Building space-based AI. Designing new molecules with quantum computers. Why pink noise is bad for your sleep. All that and more in this week’s edition of The Prototype. To get it in your inbox, sign up here.

Space is the latest place for AI, with major tech companies like SpaceX, Nvidia, Google, Blue Origin and more announcing plans for space-based data centers. But as I discussed a few weeks ago, orbital data centers face major economic and engineering hurdles thanks to the laws of physics.

San Francisco-based Loft Orbital, which builds standardized satellite platforms for a variety of customers, is offering up a different vision of AI in space, at least in the near term. Later this year, it plans to launch a 10-satellite constellation focused on doing AI in space for real-world applications like environmental monitoring or law enforcement.

It won’t be easy, Paul Lasserre, who heads the project, told me. Doing AI in space with current technology comes with significant constraints. One is power–the satellites are only going to have about 500 watts available to them. That’s less power than a high-end gaming computer. Plus all the chips and hardware need to be built to withstand radiation in space. On top of all that, there are severe limitations in bandwidth that slow the speed of data uploading and downloading.

The upshot of these constraints is that Loft is not going to be using large models like Claude or ChatGPT in space. Instead, it will be working with lightweight AI models that can run on the satellites themselves. These models will take information from cameras and other sensors to alert people on the ground to actionable information, such as possible wildfires or piracy. That could mean immediate action by law enforcement, or they might instead opt to prioritize downloading satellite data about that area to crunch at Earth-bound data centers.

Lasserre acknowledges that because these models are smaller, they could be prone to false positives. But he likens the approach to early AI systems in medical imaging, which were used to prioritize patients for doctors to examine. Even when they weren’t always right about the risk, they made it more likely that doctors were seeing patients with more urgent needs.

In addition to providing these services to customers, Loft is also planning an AI application marketplace on its satellites. This would enable government and commercial partners to develop and deploy their own algorithms on Loft’s systems. Lasserre also said the company is working on developing better visual AI, which could combine data from different kinds of sensors to better understand what’s happening on the Earth’s surface.

“This is about turning satellites........

© Forbes