Scientists have built an AI inspired by HAL 9000 and what could go wrong, really

Humans go places. NASA's latest plan is to launch crewed missions on Mars in the 2030s. We will need the most advanced and reliable space technology to help us do this safely.

That's where HAL 9000 – the crazy and insane killbot of 2001: The Space Odyssey – intervene. Believe it or not, the most famous sci-fi murder machine has been the inspiring foundation of a new cognitive computer system resembling HAL, designed to autonomously manage planetary space stations for a day.

If you think oh my god no god please god no do not worry

The Houston-based developer of AI and robotics Pete Bonasso of TRACLabs, explains that his new prototype CASE ("cognitive architecture for space agents") mimics HAL in a purely technological sense – that is, say. less paranoia and betrayal.

Aside from these psychological weaknesses, the power and reach of the iconic character impressed Bonasso half a century ago.

"I saw Stanley Kubrick 2001: The Space Odyssey during my last year at West Point in 1968, "recalls Bonasso in a new document describing the research.

At that time, the engineering student only had access to one computer (the university had one): a General Electric 225 had only 125 kilobytes of memory.

Despite the limitations of the machine, Bonasso found a way to program it to play virtual pool games, but witnessing HAL was a revelation of a whole new level.

"When I saw 2001, I knew I had to turn the computer into another being, a being like HAL 9000," says Bonasso.

Decades later, the programmer has actually achieved this goal.

The prototype AI developed by Bonasso has so far only managed a planetary space station in a four-hour simulation, but early results are promising: in the experiment , no virtual astronaut has been ruthlessly dubbed and shot.

A rendering of the CASE controls of the virtual global base (Bonasso / Science Robotics)

"Our colleagues and our NASA counterparts are not afraid that our HAL will go out of control," Bonasso told Space.com.

"It's because he can not do anything for which he is not programmed."

CASE can plan and control technology activities and operations to maintain a colony base 24 hours a day.

The system is composed of three layers. The first control materials, such as survival systems, feeding systems, planetary rovers and, in theory, the doors of pod bays.

The second layer runs the software to run this hardware infrastructure, while the third one calculates solutions around daily goals, including routine maintenance and tasks around the station, as well as to respond to potential emergencies, such as that gas leaks, engines broken down or global problems. Sandstorm.

Presumably, the third layer of CASE also comes into play if something resembling a haunting extraterrestrial monolith appears unexpectedly near the settlers, although the newspaper does not mention such an eventuality.

In addition to the layer system, the agent also features an ontological system capable of reasoning and making judgments about the data (knowledge) that he possesses, as well as interfaces allowing the man to 39 interact with AI, including visual displays and a dialogue manager answering speech.

Of course, most of these features are contained in a prototype that practically exists only for the moment, but Bonasso and his fellow TRACLabs researchers – who are looking for systems for customers such as NASA and others who are not familiar with it. other government agencies – hope to integrate testing in the real world soon.

"We are working with what NASA calls analogues, places where humans gather and claim to live on a distant planet or on the moon," the paper said.

"We hope to work slowly, piece by piece, in one or more analogues to determine its value for future space expeditions."

The thought goes, if such real-world tests with CASE prove successful – and if this HAL-spawn is deployed on the Moon or Mars – this would already have a good starting point to support a human crew ", wrote Bonasso.

Again, no need to worry about that. Really. Honest.

This system has strict limits, what he can do and what he knows.

"His abilities are very narrow, focused on events happening on a global basis," Bonasso told Space.com.

"Even though he can maintain the functioning of the survival systems, he has no idea who won the last presidential election."

Envious?

The results are reported in Robotic science.