Affiliate Disclosure: By buying the products we recommend, you help keep the lights on at MakeUseOf. Read more.
Imagine that you’re a figment of your computer’s imagination. Your brain is a detailed computer simulation — an artificial intelligence that connects to simulated eyes and simulated muscles and simulated nerve endings, which interact with a simulated world. You think and feel exactly as you do now, but instead of being implemented in grey meat, your mind runs on silicon.
Simulating an entire human brain like this is a ways off, but an open-source project is about to take a vital first step, by simulating the neurology and physiology of one of the simplest animals known to science. The OpenWorm team, which just completed a successful Kickstarter, is months away from building a complete simulation of C. elegans, a simple nematode worm with 302 neurons. The simulated worm will swim in simulated water, react to simulated stimulus, and (to the degree that such a simple organism can), think.
In this interview, we’ll be talking with Giovanni Idili, the co-founder of the OpenWorm project about their work in artificial intelligence. The OpenWorm team is a multinational team of engineers, who’ve been working on the worm simulation for several years. They use file-sharing tools like Google Drive and Dropbox to collaborate, and their meetings are streamed publicly as a Google+ Hangout.
The Future of Artificial Intelligence
MUO: Hi Giovanni! This is obviously a very complex and challenging project — could you describe the progress you’ve made on the simulation so far, and what’s left to do? What do you think will be the most significant challenges going forward?
Giovanni: We have made a lot of progress on the body of the worm and the surrounding environment that will represent our virtual petri dish. We believe in embodiment, meaning that a brain in a vacuum would be less interesting without a simulated environment – the “worm matrix” if you will – that the brain can experience via its sensory neurons.
That’s the reason why we started with putting a lot of effort into the worm body first. What we have so far is an anatomically accurate, pressurized cuticle that contains contractible muscle cells, and is stuffed with gelatin-like fluid to keep everything in place. In parallel, we have been working on getting the brain running, and we are currently running the first tests of the entire C. elegans neuronal network (the famous 302 neurons).
We are now approaching the point that we can start plugging the brain into the body and seeing what happens. This won’t mean the worm is “alive”, because it doesn’t have organs and a lot of biological detail is still missing too, but it will enable us to close the loop on the motor system, so we can start experimenting and tweaking the brain and muscles to generate different kinds of worm locomotion. This alone will keep us busy for a while.
There are two different kinds of challenges – research challenges and technical ones. Research challenges are the ones typical of any scientific venture. You don’t know when you’re gonna get stuck or on what, but one obvious challenge here is that even though the brain is mapped and the connections between neurons are known, we still don’t know a lot about the individual neurons themselves and their characteristics, which leaves us with a lot of work to do to fine tune them — doable, but hard and time consuming.
This is difficult because the animal is very small and so far it has been impossible to do in vivo imaging of the firing brain. Luckily, and this is very recent news, new techniques are surfacing that may help us fill some of the gaps.
In terms of engineering, there are many technical challenges, but I’d say that the main one would be performance of the simulation. We are running the simulation on GPUs and clusters, but still it takes a lot of time to simulate; there is a lot of work to do there.
Browser Worm Simulation
MUO: One of the Kickstarter rewards you made available to your backers was access to a partial simulation of the worm in your browser, including musculature. As you complete more of the simulation (like the brain), do you plan to make those elements available in the browser as well? How intensive will the full simulation be to run?
Giovanni: Yes – this is exactly the idea. The WormSim will be a window into the latest simulation available. Once we make some significant progress, like plugging a brain into the simulation, this will be rolled out to the WormSim. The simulation will be pretty intensive, but the WormSim architecture is currently decoupled from that, in the sense that we will run the simulation on the necessary infrastructure (GPU clusters etc) and then store the results. These results will be streamed to the WormSim, so people will be able to scan back and forth in the simulation, use 3D camera controls and click on things and access simulation metadata.
MUO: Since C. elegans is just the start, after nematodes, what’s the next step? What challenges arise between the nematode and a more complex organism?
Giovanni: Correct. We are trying to build our technology planning for the future, and we want our engine to be a bit like LEGOS for computational biology, ideally, so that after C. elegans we don’t have to start from scratch, but can assemble a more complex organism leveraging what we have already built.
Candidates are the leech (10k neurons) and the fruit fly or the larval zebrafish (both around 100k neurons). It’s not just a matter of how many neurons, but also how well-studied an organism is. It’s certainly gonna be quite a few years before we can even think to tackle other organisms, but if some other group wanted to get started on any of those organisms, we’d be happy to go above and beyond to help in any way we can — all our tools are open.
The main challenge is that as the brain of an organism gets bigger and bigger, like a mouse with its 75 millions neurons, you are kind of forced to work with populations rather than with well defined neuronal circuits made up of reasonable amounts of neurons. “Closing the loop” becomes a bit trickier. Also you need more computational power, and doing something like what we are attempting with C. elegans, cell-by-cell simulation not limited to neurons, is outright unthinkable. Once you get to that macro level, you are forced to work with something more coarse grained. But it’s going to happen, no doubt!
Validation and Testing
MUO: Given that the software you’re developing is very complex and involves simulation at many levels, how do you validate your models to determine success? Are there tests you’d like to perform, but haven’t been able to yet?
Giovanni: At each level of granularity we “unit-test” our software components against experimental results. The experimental data is either already available in the open, or comes from labs that decide to donate it to us. Neuronal simulations have to match experimental measurements on neuronal activity. Mechanical simulations for the body of the worm and its environment have to follow the laws of physics.
In a similar way, macro behaviours of the simulated worm (swimming / crawling) will have to follow experimental observations at that level. There is in fact a group of us who are working on getting ready an incredible amount of data so that we can quantitatively say for sure that our worm is wiggling the same as the real one as soon as our simulation is ready to be tested.
Applications of Research
MUO: Which application of this kind of simulation is most exciting to you? What are the most important uses of this technology going forward?
Giovanni: This kind of simulation, when validated, could enable us to conduct experiments on a computer instead of live animals. This has obvious advantages in terms of reproducing experiments and the sheer number of experiments that can be conducted. C. elegans is a model organism for human disease, so we are talking about possibly gaining bottom-up insight into diseases like Alzheimer’s, Parkinson’s and Huntington’s, just to name a few – and hopefully accelerate the cure as a consequence. The same technology could be used to simulate healthy or diseased populations of human tissues just by loading different models into the engine.
Personally, I am extremely excited by how what we are doing could help us understand how brains work on a very tractable scale. Just imagine what it means if we can capture the brain of a worm as a set of parameters (which is becoming increasingly possible with new imaging technologies) and feed those same parameters into our simulation. This may sound like science fiction, but memories have been implanted in live animals already.
What OpenWorm Means For You
The technology behind the OpenWorm project is exciting on many levels. The technology to map and simulate the brains of whole animals has profound and eventually world-changing implications for the human condition.
On a more immediate level, the ability to experiment on simulated animals and study diseases in meticulous, computational detail may well enable an entirely new kind of science – experiments performed, en masse, by computers, on computers. The technology of OpenWorm, scaled up to larger organisms, could allow us to study hard-to-grasp diseases like schizophrenia and cancer in entirely new and exciting ways.
What do you see the human race achieving with this technology in ten years? Fifty? Let us know in the comments! You can follow the OpenWorm team at www.openworm.org