Tech Leaders Wrestle With the Opportunity and Risk of Using AI in Satellite Networks

Artificial Intelligence (AI) has been one of the most talked about topics throughout the year. At SATELLITE 2024, satellite players debated the role that AI plays in global satellite technology.

One of the interesting new companies in this area is Space42, which will be the result of the merger between Yahsat and Bayanat, to form a global AI space company. Karim Michel Sabbagh, managing director designate of Space42 said the company sees AI “as part of the company’s DNA” going forward.

Sabbagh believes there are a number of issues and challenges when looking at implementing a successful AI strategy. He spoke about the data reliability question, and when introducing AI into Geostationary Orbit (GEO) and Low-Earth Orbit (LEO), and having a lot of historical data, and whether that data will be reliable enough. He also spoke about the line between autonomy and control, a key focus of the discussion. “When do you start trusting AI to do some of the decisions? Where do you draw the line?”

He also spoke about safety and reliability, and how you could “do a zillion things” right, but get one thing wrong, and the need to mitigate this. Being successful means doing away with the industry norms of the last 60 years.

“The end goal is can you achieve an adaptive communications system? How quickly can we deploy AI and let go of control? The industry was based on the parameter of control. You are wiping away 60 years of practices,” Sabbagh added.

Sabbagh also bought up the issue of standards. He said, “We are putting our foot on the pedal in terms of AI. You could argue we are dealing with a far greater risk. The one thing I would say is that we need to bring standards in what we are doing.”

Mark Lorden, director of AI Customer Success for Sidus Space, spoke about giving the power of space data into the hands of users within seconds. Lorden bought up a number of interesting questions. “How can we trust the hardware that is running these? We build in protections in the hardware. There are additional protections through software. We use AI on top of that to find faults in potential data. As the commercial market expands, we plan to expand with it. We may be running multiple missions at different times. We will need a software system. AI/ML will become essential.”

Lorden said it is really about creating an environment where a wide variety of AI and non-AI applications can be hosted and decisions can be made on that analysis, and where there might be models cross checking each other. “The issue of constant work to develop and maintain trust in our AI systems is key. I think this technology will take our capabilities to places we have not conceived of. It is a very exciting time to work in space,” he added.

Shadrach Benny Retnamony, founder and CEO of Neuron said the company is a connectivity management platform, helping with data driven procurement. He said at the heart of AI is data veracity and data quality as the company has 20 million records a day. “Labeled data is gold is the most important aspect in this whole process. We have to gain confidence. AI/ML unlocks a number of use cases. A lot of engineers who have done this for decades don’t trust AI; they still want control. We haven’t completely turned it off. However, system design is getting wider than before. As applications evolve, the drive to AI/ML will be inevitable.”

Retnamony also brought up the issue of standards saying at some point, there will have to be standards around this. He admitted contracts with AI elements have become more complex. “At some point, there has to be some form of standards around this. “Every contract with an AI component takes three times longer to get done. We don’t want to be entangled in our own mess,” he added.

Brad Bode, CTO of Atlas Space Operations, a company that offers ground software-as-a-service said the company has to be very careful because it doesn’t want to be spamming its end users with “false positives,” which highlight problems where there aren’t any. However, Bode believes there is an inevitability about it impacting companies like Atlas Space Operations.

“Speaking from the ground side, ML will speed things up. If you have a 300 satellite constellation, you are going to have leverage ML. To cede control over to AI/ML is a scary proposition at this stage,” he says. “There is legal risk, security risk, hallucination risk. We can’t have something where something appears to have gone wrong, when it hasn’t. You can die if you don’t use AI in some ways, but you can be successful if you learn to swim.” VS