Originally published at: Scientists created a new neural network based on dragonfly brains | Boing Boing
…
8 Likes
A dragonfly ain’t gonna go all Skynet on us, right?
Sure what could go wrong?
Dragonflies successfully capture up to 95 percent of the prey they pursue
22 Likes
It’s got my brain buzzing…
4 Likes
An AI based on the brain of a dragonfly? Do you want a Lexx?!?? Because this is how you get a Lexx!
Lexx: the most powerful weapon of destruction in the two universes
12 Likes
I was about to ask if they named it Lexx, but I see you are way ahead of me.
6 Likes
Insects, apparently simple but often astonishing in what they can do, have much to contribute to the development of next-generation computers, especially as neuroscience research continues to drive toward a deeper understanding of how biological neural circuits work.
right on.
Does anyone know if Sandia National Labs has a policy preventing researchers from publishing?
Frances Chance has published decades of good theoretical and computational neuroscience, but nothing on insects, prey capture, or collision avoidance. This is the most detailed pub I can find on the work that she’s done in her years at Sandia. I’d love to get more info on this project, but there’s nothing online.
4 Likes
Should have applications in debugging, if nothing else.
12 Likes
…and all it can think about is sex and bugs.
2 Likes
“It’s not a bug: It’s a feature!”
8 Likes
6 Likes
Not everything that Sandia does is locked down, but getting publication approval is, as you might expect, not trivial.
On the other hand a lot of the researchers at Sandia are contractors and aside from the usual strings attached to anyone with a security clearance they can publish as they like. I’ve sat in on several seminars given by Sandia folk and know rather well someone whose work is in conjunction with them.
Also, the national labs in New Mexico have an arrangement with NM Tech to allow lab staff to switch hit with the faculty; one of my advisors left Tech while I was there to go back to Los Alamos. Happens a lot.
5 Likes
Dragonfly today, sand cat tomorrow.
this quote has me curious. do we really have a deeper understanding, or are computers just more capable of replicating - really, reaching - the complexity of simple neural circuits
it seems to me we don’t “know” or “understand” how facial recognition, language, or deep fake algorithms work - what we know is how to write systems that learn to mimic them
i imagine that insect systems are the same. we learn to model the neurons, we learn to model the environment ( the particle smells or whatever ) but do we know more than we used to?
2 Likes
There seems to be 2 separate questions here:
- are we gaining a deeper understanding of how biological circuits work?
Yes, absolutely! - are we doing so from creating artificial neural networks that replicate biological function?
No, we’re not
We are not gaining better understanding of how brains work from artificial neural nets, but her point is that we’re gaining understanding of biological nervous systems (from studying them directly) which then is applied to new biomimetic systems
Take the dragonfly, they have particularly fast photoreceptors that can encode changes in light ~8 times faster than human photoreceptor which was learned decades ago from direct photoreceptor recordings. This leads to study of the cellular details of their receptors and how we can apply that to speed up man-made sensors.
The visual circuit anatomy of insects is mostly maintained across species, but the dragonfly has a specific set of dedicated neurons for prey tracking. We’re now in the third decade of studying these neurons and related circuitry to learn how the circuit filters the visual information and how the neurons’ cellular properties produce the precise discrimination between collision and non-collision trajectories. Her model for artificial tracking is presumably incorporating this research
So far there is less work on how dragonflies integrate the descending signals with flight control circuitry to produce the real-time adjustments that factor in both target and predator movement, but its an answerable question and someone will get there. When they do, that can be used to improve design of man-made devices
5 Likes
I should probably clarify that I’m all for what sounds like really interesting R&D that could have some really amazing applications.
I’m also for a global comprehensive ban on autonomous weapons that we all take as seriously or more seriously than nuclear proliferation.
I mean the Terminator HK doesn’t even function anything like a dragonfly, right?
Doomed I tell ya!
I would argue that our systems for facial recognition et al. do not even mimic these capabilities of our brains or those of other animals, but (sort of) accomplish them in a very different manner, such that the trope “learning from computer systems how nature works” is entirely moot, at least for those capabilities you mention.
2 Likes
I, for one, welcome our new insect overlords.
This topic was automatically closed after 5 days. New replies are no longer allowed.
