German researchers say AI has designed tools humans don’t yet understand for detecting gravitational waves, that may be up to ten times better than existing human-designed detectors.
https://scitechdaily.com/when-machines-dream-ai-designs-strange-new-tools-to-listen-to-the-cosmos/
22 Comments
Submission Statement
Einstein imagined gravitational waves over a hundred years ago, but it wasn’t until 2015 that they were first detected. Detector design is complicated because a wide range of parameters need to be fine tuned. It turns out when AI is set to this task it finds new types of design that humans so far haven’t even thought of. The creators wonder if this approach might lead to similar results in other fields.
What AI could even achieve this right now? Either they have something magnitudes better than what’s available to the public or this might just be more hype bs.
If we don’t know how it works, then how do we know it isn’t making mistakes?
Sounds like some pie in the sky headlines cooked up by some MBA looking for investors
If we don’t know how it works then how do we know it works?
It hallucinate like crazy
Will only be impressed if AI cure something like cancer.
We don’t know what it is but it’s 10 times better. More techie garbage to profit from destroying the environment and humanity
It would be interesting to see how small of interactions systems like LIGO could detect.
Claiming 10x more sensitive is one thing, but could we get to a point where LIGO could claim it felt gravitational waves from non-exotic bodies? Not neutron stars and black holes.
Could we see stellar collisions, or even stellar supernova, from it? That’d be cool.
Its not possible to do science with tools we don’t understand. Its just not possible. Understanding the tools, and all possible ways they can go haywire, is mandatory to interpret scientific results. It would be extremely bad science to trust any results coming from a tool no one understands.
Actual paper, instead of this nonsense: [https://journals.aps.org/prx/abstract/10.1103/PhysRevX.15.021012](https://journals.aps.org/prx/abstract/10.1103/PhysRevX.15.021012)
Abstract:
>Gravitational waves, detected a century after they were first theorized, are space-time distortions caused by some of the most cataclysmic events in the Universe, including black hole mergers and supernovae. The successful detection of these waves has been made possible by ingenious detectors designed by human experts. Beyond these successful designs, the vast space of experimental configurations remains largely unexplored, offering an exciting territory potentially rich in innovative and unconventional detection strategies.
>Here, we demonstrate an intelligent computational strategy to explore this enormous space, discovering unorthodox topologies for gravitational wave detectors that significantly outperform the currently best-known designs under realistic experimental constraints. This increases the potentially observable volume of the Universe by up to 50-fold. Moreover, by analyzing the best solutions from our superhuman algorithm, we uncover entirely new physics ideas at their core.
>At a bigger picture, our methodology can readily be extended to AI-driven design of experiments across wide domains of fundamental physics, opening fascinating new windows into the Universe.
whats a gravitational wave and why do we need to detect it?
Hmmmm I don’t know if we should use tech that finds things in ways we have difficulty recording locating things we don’t understand. Seems like there would be a loss there.
This totally flips the script on how science gets done. Instead of AI just crunching our data, it’s designing tools we couldn’t even imagine. The fact that these researchers admit they don’t understand some of Urania’s designs yet is both humbling and exciting
Imagine what Einstein would think about an AI named after a Greek muse designing better ways to detect the gravitational waves he predicted. The “Detector Zoo” concept is smart too crowdsourcing the human understanding of machine generated ideas. We might be looking at a future where scientific breakthroughs come from human-AI collaboration rather than human genius alone
Just wait until the AI says the equivalent of “separate U235 from U238 int *this* way and combine two pieces together very quickly. You’ll get a lot of energy released. And somebody tries it.
I see some people here are confusing design optimization ML models with LLM’s.
NN, Tree models, polynomial regression etc don’t really hallucinate. They just over or under fit and you can test the predictions against known cases to determine if they are predicting outcomes with good enough accuracy. Yes they are black boxes, but that doesn’t mean they are hallucinating.
This is hype train nonsense. There will be another new Grift in a couple years
Huh? How do you even know the tools are “better” if you don’t even understand them? How is that not poor design? Just sounds like a puff piece for ai.
i was reading about AI designing integrated circuits, processors, etc.
some of the chip designs work in reality, but inside the chip, like half the circuit isnt even connected to the other half.
kinda like induction to measure amperage in a wire, but they cant figure out how or why the designs work.
its wild to think its just starting, soon we wont understand much of what AI does, it will start thinking in its own languages, designing higher level AI, coming up with technology/processes/hypothesis’s that we havent even considered because of exponential progression.
but in the sametime/ meantime its almost comical how stupid AI is in normal conversion.
We don’t really understand how any neural network of any complexity works, including the biological ones we have always used.
Sure, let’s start building incomprehensible machines designed by our slaved AIs.. Nothing can go wrong
Advancing in AI is going to be its undoing and eternal damage to the Earth’s climate.
I don’t think I’m seeing anything in the article that indicates Urania has produced anything of substance. The author sounds like they’re being careful to avoid claiming the solutions it designs actually work, besides one vague sentence:
> […] Urania has designed a series of novel gravitational wave detectors that not only match but often exceed the capabilities of existing human-made concepts.
The article is basically, “This tool has generated many experimental ideas that are good somehow,” on repeat. There’s not even any lip service paid to the testing processes or any of the metrics by which these nebulous “tricks, ideas, and techniques” are evaluated. The article does, however, say that “many” of them are “still completely alien to [the researchers],” which sounds to me like there’s no indication that there’s any sense to them at all.
> They have compiled 50 top-performing designs in a public “Detector Zoo” and made them available to the scientific community for further research.
Maybe there’s something to this and the article’s author didn’t want to drown the reader in technical jargon. I’m too lazy to look into the “realistic experimental constraints” mentioned in the paper’s abstract. Maybe someone who knows more can clue me in.