AI, as in neural nets
August 31, 2017 7:12 PM   Subscribe

 
this seems like a beautiful use case for neural networks. it's supervised learning and they have tons of training data.

there was a recent high-profile success using a system called Celeste (a Bayesian generative model, not deep learning) to analyze astronomical data. the authors of that paper suggested astronomy is a field that's been underserved by machine learning -- tons of data, not a lot of work done -- and called for researchers to attack other problems in the field. good to see that call being heeded.
posted by vogon_poet at 7:50 PM on August 31, 2017 [2 favorites]


I'm all for using improved computational techniques, including AI to help understand the world. I think that a lot more observatories need to be established... and networked.

Recently NASA deployed a team to Argentina to observe the occultation of a star by an object which was a potential observation target for the new horizons probe... They had 24 telescopes deployed in a linear array, a few miles apart (if I understand the details correctly), and were able to observe the object, and get a rough determination of its shape because of this line.

If we had thousands of telescopes deployed across the world, that could be synced up and observe the same stars at the same time, other objects could be detected, orbits calculated, etc. I think it would be a big win for science.
posted by MikeWarot at 8:11 PM on August 31, 2017


I think we need to get off this planet. Satellite scopes are great but I keep fantasizing about a facility on the back side of the moon with all of the moon as shielding from the noisy earth. Maybe a satellite scope in a synchronous orbit on the back side of the moon as a stepping stone?
posted by aleph at 9:06 PM on August 31, 2017 [1 favorite]


I think, as an absolutely uninformed observer, where we're at here is Linnaeus; these tools allow for the rapid identification and classification of specific objects in the astronomical data collected by various sensors. It's basic, fundamental work that opens up galaxies of new discovery and now can be done quickly and cheaply. Think how far biology has come since Linnaeus.
posted by notyou at 9:31 PM on August 31, 2017 [1 favorite]


This, to me, proves one point - neural networks are good at one thing that makes us human: classification and establishment of patterns.

Now, for the next challenge of humanoid AI: creativity. My hunch is it will have to do with real randomness, so getting hold of a good random generator and spicing the current AI with it could move us one step further.
posted by Laotic at 9:40 PM on August 31, 2017


I keep fantasizing about a facility on the back side of the moon[...]

Also known as the dark side. They wouldn't be able to see anything!
posted by Joe in Australia at 9:43 PM on August 31, 2017


I'm assuming you're being facetious with that comment, but there is a logistical concern about a far-side lunar observatory, that being actually relaying data to and from Earth. Presumably you could use a set of satellites in orbit around the moon?
posted by NMcCoy at 11:14 PM on August 31, 2017


There is no dark side of the Moon really, as a matter of fact it's all dark.
posted by Meatbomb at 11:16 PM on August 31, 2017 [4 favorites]


There is no AI really, matter of fact 'tis all hype and misnomer.
posted by GallonOfAlan at 11:54 PM on August 31, 2017 [2 favorites]


Now, for the next challenge of humanoid AI: creativity

I tend to think that is a bad, bad idea personally. We will inevitably be at odds with any other intelligence.

But I'm all for using AI to do stuff like this.
posted by fshgrl at 2:05 AM on September 1, 2017


Regarding the dark side of moon,
To believe in it you'd be a loon,
There's no permanent night,
It gets flooded with light,
An eclipse is it's version of noon.

posted by quinndexter at 2:44 AM on September 1, 2017 [1 favorite]


The far side of the moon would be a good place to put radio telescopes, as the mass of the moon would block all the signals we emit. This assumes there isn't any way to effective way that nature bends signals around the surface of the moon, like our ionosphere.

A day on the moon is 28 days, and the phase of that day is visible from earth.
posted by MikeWarot at 3:59 AM on September 1, 2017


I tend to think that is a bad, bad idea personally. We will inevitably be at odds with any other intelligence.


There would be an upside and a downside, I reckon. The upside can be realized if we can harness an intelligence vastly superior to ours in terms of both data processing capability and creative output and selection. As long as this intelligence wasn't asked anything regarding our role on this planet, we could benefit from it greatly.

The downside would be, that if we managed to build an intelligence similar to human, with all its necessary sensory inputs, creativity and data processing capability, my hunch is that it would necessarily begin to suffer the same, and magnified, mental defects that affect humans. And producing a huge AI just to find out you've created an autistic sociopath would be, well, disheartening.
posted by Laotic at 4:06 AM on September 1, 2017 [1 favorite]


I think we need to get off this planet.

The James Webb Space Telescope is launching next year. Almost 3x the diameter of the Hubble, it will be parked at the Earth-Sun L2 point, which is out past the moon, about 4x as far away as the moon.

Looks like it's going to be amazing.
posted by jjwiseman at 10:34 AM on September 1, 2017 [1 favorite]


« Older Massive "Wooden" Roller Coaster Built from 90,000...   |   Exuma, the Obeah Man Newer »


This thread has been archived and is closed to new comments