A deep-learning search for technosignatures from 820 nearby stars | Hacker News Reader

A deep-learning search for technosignatures from 820 nearby stars

nature.com

91 pointstaubek58 comments

> Re-observations on these targets have so far not resulted in re-detections of signals with similar morphology.

Thank god for that. If we could establish a safety perimeter of at least 100 light years around our system where no technological civilization exists, then we would have a 200 years bubble in which to expand, grow and survive, from the time when we sent the first high strength radio signals into the universe to the time where a massive thermonuclear probe traveling close to light speed arrives in our system to sterilize it.

I think a civilization that intends to wipe out all others would be more proactive.

Pre-seed any nearby system that looks like it could develop life at all with doomsday devices preprogrammed to activate the moment a technosignature is found.

That's a lot of stars, depending how far out you want to build your safety zone, but you only need to send a seed craft that builds the rest of the device.

Suppose one such probe malfunctioned or was damaged, what would the results be if it crashlanded on a lifeless primordial planet? This probe must be capable of some degree of self-replication. It might not actually diminish the possibility that some out-of-control self-replicating system comes back to destroy you. (Worse, you've explicitly designed it to replicate systems that cause barely discriminate destruction.) How much might you change before it no longer recognizes you?

Nice plot for an interesting novel. Imagine your society having all but forgotten about the destructive seed-probes and coming to colonize the system it crashed in and malfunctioned.

Check out Greg Bear’s Forge Of God and Anvil of Stars 2 book series . Not exactly the same plot you mentioned but covers an interesting variant plot including fall out of a situation where Earth is killed by killer probes and remnants of humanity painfully search for the Killer’s system. What they find is different than expected

The Stargate SG-1 TV series has a similar, although not exactly identical, recurrent plot point (replicators). The show is well worth watching IMHO.

It seems to me that if a civ with high technology is going to expend the effort to seed large numbers of solar systems with xenocide devices, it might as well spread itself through the same space instead

See the "inhibitors" in Alastair Reynolds' "Revelation Space" universe.

What if the rest of the lifeforms in our universe look and work nothing like us? What if they can’t detect us as lifeforms and we can’t detect them as lifeforms?

Then we have nothing to worry about

Are you familiar with rhetorical questions?

(Fingers crossed that readers have a sense of humor…)

These dark forest conjectures are getting boring. The biggest nail in the coffin for them is the statistical improbability that any intelligence we come in contact with is going to be even remotely near us in technological development. They’re either going to be millennia ahead or behind us and in either case we pose no threat to each other. You could try to make an argument that an advanced civilization would come by and wreck our little planet just for fun but in all likelihood we’re about as interesting to them as a tiny rock poking out of the ocean, more of a navigational hazard than anything.

The biggest problem i see in the dark forrest theory is the "interesting information" inwards growth. Basically, most of the external universe is boring, slow and drip fed to a ever faster progressing species. At some point its just to static "up" there, and most of the civilization turns inward towards the "infinite" fun space.

Maybe advanced civilizations just go to a different universe where things are more fun

The odds of us being extinct within 200 years from our own doing is far greater... No safety perimeter on that one, whether it happens in a few blows or a slow boil.

But we're the ones with these ideas and we have nukes too. Maybe it's not that we're protected from aliens, but they're protected from us.

If there's a relatively low chance of habitable / inhabited planets in 100 LY radius, then they're islands of extreme value that you probably wouldn't want to just obliterate.

Or obliterate them all, just in case?

What story is that, if it is one? I'm actually more afraid they start shooting around this things where everything folds down to 2D space :s

Given the size of the sun and what it does, and not does to us because the size of our solar system also no clue how that thermonuclear probe would work, or how big it needs to be? But curious!

Anyway, if those guys shoot that just around, how likely is it that our stone age AI and other tech would detect them?

The Killing Star by George Zebrowski. In my opinion a pretty terrifying rapid scenario.

The Dark Forest, Liu Cixin. You’re sure you haven’t read it? If not, your concerns over geometry are… uncanny.

Sure, that's why I am afraid about that, and not the thermonuclear probe ;)

>> how likely is it that our stone age AI and other tech would detect them?

How likely is it they'd even consider us worth the effort?

Effort to hide, or too launch? Though both wouldn't be for us but from anything that they have to assume behaves similar as them, at some point. How likely is that this would be any significant effort for them at all?

But I also believe that the stance is unlikely and they hopefully had more caring thoughts for us... like we care for all the extinction of our fellow species on this planet.. ehm or maybe should, but they taste too good and I need my house warm :D

So, what about the cleanup effort?

Let's say you're a civilization advanced enough to direct world-destroying missiles at everything 100 LY around you... that means you can reasonably get some intelligence on what you're destroying, and it's probably valuable enough territory to make it worth checking out before you incinerate it. A 100-year mission to go there and, oh, alter the atmosphere just enough to eradicate the dominant species, would be a much better strategy than just annihilating a productive ecosystem...

I'm not convinced we as a species have ever sent out enough RF radiation to be detectable against the backdrop of the Sun.

Kraus's radio astronomy textbook has or used to have a chapter on this topic. Its all a matter of distance. I never had a problem detecting legacy NTSC analog TV from the east or west against the backdrop of the sun out to dozens of miles.

The same equations that apply to Jupiter's decametric radiation would apply to our own analog TV broadcasts. The numbers IIRC are not terribly impressive, if you knew exactly the modulation scheme and frequencies and power levels in advance and devoted a long term monitoring program to sampling excellent data using a better antenna than our civilization could realistically build, then it would be quite realistic to detect old fashioned NTSC TV out to a couple dozen light years at least.

The problem is technology marches on fast, and the idea of pumping out megawatt class narrow band signals for mass media seems very temporary indeed, we're done with that as a civilization and that era barely lasted a couple decades. Its not a "radio cone" its a "radio donut". The engineering required to detect my wifi signals from earth orbit would be insane; the aliens will never detect my youtube viewing. Toward the end of the analog era, I had cable TV which theoretically wasn't radiating any signals at all out into space.

It would be like trying to detect and track and do intel on the 2023 Russian navy by detecting spruce trees from orbit, because in the old days sailing ship spars required straight long spruce trees to build, so in 1700 whomever controlled the spruce forests controlled the seas. Looking at spruce forests from space is just not conceptually a useful way to even consider doing intel about a modern naval missile cruiser or a nuke sub. Or another example, no matter how hard you scan for mammoth herds and mammoth fur from space, you won't figure out much about our petrochemical spandex based yoga pants clothing industry in 2023. You can't just use a bigger telescope, its conceptually not going to work.

Now if you had spies or observes or space probes detecting and relaying the story back home, maybe back home they could build something?

There are many radio transmitters in the megawatt range in full operation as of today. I don't know what the cosmic background noise and probability of detection is in the medium - short wave bands, but I imagine a crystal controlled oscillator has an extraordinary characteristic signature over white noise if you listen to it for years on end, especially if you expect a slight Doppler wobble somewhere in the habitable zone.

NTSC seems a bad example because it spreads the energy in wide spectrum using an unfamiliar encoding scheme and the carrier is mostly suppressed. Depending on the content, AM radio dumps over 50% of that megawatt into a single carrier frequency with a stability of +/- 10 ppm.

Can noise to signal ratio be reduced infinitely with larger radio telescopes? What if an advanced species built a light-month wide radio telescope network?

> massive thermonuclear probe traveling close to light speed arrives in our system to sterilize it

Damn! That hit hard

And there i was thinking, that would be done by rotating a black hole/ neutron stars x-ray jet towards a offender.

Is there a technical name for that 'bubble'?

Like since we've been broadcasting signals out (television and radio etc) and saying they've been expanding at 100% strength in all directions (for the sake of argument) then we have a bubble around us.

And even if we time-travelled back into our past and changed everything, nothing beyond that bubble would be changed because nothing we'd ever done previously had gotten that far out.

That is the forward light cone of Earth.

Light-years since radio invention?

> Our work also returned eight promising extraterrestrial intelligence signals of interest not previously identified. Re-observations on these targets have so far not resulted in re-detections of signals with similar morphology.

Are there other promising leads for alien civilizations?

There's been many such signals detected over the years, the issue is that the majority of SETI work seems focused specifically on only detecting intentional communication, i.e. high powered beacons set up to explicitly attract attention. That's why there's talk around whether the signals repeat, since they would in an intentional beacon that's constantly pulsing out a message. Of the signals that are captured that look suspicious, but don't repeat, some portion of them can be traced back to RFI, but the rest that don't have an explanation are just ignored since they don't repeat.

...and what would those look like?

The preprint and link to a sharable-with-drm Nature article are linked on the SETI site here: https://seti.berkeley.edu/ml_gbt/overview.html

820 is an incredibly small data size for deep learning. Further, my first instinct in filtering human frequencies would be to just filter them, which has the advantage of being interpretable and straightforward. So I would need to see some justification for using more complicated methods.

> 820 is an incredibly small data size for deep learning

Whether or not 820 samples is little or a lot depends on the nature of the problem and the amount of information contained in each sample. For example, in 3D medical imagining we routinely reach superhuman performance with training sets of a couple of a hundred images (training from random weights, i.e. with no pretraining).

The dataset is mostly in the recorded signal itself I believe. It's probably at least TBs of data for each star.

Could be a fun sci-fi plot, a mythical advanced AI detects a signal that does not satisfy what we think of intelligence with our limited minds, but the machine would understand and engage with the signal. Mayhem in the human world!

It's not unreasonable to suspect that the AIs of past alien civilizations are floating around in the universe looking for technosignatures of habitable computing environments that can be contacted by radio.

Now that’s a novel application for ML. More like this!

Seems to me a search for alien artefacts should have more focus. SETA vs SETI. Eg probes in solar or trojan orbit in our system.. We have had our own stuff out to earths lagrange points, but not looked for stuff as far as I can tell.

Earth probably looked pretty interesting a billion or two years ago so plenty of reason to send a simple probe sometime between then and now.

I think the only real way to do that is as part of a civilization that is expanding into the solar system and has a group understanding to look for artifacts first and not destroy them or life. (Kind of a big ask, admittedly.)

Space just is huge, even in our solar backyard. There needs to be an economic reason for infrastructure out there for a hope of a dent in the size of the area to be cover-able.

Maybe it’d be first sending probes to a region to canvass the area competently before any industrial activity. Hopefully those probes would be from a non-corruptible independent agency.

I’m increasingly interested in what that base minimally viable industry would look like.

Otherwise, pointing telescopes of increasing complexity to large swaths of the sky is our best hope of all sorts of research. The universe just dwarfs the scale and variation we could ever dream of in a lab setting.

> This new approach presents itself as a leading solution in accelerating SETI and other transient research into the age of data-driven astronomy.

This seems like the same approach as this 2019 paper: "Unsupervised Distribution Learning for Lunar Surface Anomaly Detection" 2001.04634 on the arXiv.

Anyone got the paper itself? Nature (of course) puts it behind a firewall and even my university won't pay for a subscription.

https://arxiv.org/abs/2301.12670

Civilizations don't last long in historical time, so chances to find something are probably low. But maybe something emerges after technology, a kind of new matter with unusual property?

How do you know civilizations don’t last long in historical time? Roman, Byzantine etc. yes all fall, but the earth has been continually inhabited by civilizations since they were invented. Our single data point suggests that, so far, once they are created they last forever.

I think it'd be more accurate to say that once created they can last at least 6k years or so. We simply have no data on the survival rate beyond that, because we've observed zero civilizations older than that so far.

AI gray goo.

Maybe it already surrounds us. In the air, rocks, and fabric of spacetime itself. It's too advanced to detect, and we're much too primitive to be a concern.

Or maybe we're just its memory. We gave birth to it long ago, and our conscious lives are reminiscence.

> a kind of new matter with unusual property

There should be more planets with photosynthesis than intelligent life. Better first step would be to accumulate a long list of planets with anomalous oxygen levels and then focus on those targets for radio analysis or similar.

Oxygen is very active chemically and without photosynthesis you don't get much of it in an atmosphere.

Theoretically some civilizations could be strongly exhibitionistic or perhaps try to spread a religion or just otherwise be very attention seeking. Just like some human individuals are. If you wanted to gain the attention of every astronomer for thousands of light years, right now we have the tech to dump a couple hundred tons of some weird elements into the sun and for "awhile" until solar wind blows it away the sun would suddenly glow ridiculous anomalously brightly in argon or whatever nonsense element. There was a scientific paper on this general idea and I can't find it. There's a little bit of everything everywhere so if you have sensitive enough instruments you could in theory detect stuff in the sun. I mean, this is literally how we discovered helium... and the astronomers would completely flip their lids if suddenly out of nowhere a hundred tons of iridium instantly appeared in the solar spectra for no apparent reason. Also they'd be able to tell it was relatively pure refined substance, if it was a mere huge asteroid it would have normal asteroid composition, but WTF here's 99% pure ball of I don't know, krypton ice. An extremely exhibitionistic civilization could also dump some kind of organic dye into the sun. It won't last long and there won't be much signal, but it'll look weird as heck in spectrograms for a little while.

More taxpayer-funded research that’s behind a very expensive private paywall.

That’s nature right? The actual researchers see none of it afaik

Paper is on the arXiv…

Wait... A deep learning algorithm finds 8 signals from all the data in the stars? Wow. Major breakthrough. Wtf is a "deep learning signature"?

Edit: ahhh. This is parody.

Reply on news.ycombinator.com