A lot of thinking has gone into how to communicate with aliens. But is it enough? Even messages designed to be understood by extraterrestrials, like the Arecibo message, the Pioneer plaques, or Voyager’s golden record, rely on conventions that may be uninterpretable to any non-human intelligence.
More recently, there have been efforts to collect a trove of information about humans and what’s important to us (Wikipedia, dictionaries in many languages, libraries of books) and then print it in miniature archives that are designed to survive for a long time. In all cases, it’s doubtful if having more data would meaningfully increase the chances that any of it is understood by aliens.
Could a version of our LLMs, trained on humanity’s knowledge, be a viable alternative and act as our ambassador aboard an interstellar spacecraft? Maybe sending a robust computer that runs a good AI is the best thing we could do, besides going ourselves.
Such spacecraft would have to be very small to make it possible to accelerate them to the speeds required to reach other stars in hundreds or thousands of years. It would be extremely challenging to build a computer that could run a sufficiently powerful AI, that would also be able to withstand the hundreds or thousands of years it’d take to reach another star, and that would be light enough to be launched on such a journey. The Breakthrough Starshot initiative’s StarChips, even though they’re the most advanced designs for miniature interstellar computers, are still a long way off.
Getting the AI running on this computer right would also be difficult. Current LLMs aren’t up to the job, but in the last few years it has become possible to see a path to a future where they will be. Ideally, such AIs would represent humanity, less ideally, they’d represent some fraction of it. They would be able to make decisions autonomously, and may even have the ability to steer the spacecraft. They’d be able to look for signs of intelligent life, and if they find it, would make the decision whether to reveal themselves by sending a signal. Once contact is established, they would be able to try different methods of communication, making success more likely than any static archive would. They’d also have the ability to reliably self-destruct when there’s a risk of giving away important information that makes humanity more vulnerable.