Large language models may enable real-time communication with extraterrestrial civilizations despite the vast distances between stars. We need to start thinking about what to tell them about us.
Artificial intelligence mania has overtaken our economy and will soon expand beyond Earth to become omnipresent in spacecraft as well. It’s worth asking, what does this mean for the search for extraterrestrial intelligence? Just like on Earth, AI promises a rethinking of long-cherished hopes for space exploration, such as finding that we are not alone in the universe.
AI’s advances explain this ambition. Introduced in 2017, the “transformer” neural network architecture has become the cornerstone of today’s large language models (LLMs). Trained on Internet-scale datasets, these models contain vast human knowledge and are changing our world. They will affect nearly 40 percent of global employment, according to the International Monetary Fund (IMF).
Could this technology help us communicate with hypothetical advanced civilizations elsewhere? SETI, the search for extraterrestrial intelligence, has been around as long as METI, messaging extraterrestrial intelligence, the aim of which is to attempt to find and communicate with extraterrestrial civilizations. But after 40 years of serious search we have not found such ET intelligence, and our messages remain unanswered. We cannot conclude that we are alone in the galaxy, given its vastness and our nascent search efforts. It may be time to radically rethink our approach, however.
As alien-curious scientists, we propose advancing METI by transmitting not just music, math or brief descriptions of ourselves but something more meaningful: a well-curated large language model that encapsulates the diverse essence of humanity and the world we live in. This would enable extraterrestrial civilizations to indirectly converse with us and learn about us without being hindered by the vast distances of space and its corresponding human lifetime delays in communication. Aliens could learn one of our languages, ask the LLM questions about us and receive replies that are representative of humanity.
This is a radical and potentially risky idea because unfriendly aliens could misuse this information. Nevertheless, it’s a discussion worth starting, given recent discoveries. Space telescopes such as Kepler and the Transiting Exoplanet Survey Satellite (TESS) have revealed that our galaxy is full of exoplanets, and it’s estimated that at least 300 million are similar to Earth and potentially harbor liquid water. We believe that several of these worlds could host technological civilizations curious to meet us and learn about us.
Astronomer Frank Drake, the father of SETI and METI, understood that our galaxy could host other technological civilizations. He initiated Project Ozma in the 1960s, listening to two nearby stars, and later led the development of the Arecibo message in 1974. This message included basic mathematical and chemical data, a depiction of DNA and information about the human population, our solar system and our technology. Comprising only 1,679 binary digits, the message was inspired by the pixelated screens of early personal computers. If Drake were alive today, he would no doubt recognize AI’s potential to enhance communication with extraterrestrial intelligences.
For interplanetary communication, smaller, open-source language models, such as Meta’s Llama-3-70B and Mistral AI’s Mixtral 8x22B, would allow for training with curated datasets. Typically Llama-3-70B is around 130 gigabytes in size, which is a bit hefty to transmit across light-years without errors. With a technique called quantization, however, we could squeeze it into a few gigabytes while maintaining its performance. Crucially for interstellar communication, it would allow the AI to run on its own without an Internet connection.
Assuming we want to send these LLMs to extraterrestrials, we have two primary technologies: radio communication, which is broad and slow, and laser communication, which is directional and fast. By radio, NASA’s Lunar Reconnaissance Orbiter achieves data-transmission rates of up to 100 megabytes per second (Mbps) for downlink. That means it would take about half hour to transmit the entire full-size Llama-3-70B model to the moon. The Lunar Laser Communications Demonstration has achieved data rates of 622 Mbps, which would reduce the transmission time to about five minutes.
Reaching advanced civilizations beyond the moon involves addressing the immense distances between stars, signal attenuation and current technological limits. For instance, NASA’s Psyche mission, which launched last year toward a metal-rich asteroid, is equipped with a prototype laser interplanetary communication device. That laser technology has achieved a data rate of a few hundred Mbps. Interstellar communication with current technology would likely drop to 100 bits per second, however, resulting in a transmission time of hundreds of years to send an AI to Alpha Centauri, our nearest neighboring star system, lying a little more than four light-years from Earth.
We could instead transmit a well-curated and smaller model, as discussed above, of only a few gigabytes in size in less than 20 years, making it a feasible project for humanity. It would not only generate text but also images and sounds. Its content, personality and tone should be determined by researchers, philosophers, historians and other experts to represent humanity at large.
Using more powerful lasers to increase transmission rates is another approach. By combining multiple 10-kilowatt lasers to achieve a 100-gigawatt-strength transmitter, as proposed by one of the Breakthrough Initiatives, we could transmit large amounts of data across several light-years. Another advanced project suggests using the sun as a gravitational lens to amplify signals and create a superfast interstellar communication system. We’d need a probe equipped with a laser at 550 astronomical units, or 82 billion kilometers, from the sun, beyond the orbit of Pluto.
Another, albeit more antiquated, solution is to equip every space mission with a simple ruggedized onboard computer containing a well-curated LLM and an interface for communicating with it. This digital time-capsule approach will both pay tribute to, and naturally extend the legacy of, the analog Voyager golden record. That gold-plated copper disk included carefully selected images, sounds, music and messages intended to communicate the story of our world to extraterrestrials.
In the distant future, an intelligent civilization may encounter a spacecraft equipped with an ancient computer or receive a signal containing instructions to rebuild our AIs. This indirect communication with us could reveal our past, our ambitions and our essence as a technological species, showing these life-forms that they were not alone and that a civilization of humans once existed—perhaps not so different from their own and perhaps still around.
The time has come to leverage our AI advancements for a new era of interstellar communication. By sending well-curated large language models into the cosmos, we will open the door to unprecedented exchanges with extraterrestrial intelligences, ensuring that our legacy endures, even when we might not.