Using AI to Find ET

Original article was published on Artificial Intelligence on Medium

Growing astronomical datasets can analyzed by machine learning, potentially identifying new signals

(Pixabay, geralt)

A universe of data

We are — according to some sources — living in the age of data. More and more aspects of our lives are encoded, encrypted, and informationally exploited.

The data growth is not limited to our lives — including our brain and our genes. Or even limited to our planet. The universe itself is bursting with data, ready to be explored.

Recent efforts in the search for extraterrestrial intelligence (SETI) are collecting that data, searching for signals, for any signal that might potentially signify something more than ‘natural’ phenomena (even though one could make the argument that the natural vs artificial dichotomy is spurious, but that is an entirely different discussion).

These astronomical datasets are not easy to analyze, though. For one, they’re huge — astronomically huge (sorry about that one). They also contain many different ‘types’ of data (light spectra, radio frequencies, gravity signatures…).

Time for machine learning, methinks.

SETI, meet AI

With better detection technologies it has quickly become clear that handling those enormous astronomical datasets is not a one-man job. SETI was well-aware of this and even twenty years ago they explored options to analyze their data in one of the largest volunteer computing projects ever with SETI@home.

With this project, everyone could volunteer computing time by allowing SETI to use the computing power of their PC/laptop during downtime. A few months ago, the project was put on hiatus.

SETI@home, however, was simply harvesting computer power, not necessarily implementing innovative analysis methods.

In 2017, this changed: SETI organized a hackathon to implement AI/machine learning for the analysis of astronomy data. Not too long later, SETI associated researchers used AI methods (a convolutional neural network) to explore data from the Green Bank Telescope (GBT). Likewise, the SETI institute has been using IBM Cloud and AI algorithms to analyze signals captured by the ATA (Allen Telescope Array).

In the words of SETI president Bill Diamond:

… results hint that there could be vast numbers of additional signals that our current algorithms are missing and clearly demonstrate the power of applying modern data analytics and AI tools to astronomical research…”

(For those wanting more technical detail: here and here are two studies on implementing machine learning in the study of SETI data.)

The final frontier

Having AI scrutinize astronomical data for possible signals is not without its dangers, of course. A false positive can be enough to make people go crazy. We’re very good at seeing patterns and meaning where there aren’t any. So, if an AI system spots potentially artificial signals, little green men will be dancing around in our minds. AI-uncovered signals should thus be vetted by human experts (mirroring the potential of human-AI hybrid teams in art).

The opposite is also possible: a false negative. In other words, even if there is life out there, we — and our AI systems — could miss it. Maybe AI dismisses a signal because it doesn’t consider it associated with life (either through self-evaluation or through limits in the training data we provide — which is necessarily limited). Maybe the data we receive through telescopes is (currently) insufficient, for example of ocean worlds covered by a layer of ice.

But can we move AI beyond detection and into space exploration?

Astronaut Karen Nyberg with her colleague Robonaut (Wikimedia commons, NASA)

After all, it’s been argued that the future of space exploration is robotic.

Humans — in their current form — need bulky suits and vehicles to protect themselves from radiation. We also need to carry a lot of fuel, food, water, and other resources. And, until we figure out human hibernation or self-sufficient generation ships, we’re constrained by our — again current — lifespan. A century in space doesn’t get you very far with the means of propulsion we have now.

Robots, though, have their limitations as well, leading some to argue that humans in space are still a requirement for space exploration. But what if we combine AI and robotics? There are already budding AI scientist robots. Preparing them for space will require some tweaks, but nothing that appears impossible.

Maybe they’re also better equipped to say hello if they meet someone (something?). Not just because they can parse and analyze communication signals much faster than a human emissary could, but because it might be their cousins they meet. Indeed, given the above considerations about the vulnerability of biological lifeforms in space, the argument has been made that ET will be robotic/AI. (But again here, let us not discount the possibility and potential of hybrid forms/cyborgs.)

Of course, having AI explore the cosmos doesn’t have to mean we can’t experience the wonders it comes across.

We can always tell it to phone home.