The telescope that points itself

Lisa Lock
Scientific Editor
Andrew Zinin
Chief Editor
I'll admit to using AI for things it was probably never designed for. When my van threw a fault I couldn't immediately figure out—one that seems notorious for trips back and forth to a repair shop—I ended up in a long (but cheaper) back-and-forth with an AI system, working through the symptoms. The same happened with a run of gutters and with stud framing in my house. It isn't magic, and perhaps more often than I'd like, it's wrong, but as a way of thinking through a problem with something that's read more manuals than I ever will, it has earned its place in the toolbox. So I read this next story with some sympathy and interest.
AI has now been turned to the sky because astronomers have a scheduling problem, and it's a harder one than mine. Time on a major telescope is rationed. You might wait months for a handful of hours, and those nights arrive with conditions you can't control. The moon might be too bright for a faint target. The seeing might be poor. Clouds might roll across the part of the sky you'd planned to work in. Many of these can be planned around, but it's a tricky, moving beast, and every hour is a judgment call about which of your targets is worth the conditions you've actually got. Getting it wrong means soft images, washed-out data and a wait until the schedule comes around again.
Alex Drlica-Wagner of Fermilab and the University of Chicago, and Aravindan Vijayaraghavan at Northwestern, have built something to make that decision. Working through the SkAI institute, they trained a deep learning model on years of observations from the Dark Energy Survey. They didn't encode the rules astronomers use. Instead, they showed the model where the telescope was pointing at a given moment, asked it to predict what happened next, then compared its guess with what the humans actually did and made it correct itself. Repeat that a few million times, and the model works out how moonlight and the atmosphere shape a good decision without ever being told that they do.
Having now learned how "astronomers do it," they scheduled two real observing campaigns this spring and summer on the Blanco 4-meter at Cerro Tololo in Chile. The system was able to drive the 570-megapixel Dark Energy Camera, producing the plan and then adapting it live as conditions changed.
The honest assessment from the team is that it currently performs about as well as a human. That sounds modest until you consider that matching an experienced astronomer was the entire goal of a first deployment. The next objective is to exceed them by trying strategies no human would think to try.
There's a practical reason this matters now: With the Vera Rubin Observatory about to start producing more data than anything before it, other telescopes will need to react quickly to what it finds. Scheduling by hand doesn't scale to that, but an AI tool can react and adapt at speeds no human can match.
Drlica-Wagner puts the point better than I could: "... automate the operational work, and astronomers get their time back for the interesting part." Which is roughly what I want from my AI system, although my ambitions stop at the van for now.
Provided by Universe Today

* Alex Drlica-Wagner and Aravindan Vijayaraghavan built a deep learning model using Dark Energy Survey observations.
* The model was trained to predict outcomes based on telescope positioning.
* The training involved comparing the model's predictions with actual human scheduling choices, allowing the model to self-correct.
* The system utilized real-time data regarding moonlight and atmospheric conditions to inform scheduling decisions without explicit rule encoding by humans.
* The system was used to schedule two observing campaigns on the Blanco 4-meter telescope in Chile during spring and summer.
* The AI drove the 570-megapixel Dark Energy Camera.
* The team reported the current performance of the system is about as good as a human astronomer.
* The goal for future development is to exceed human performance by testing strategies humans might not consider.

Astronomers have developed a deep learning model trained on Dark Energy Survey observations to schedule telescope time, addressing scheduling constraints imposed by limited observing windows and variable atmospheric conditions. The system operates by predicting future conditions based on current positions, comparing its predictions against human actions, and correcting itself through repeated iteration. This process allows the system to navigate unpredictable variables like moonlight and atmospheric seeing to optimize target selection. The model was successfully deployed in real observing campaigns on the Blanco 4-meter telescope in Chile, adapting scheduling for the Dark Energy Camera as conditions changed. The team assessed the current performance of the AI to be comparable to a human astronomer. The potential benefit is the ability to automate operational scheduling, aiming to free up human astronomers for more interpretive tasks.

The narrative positions AI as an external tool solving complex, multi-variable decision-making problems inherent in observational astronomy, moving from mere pattern recognition to operational autonomy. The core tension lies in the transition from human judgment—which is constrained by experience but capable of intuition—to algorithmic optimization based on learned correlation. This raises questions about what constitutes "astronomical insight" when an opaque system achieves empirically equivalent results. The implication for human agency centers on redefining the value of expertise: if the AI can match or exceed human performance in scheduling, where does the unique value of human contextual understanding reside? Furthermore, the push to automate operational work suggests a pattern of externalizing labor and decision-making, prompting a need to analyze who benefits from this efficiency—the astronomers, the institutions funding the research, or the infrastructure developers. The focus on reacting quickly to new data from facilities like the Vera Rubin Observatory highlights an accelerating need for computational adaptability beyond what traditional human scheduling methods can handle, suggesting a systemic shift toward computationally mediated knowledge processing in scientific practice.

The telescope that points itself

Facts Only

Executive Summary

Full Take

Sentinel — Human

The telescope that points itself

Facts Only

Executive Summary

Full Take

Sentinel — Human