AARON: AI Art Generator from 1973 | Hacker News Reader

AARON: AI Art Generator from 1973

katevassgalerie.com

174 pointssmusamashah22 comments

This used to be downloadable on windows as a screensaver/standalone version. Here is the gallery of images it made https://www.kurzweilcyberart.com/aaron/static.html The windows version can be found on Archive[1][2] and it works fine on Windows 10. Generated images are stored in temp dir in text format. Made a tool to view and save them in JavaScript[3].

Images generated by this are very coherent and artistic. You have to try this yourself to witness it. Obviously the windows version is not the full thing but it does create people, plants, walls with paintings on them and does this very well.

1: https://archive.org/details/AARON-2-0-01420

2: https://archive.org/details/kurzweilcyberart_aaron

3: https://github.com/SMUsamaShah/Kurzweil-CyberArt-AARON-Image...

Wow since that page was made we’ve come full circle on “ your web browser is not Java-enabled”

It's crazy that this software (and its installer) from 2001 runs out of the box on Windows 11.

I first learned of AARON 20 years ago, when prints were being handed out at a Lisp conference in New York. (Harold Cohen switched from C to Lisp so that he could add colour to its artwork.)

There's a lot more information about AARON on its web page: http://www.aaronshome.com/aaron/publications/index.html

There's a good book from 1990 with lots of detail, history, and examples of AARON and Harold Cohen's work: Aaron's Code by Pamela McCorduck. Recommended reading for anyone into generative art, or more generally the intersection of computation and art.

Out of print, but available second-hand on Amazon: https://www.amazon.com/Aarons-Code-Meta-Art-Artificial-Intel...

For those in (or soon traveling to) Southern California, the LA County Museum of Art is now featuring a lovely computer art exhibit that includes physical outputs from Harold Cohen, among other works from algorithmic luminaries of art: https://www.lacma.org/art/exhibition/coded-art-enters-comput...

Only tangentially related, but I think it would be fascinating to see an image-generating model trained only on imagery from before a certain date.

Or, for that matter, a LLM only trained on content before a certain date, presumably exhibiting attitudes of the time.

I'd say that's very manageable; there's open source tools and datasets to train a tool just like that. I do wonder if generators like dall-e would generate different results then, that is, if you ask it to do a work in a certain style without any other style in its databases.

I’m intrigued by how he achieved the hand-drawn sketchy line style look in his early 80s work when so much computer art was geometric lines or Mandelbrot sets

Some of them seem to have been drawn with a plotter.

I saw several plotter-based AARON demos in the 90s. It was interesting and the sketchy line style was nice but not itself impressive. There were many add-ons to CAD software for instance to render drawings in various "hand drawn" looks for reviewing drafts with clients or delivering them as part of the visualizations. I even remember a pen plotter that my dad had that had thick highlighters in some slots for doing color elevation renders

> His creation of this program grew out of what he has described as his need 'to understand what art is'

Is not this in a way also the approach of Marvin Minsky?

Ray Kurzweil has a long-standing interest in Harold Cohen, and is at least partially responsible for the creation of the Windows software version of AARON.

How do I know this? I worked for Ray 25 years ago & helped promote AARON. Our ancient & untouched 2001 website remains available today at https://www.kurzweilcyberart.com/

“AARON cannot learn new styles or imagery on its own; each new capability must be hand-coded by Cohen.”

So… not AI?

Artificial Intelligence != Machine Learning

More precisely, ML ⊂ AI

When I was studying AI at university 20 years ago, there was a running joke in the AI department that things were AI until they worked reliably, then they became Computer Science and the AI folks never got the credit.

Like the best jokes there was a grain of truth to it: novel search algorithms, expert systems, heuristic systems, fuzzy logic etc. were all ‘AI’ to begin with. The field really only means trying to get computers to behave intelligently. But people want AI to be somehow mysterious. The current large neural networks are inscrutable enough that the joke may be a reached its limit now (I’m not sure of that though…)

> inscrutable enough

The "problem of transparency" is also relevant to checking that the solutions proposed by the AI system actually are "intelligent".

Decades ago, it was common that "AI" specifically referred to rule-based, symbolic systems. If you were working on "neural networks" then you would explicitly say so.

Isn't AI just a fancy word for a ton of if statements?

> So…

Simulating intelligence → is AI

That new capabilities must be hardcoded just means the AI is not "general".

Can someone explain how this works? Read the article and have no clue.

What an interesting conference! Thanks for sharing

Reply on news.ycombinator.com