds have to expand at the expense of some outer realm, as you cannot create more GBs of memory out of thin air.

This problem with determinism has a theological angle (I've warned you). The so called "fallen angel" believed that consciousness can exist in deterministic worlds, and he had an impressive design of such a place. That world was created, but his vision never materialised: his mind was vast enough to imagine such a world, but not enough to see its subtle deficiency.


But what's the deficiency?


Ooh, i think i remember this one: It had the cosmic coincidence of one of its researchers being called Timothy MOSCA.

https://www.smithsonianmag.com/smart-news/see-the-first-comp...

good to see it got some more attention.


He was in charge of ethics.


Codex provides access to proofread static snapshots of the FlyWire full-brain connectome - wiring diagram of Drosophila brain.


This is very interesting. Any wild guesses how far (in years) away from building a synthetic fruit fly? 10 years?

By synthetic I mean a fly that doesn't die and that can be reproduced again.


I’d estimate We are 5 to 10 years away from somewhat deciding the effective connectome and “algorithms” of the roundworm which just has a few hundred neurons. We know every neuron in it by name and know how they all connect to each other, yet we have very little clue how they all end up creating the behavior we see in the worm. We are quite far away from truly mapping every neurons connection and functional effect in a fruit fly which has tens of thousands of neurons.


I'll ask this here, because I'm curious, and it seems that you've read at least a little on this subject...

Is it known how much individual variation there is between different roundworm brains? If you grow two roundworms, even from the same DNA, does the resulting brain have all the same wiring, or is it different based on environmental factors?

I ask because I thought that brain growth involves a feedback loop, where individual experiences during maturation influence the actual physical structure of the brain to some degree.


In a round worm, the neural structure is remarkably constant. The exact same number of neurons emerge and form the exact same network in every worm! But in each worm they might move a bit around slightly especially in their head region which makes quick identification a little hard!

But you can see why this makes the worm the best first organism to decipher the full neural network in. In my opinion not enough people are working on this. There are maybe 5 labs that are working on this topic at this moment!


10 years is barely enough for building a working connectome (a model of brain which use to react in a somewhat similar way to a C.Elegance's brain while consuming tens of KW or more).

Building not even an organism but at least a brain in vitro is impossible in any observable future even for C.Elegance, nothing to say about a fruit fly.

We can not even build a working model of Solar System with ability to Rev/FF in at least a 1000 years and you are talking about one of the most complicated things in Universe.


There's already been "synthetic organisms" built ... Start with the DNA of a simple organism, edit the genetic code, then synthesize this new DNA and insert into a host cell who's DNA has been removed.

I think we're still a long way from designing an organism from scratch though, as opposed to "editing" an existing one.


Is this an actual whole-brain connectome or is it that same work based on a larval brain with several orders of magnitude less complexity than an adult fly brain?


adult


There's lots of very exciting work going on around the fully mapped fruit fly connectome. For example, I'm a CTO of a stealth startup that aims to do for utilitarianism what carbon credits did for environmentalism. We are selling 'utility credits' which translates directly into us simulating trillions and trillions of fruit fly brains in a state of constant orgasmic bliss, which you can then buy to offset any actions your company has undertaken that damage global happiness or well-being. We've seen a lot of interest from some pretty large industry players.


My subjective probability that this is sarcasm is .87 or so


Throw some ML and NFTs in there and I'm sure you could get this funded.


beautiful satire!


This would be comedy gold if only it wasn’t so believable.


This is great, but suffers from the same issue as carbon credits in that any secondary market is likely to allow multiple pledging for the same underlying activity. If someone is already simulating millions of orgasmically happy fruit flies, for instance, they can earn utility credits by promising not to deactivate those flies - but how can we verify the same promise hasn't already been extended elsewhere?

I think you need some NFTs.


> I think you need some NFTs.

And if you build them on a proof of work system you even increase the demand for carbon credits. It becomes a wonderfully self-sustaining and robust ecosystem unlike that pesky delicate one in the amazon getting in the way of good farmland.


This is the exact problem that my startup NeuroNFT is trying to solve

NeuroNFT is a platform that uses blockchain technology to generate and trade NFTs (Non-Fungible Tokens) based on the whole brain connectome of the fruit fly.

The core of the platform is a detailed, dynamic model of the fruit fly's brain connectome, mapped with high precision using cutting-edge neuroimaging techniques. Each unique connectome can be tokenized into an NFT and owned by an individual.

These NFTs aren't just pieces of digital art. By leveraging advanced machine learning algorithms and computational models, each NFT can act as a 'brain' for a virtual fruit fly in a digital environment, offering different behavioral patterns based on the unique neuronal configuration represented in the NFT.

This platform serves as a bridge between the worlds of neuroscience, blockchain, and digital art. It provides a novel way for users to learn about neuroscience, while also offering a unique, scientifically rooted form of digital collectible. Moreover, it paves the way for groundbreaking research in understanding brain dynamics and behavior, with users' interactions with their NFTs contributing valuable data to the field.


Parallel invention, missing attribution or same person?

https://twitter.com/amasad/status/1675393593953878016

Utilitarians: what's stopping you from spinning up a 1M H100 cluster and putting these fruit flies in an infinite state of orgasmic bliss?


Honest question: What about someone doing the opposite, simulating trillions of suffering flys? Shouldn‘t we worry about that?


Utilitarian blackmail pay us x amount or we torture fruit flies.


Create sentient AI and hold them to ransom? Sounds like a Neal Stephenson subplot.


We should have coffee.

I work with many well known state and local governments that have programmed which pay people not to steal and commit acts of violence. The idea is that every person who's not actively hurting someone should earn financial rewards. People with a proven track record of hurting people get higher rewards.

We've been trying to figure out a way to charge the general public for these positive externalities and I think you're on to something.


Sounds about as legit as carbon offsets genuinely are.


So sad SBF is too busy to be able to invest in your startup.


A philosophical challenge to utilitarianism.


Sprinkle some blockchain bullshit around this and you'll be drowning in money :)


Blockchain and machine learning in addition to the fly connectome so you always locate the highest peak happiness state.


I'm cutting out the middleman and just feeding fruit flies myself.


That is very interesting. Where can I read more about your company?



Absolutely brilliant.


> We are selling 'utility credits'

I'm sorry but this is where you lost me. We should be democratising utility credits, not selling them.


Yes, if we can find a way to mine bliss utility simulation credits on our smartphones, we could offset personal moral failures.


Of course, in the beginning, it could only offset base actions like pointless violence and aggressive, rude behaviour. But in the future, as we hope to simulate more advanced brains, we could offset for instance cognitive dissonance, double standards, and perhaps one day even vanity!


This could massively save on the social security budget.


Isn't this concept lifted directly from Ned Beauman's Venomous Lumpsucker? I'm hearing bells. Church bells. :)


Is your company called Black Mirror ? :-)


What a coincidence, I recently founded a startup with a similar mission. It's called TormentNexus.


Aren't carbon offsets (credits) a result of laws being passed by countries?


Aren't utility credits called money?


No. If I have 100 dollars, I might have gotten those from betting on old person fight ring in a care home - a net utility negative, or I might have gotten it from teaching disadvantaged teenagers PHP - a net utility positive.

If you have 100 UC, the only way those credits could have been minted is through 100 hours of 1 million simulated fruit flies having the time of their life.


> I might have gotten it from teaching disadvantaged teenagers PHP

Aren't they disadvantaged enough??


Bravo!


>...old person fight ring in a care home - a net utility negative...

Given the potential for freeing up much needed beds, I think this is debatable


If you think about it toughening up our elderly could increase their health and cause them to not need care beds as much. After the initial beatings I mean.


Hmm good point, I didn't think about that - it's a complex issue I guess


We should take a break and go down to the old person right ring.


why do you think the elderly ought not enjoy some full-contact shenanigans!?


Ah, a Venomous Lumpsucker fan.


fly.org by any chance?


[flagged]


I’m still not sure if the OP was joking.


It’s from a book.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: