Ourtificial Intelligence

The ethics of diffusion models

Bugs Bunny capitalist & Bugs Bunny communist arguing over artificial intelligence

In an episode of BBC’s “The Code” they asked 160 people to guess how many jelly beans were in a jar, the guesses varied hugely from a low of 400 to a high of 50,000. Only four guesses came remotely close to the actual number 4510; however, if you took the mean average of their guesses they came out to 4514. Despite their wildly different answers, their cumulative conjecture came extremely close to the correct number.

Although together they reached the correct answer I don’t believe it was the most creative one. Certainly, the woman who guessed 50,000 beans could fit in such a small receptacle had some more wildly imaginative side than those with more sober accurate guesses.

Artificial intelligence can currently be imagined in this way of mean averaging. It’s given 10,000 pictures of a chicken and starts to hone in on this averaged construct of chicken. It learns through even distribution how to draw the most chicken chicken. This is obviously a vast oversimplification but for the most part accurate.

Creativity exists somewhere beyond just technical ability. It’s hard to define, creativity can be many things, but as I see it creativity is spotting a pattern where there is not one; linking two or more foreign concepts in an abstract way that might enhance each of them. Mowgli is an orphan 1 who is raised by wolves 2 at a glance these things don’t go together yet Rudyard Kipling manages to marry the two concepts in a compelling world where they do.

When AI is confronted with two prompts that it hasn’t seen together before it tends to draw them both individually in this case toe and knife rarely understanding how it might merge the concepts. A child given the same prompt would likely come up with a far more interesting amalgamation of the two but lack the skill to render it in any high fidelity.

So it seems there is ultimately some confusion here that lies between creativity and mastery of a tool; an art conservator, restorer or forger are all craftsmen of the highest order with great technical ability. Yet in spite of this, they are never considered creatives like the artists whose work they are emulating.

Highly regarded artists have always used others in the creation of their work, for example, we are still very unclear on which of Caravaggio’s paintings he produced himself and which were from his students as he took credit and signed work that wasn’t his. This practice hasn’t died out, in fact almost all world-level fine artists today work with a team executing their creative vision, with some being accused of having little to no hand in their work.

Imagine for a second that this moment is not some dystopian nightmare but a societal shift to the great democratization of art. That each of us might have a digital student at home we can train in our style. A digital hand to help us paint so that we too can sign the corner.

I believe both are true at once. Much like with the V-2 rocket one man’s bomb can be another man’s spaceship. Technology isn’t inherently good or bad, what we do with it is.

The problem instead lies in the world that has created such a volatile position for artists that they must gate-keep their least creative talent (technical ability) to keep bread on the table.

So back to those 10,000 pictures of chickens. Where did they come from? Well more often than not they came from you and me, most major AI systems have been trained on our data without our permission. AI itself isn’t the problem but the ethics of the companies that make it. In spite of being trained on our cumulative effort, the product of this labor has been kept from us entirely. No one has seen any royalties for these pictures of chickens, despite big tech companies making billions in venture capital over them.

This is why we at Toe Knife are working on our own ethical diffusion model. Made only from our artwork, our pictures, and donations from our community. Where donors will have a say in how this model is licensed and distributed along with access to use it themselves. Big business as with so many other emergent technologies has swooped in to preside over and privatize this open space into a for-profit model. So let’s make artificial intelligence into ourtificial intelligence, the work of the people for the people.


Leave a Reply

Your email address will not be published. Required fields are marked *