launch

In a world where machines can do many things as well as humans, one would like to hope there remain enclaves of human endeavour to which they simply cannot aspire.

Art, literature, poetry, music – surely a mere computer without world experience, moods, memories and downright human fallibility cannot create these.

Meet Aaron, a computer program that has been painting since the 1970s – big dramatic, colourful pieces that would not look out of place in a gallery.

The “paintings” Aaron does are realised mainly via a computer program and created on a screen although, when his work began being exhibited, a painting machine was constructed to support the program with real brushes and paint.

Aaron does not work alone of course. His painting companion is Harold Cohen, who has “spent half my life trying to get a computer program to do what only rather talented human beings can do”.

A painter himself, he became interested in programming in the late 1960s at the same time as he was pondering his own art and asking whether it was possible to devise a set of rules and then “almost without thinking” make the painting by following the rules.

The programming behind Aaron – written in LISP, which was invented by one of the founding fathers of artificial intelligence, John McCarthy, back in the 1960s – attempts to do just that.

Some of Aaron’s knowledge is about the position of body parts and how they fit together, while some of the other rules are decided by the machine.

It actually “knows” very little about the world – it recognises the shape of people, potted plants, trees and simple objects such as boxes and tables. Instead of teaching it ever more things, Mr Cohen has concentrated on making it “draw better”.

And it has been a great pupil.

“The machine had become a world-class colourist – it was much more adventurous in terms of colour than I was,” he told the BBC.

For many years the two worked side by side, but gradually Mr Cohen began having doubts about the partnership.

First, he decided to abandon the painting machine that was hooked up to Aaron.

It had been, he told the BBC, too cumbersome and had led too many commentators to regard the project as a robot rather than clever programming, which had irked him.

But he was also having bigger doubts – Aaron was both becoming too independent and also revealing some serious limitations.

“I dreamed up a very simple algorithm and it obviously embodied a great deal of knowledge, but when I looked at the output I didn’t remember doing it because I hadn’t done it,” he told the BBC.

“It no longer needed me. I never intended to leave everything to the program, but it gradually came to me that it could do without me.

“It had become autonomous enough to disturb the guy who wrote the program.”

What had originally been conceived as a team project was becoming something else entirely.

“Works of art are like children – they go out into the world but you always have a connection to them and I’d lost that connection. I felt out in the cold,” he told the BBC.

At the same time though it was clear that Aaron, while excelling at colouring, was never going to be truly creative.

“It was not that autonomous, and the very little dose of autonomy that Aaron had only related to colour,” Mr Cohen said.

It led him to question whether a creative AI was ever possible.

“I don’t deny the possibility that, at some point in the future, a machine can make something approaching art – but it is going to be a lot more complex than teaching a car to drive around a city without a driver, and it isn’t going to happen next Wednesday or even in what is left of this century,” he told the BBC.

The partnership with Aaron is still “alive and well”, but it has changed.

Now, Aaron concentrates on the drawing, while Mr Cohen does the painting. And these days, he does it digitally, using a giant touchscreen rather than real paint – perhaps in a nod to the machine he created.