Astronauts Will Use AI in Space—Here's Why NASA Needs to Be Careful

ChatGPT-style artificial intelligence (AI) is heading to space, but experts say NASA will have to iron out some bugs in the technology before it is ready for astronauts. 

The space agency is working on a system that will allow astronauts to do their jobs using a ChatGPT-like interface. The cold, unforgiving vacuum is not a place for the errors that AI is known for, however.

"The greatest risk would be conversational agents that get it wrong—that hallucinate or that omit important details," John Hale, a professor in the Tulsa University Master of Science in Cyber Security program, told Lifewire in an email interview. "Properly training these AIs, and more importantly, training their operators on their limits and capabilities, would be supremely important."

ChatGPT in Space

NASA is reportedly devising an AI chat system for Lunar Gateway, a proposed space station. The system could use AI to detect and fix problems and to chat with astronauts. 

The use of tech like ChatGPT in space travel and satellite communication has the potential to significantly improve how we interact with remote systems and users, Bars Juhasz, an AI researcher and the founder of Undetectable AI, told Lifewire via email. This advanced AI technology could revolutionize communication protocols and redefine our methods of controlling and monitoring satellite systems.

"ChatGPT has been effectively utilized in business travel to provide real-time assistance," he added. "If this principle were applied to space travel, it could provide valuable real-time support and information to astronauts and mission control teams, thus potentially enhancing mission success rates."

Integrating a ChatGPT-like assistant in space missions could make them more interactive and intuitive, David Ciccarelli, the head of AI firm Voices.ai said in an email to Lifewire. He suggested that astronauts could ask the AI for help in real-time and receive back instructions in plain English without fumbling with clunky manuals. 

"That's a massive leap forward in terms of efficiency and ease of use," he added. 

But AI isn't just about aiding astronauts with tasks or experiments, Ciccarelli said. AI could also manage various operational aspects of the space station. "Think of it as the ultimate multitasker, handling everything from prioritizing data transmission to health management," he added. "In fact, it could tackle and resolve issues like data glitches or digital outages all by itself, without needing a human to step in."

Keeping AI From Going Rogue

If AI chatbots ever actually get into space, the tech will have to fight a bad rap. In Stanley Kubrick's "2001: A Space Odyssey," the Hal 9000 computer goes rogue and acts violently towards the spacecraft crew. 

In real life, NASA must ensure that AI doesn't give astronauts any wrong information. 

"Space travel is not an industry where whoopsies and close-enoughness are acceptable," AI expert Richard Gardner, the CEO at tech firm Modulus, told Lifewire in an email interview. 

"There could be lasting consequences to AI models creating responses that don't make sense or are flat-out incorrect. That issue is exacerbated by AI models that often struggle to decipher context. That limits its usefulness during critical moments until further research is done." 

The human element would be critical when using AI in space, Juhasz said.

"Astronauts, whether they were on the craft or piloting from the ground, would need to be trained to use [AI] in a specific way which leaves no room for hallucinations, which could look like being limited to what they are allowed to use the ChatGPT-type technology for, or the exact prompting styles they must use when interacting with the model," he added. 

Even before blasting off, AI is helping NASA rockets get off the ground. The agency uses commercial AI programs to design mission hardware like structural mounts and scaffolds. The software can save up to two-thirds of the weight compared to traditional components and can be milled by commercial vendors. The only caveat is that the AI designs look out of this world. 

"They look somewhat alien and weird," NASA research engineer Ryan McClelland said in a news release, "but once you see them in function, it really makes sense."