life-drawing and astrophysics

.

Can You Make Me More Capable? Of Art, Astrophysics and AI

Since the pandemic, I have returned to an old passion of mine–drawing, especially life drawing. I have always loved it. One year I even won my high school’s art award. Over the years, academic work had crowded it out, except for the occasional conference poster or sketching faculty during meetings. During one of the dark pandemic days I wondered if there were any “drop-in” life drawing classes nearby. It turned out there were two, one in a big old house within walking distance (this was NYC). What a great way to overcome some of the social isolation of those years—even when masked. These classes were pretty full, and post pandemic, the number of offerings has grown by leaps and bounds. This seemed somewhat paradoxical to me. At a time when AI can generate beautiful drawings and paintings in virtually any style within seconds, people were still spending hours struggling to sketch a live model. Why? Clearly because it is great fun, relaxing, and an enjoyable (and, in my case, unusual) social activity; getting better at it expands our ability to impart our own creative perspectives. It’s not so much the drawing we want, but the creative power to create them in unique ways.

But I have been struck by something that was different from my drawing classes from years ago, especially during the instructed sessions. Rather than focusing so much on accuracy, the instructors were allowing, even encouraging stylistic effects and even exaggeration. Of course we were blown away by the technical skill around us, but it seemed that the most appreciated drawings were often those that distorted proportions just a bit, exaggerated a gesture or expression, communicating something about the model more effectively. Perhaps the value of subtle touches of experience and intuition, rather than realism, never absent, is at least partly because of AI. In any event, AI drawing agents and painting robots are not about to usurp humans. If only I could practice for a couple of years to attain the higher level of technical skill I would need!

I recently read a Science article, “The Last Astronomers” (June 4, 2026) about AI in astrophysics. The article describes how at a Harvard lab machine learning was originally merely supposed to automate the tedious parts of research, leaving humans to enjoy what one astronomer called “the fun part of physics: honing scientific questions.” Instead, AI is beginning to solve problems that many researchers regarded as the heart of scientific discovery–the fun stuff, as they put it. Many fear that if unleashed in all parts of the scientific process, AI tools could lead to nothing less than the death of astrophysics as a human endeavor. “A lot of people think that it’s too late to intervene—we’re done,” said a computational astrophysicist at New York University. “Anyone working in astrophysics,” he said, “is someone who wants to do astrophysics, not someone who wants to learn the answers.”

The article continues:

“The second cluster of worries involves what researchers outside of astronomy have termed ‘deskilling’ or, more bleakly, ‘cognitive surrender.’ What if AI-dependent astrophysicists, especially young ones, lose or never build their own math, coding, and reasoning skills? ‘At that point, we’ve just completely selected for the disappearance of science in 50 years, because nobody will know how to do anything,’ one of those interviewed for the article said.”

One of the Astro-AI researchers described how “she wanted to understand how the spiral arms of a distant galaxy were moving. But isolating just that motion from other patterns imparted into her data by the spin and the geometry of that distant galaxy had thwarted her group for years. She asked ChatGPT, which resolved the problem in a few minutes.” Wow! That’s impressive. Her research group is planning several papers on the resulting data set, but nothing is said about whether the AI explained why the method worked, or more specifically how the desired motion could be isolated from the other effects, improving human capability for the next time a similar obstacle arises. Did the researchers simply obtain an answer? Or did they come away understanding why the method worked? Did the AI leave them better equipped to solve the next similar problem (building a kind of error repertoire)?

I really don’t know but maybe we’re benchmarking the wrong skills. Suppose the benchmark were instead:

After six months of working with this AI, how much has the human improved with a particular type of problem? Are the scientists themselves becoming better theorists, or even better at getting the best out of AI itself? This may well be beyond the capability of current AI.

This brings me back to the life drawing sessions, and there, at least, I don’t think it is that far-fetched. I can imagine the people now building AI image systems eventually creating a tool that improves my drawing instead of simply producing drawings for me.

It might say something like:

“I’ve watched you draw twenty figures. You consistently underestimate the apparent length of limbs pointing toward you. Here’s the correction. Try tracing it once. Good. Now draw three more examples from slightly different angles. Better. Now twenty more.”

It would remember every drawing I had made over a period. It would recognize recurring mistakes long before I did. It could invent exercises specifically designed to avoid them. I can well imagine AI speeding up my improvement so that what might have taken a year, say, is accomplished in only 2 or 3 months. I’d like to acquire such a machine. What do you think?

 REFERENCE:

Sokol, Joshua (2026). The Last Astronomers: Amid a flood of AI advances, astrophysicists are questioning the soul of their field. Science, 4 June 2026. (Online)