Painstakingly, I was putting the finishing touches on my analog geometric art – when this video popped into several of my social media feeds:
We Need To Talk About AI… by Cool Worlds Podcast
Cool Worlds Podcast is run by a team at Columbia University and usually features guests interviewed by physicist David Kipping. This time, he has no guest but is reporting on a meeting he attended at the Institute of Advanced Study (IAS) in Princeton – professional home to “some of the smartest scientists on the planet”.
I will summarize some high lights as neutrally as I can. Or low downs, as Kipping has called them on X / Twitter:
I attended a meeting last week that was one of the most shocking of my career. Here’s the low down
But if you are interested in science or AI, or the future of humanity in general – go watch it yourself!
This was an informal meeting about the role of AI in science – computational physics / astrophysics specifically. Kipping mostly quotes (anonymized) senior faculty of the IAS. What shocked him the broad agreement among the audience of scientists.
The anonymous senior scientist stated that agentic AI is already capable of doing 90% of their work.
They had surrendered all their digital life to the bots, giving up on privacy: Email, files, even sudo (“super user”) access to their machines – and one third of the room agreed. Re privacy implications the attitude was: I do not care – the advantages are so great.
Astrophysics relies on numerical simulations – modeling processes in the universe. In the past, this had required a unique skill set, years of mathematical training, coding experience, and outstanding analytical skills.
However, the IAS physcist explained that the large language models (LLMs) have already achieved total coding supremacy. Kipping notes that he has also been surprised by the models’ abstract mathematical capabilities: For example, ChatGPT had been able to solve an integral symbolically, providing all the steps of the derivation – while the symbolic mathematics software Mathematica had not been able to do so.
A caveat has to be added here: Software development / “programming” is not the same as “software engineering”. As also the credible comments on this Youtube video suggest, total coding supremacy has not at all been achieved in typical corporate / enterprise environments. AI might be excellent in solving a particular task (even a very tricky one). But it is a very different thing to work on a large legacy code base that is maintained by a large – globally dispersed – team, while fulfilling for example regulatory requirements and keep the code maintainable. Any type of engineering is about finding the best technical solution while meeting lots of other constraints.
Kipping mentions that scientists are especially vulnerable. The superior technical skills that scientists needed in the past – technical skills and raw intellectual power – might not be what will set you apart in the future. The scientist of the future might be more of a manager / mentor / advisor, delegating to their bots in the way senior experts work with graduate students today.
This aligns with my anecdotal evidence that AI has been embraced early and happily especially by (formerly technical) people who have moved to a more non-technical role. The managers are able to code again, sort of.
I feel this is at the core of many discussions on AI and jobs. Some of us take pride in and find fulfillment in doing the thing – technically, at the most detailed level possible. Even if this is “tedious”. For the craftsperson it is not just about the result: It is also about the process. For a “manager” it might be more about the results and about orchestrating teams and resources to achieve the results.
This is not that different from the AI art discussions: For an artist, it is not only about creating a certain type of artwork using the most efficient way possible. It is all about the process of creation by a human being.
It was interesting that the issue of “copyright” was not touched on in this video – just as it isn’t in discussions among software experts (my anecdotal evidence again). AI had been trained on blog posts, forum discussions, and publicly available documentation – but software engineers are not upset about this, in contrast to artists. Maybe because a large part of this publicly available material is open source / in the public domain? The same would apply to scientific publications and publicly documented algorithms for simulations in astrophysics.
But other objections to AI have been mentioned: Usage of energy, climate change, AI companies in the hands of a few billionaires, people losing their jobs. The response of the IAS faculty was again: I do not care – we need to stay competitive. Everybody in this room seems to have agreed that young scientists need to fully embrace AI tools early.
Kipping is hesitant to admit it, but he says he is not sure if he would work with a graduate student who rejected to work with AI. Training a young scientists appears to be costly today, like: “I would need one month to solve a problem, the graduate student will need a year, but those AI model would need only a few prompts.”
He tries to find some positive aspects, as a potential democratization of science in the sense of “Now everybody can do science with the help of AI”. On the other hand, if all the details and all the implementation is done by AI. Only AI will be able to read the scientific papers and to maintain the software and the systems in the future. David Kipping states he does not want to live in a world that feels like magic.
Technical skills would necessarily atrophy – just as our spatial navigation skills have atrophied since we have started using GPS. My impression is that this is not considered negative. It would be absurd to say: Let’s keep our map-imagining skills alive.
Or is it? It has been ten years now that Nicholas Carr had published his book on automation – The Glass Cage: How Computers are Changing us. It has been even longer he wrote about how Google is making as stupid.
I have reviewed “Glass Cage” – the book on automation – here, and added my thoughts. Quoting myself:
‘Automation’ can thus be understood in a very broad sense. I have written about Newton’s geometrical proofs that even Richard Feynman found very hard to reproduce. Now we have been spoilt by the elegant code-like symbols of calculus. Do really miss out if we not haven’t acquired such ancient skills? Carr believes so as we are human beings made to interact with the world directly, not via a cascade of devices and abstractions. A physics professor who has embarked on “a self-imposed program to learn navigation through environmental clues” finally concluded that the way he viewed the world had palpably changed. Architects felt that they needed to stay away from electronic help or bring in the computer late so that the creative process is not (mis-)guided too early. A photographer tells his story of returning to the darkroom as he felt that the painful manual process forces him to make more conscious and deliberate choices – with a deep, physical sense of presence.
…
The physicist turned stone-age pathfinder said that …
… “primal empiricism,” struck him as being “akin to what people describe as spiritual awakenings.”
We enter the realm of the non-efficient, obviously. Everybody can have their inefficient, non-AI or even non-software hobbies. But the discussion in Princeton centered around staying competitive, that is: Staying employable and being paid for providing value – by certain standards.
“Ironically”, art might be an exception: While artists are extremely vulnerable to AI as their creations are easily scraped and re-used, their results might be the only ones for which – at least some! – buyers are willing to also pay for the story and the process. Even Kipping mentions this at the end (despite considering using AI-generated visuals earlier in the video).
~
I do not have any uplifting conclusion to offer – the video had neither. I have just realized again that my practice of analog art might be an act of defiance. I do not want to stop exactly where AI started; I go back to the times before software, where the physics professor in Carr’s book found meaning.
Working on Lissajous Trilogy was the perfect antidote to AI: A drawing on canvas, representing a series of analog steps of calculation. Here is the story of The Making Of.
In a final step, I prepared a photo of the artwork for canvas prints. The original features wavy extensions of the construction lines on the sides of the canvas. To replicate this effect on canvas prints, I extended the photo with flipped versions of the drawing – and finally distorted the part on the canvas sides digitally. I applied the distortion effects manually, with a digital pen, using the iPad software Procreate, a company whose CEO famously said he f*cking hated generative AI.
Quoting from Procreate’s AI statement page:
We’re here for the humans. We’re not chasing a technology that is a moral threat to our greatest jewel: human creativity. In this technological rush, this might make us an exception or seem at risk of being left behind. But we see this road less travelled as the more exciting and fruitful one for our community.
Leave a Reply