Self Portrait 2024 - Feedback
I did my first “self-portrait” of a robot back in 2006.
Thought it was an interesting idea and was always looking for things to paint. This was an early piece when it came to my painting robots. The robots themselves were made of wood and parts I found at the hardware store and junk yard. This particular robot head was made with a solenoid from a dishwasher. I still have it on my desk as a paper weight.
An everyone knows how over the years I kept making my robots better and better, adding more and more AI. With this early piece, the most significant AI algorithm was k-means clustering, which is such a basic algorithm, most people don’t even think of it as AI anymore.
The biggest advance in my robot’s AI came sometime around 2010 when I added cameras to watch itself paint and use what it was painting as feedback for the next strokes. Then around 2014 Deep Learning came, then GANS and the rest is what most people understand as AI Art these days.
Throughout this time I have made a number of “robot Self Portraits” - Each focusing in the advances my robot’s had recently made towards full autonomy. Most have been minted on SuperRare, though I imagine I have lost many to the trash bin in my studio.
Since 2010 all my robot art has been agentic, which is the current buzz word. And I used to describe how each painting was painted by dozens of creative agents fighting with one another for control of the brush, the winning agent making the next strokes. And I would talk about how the agents did different things like measure contrast, look at complementary colors, add a dose of green at the end, and even decide when a painting was done. Two dozen different agents fighting over the aesthetic outputs of the robots, making decisions based on live feedback, and creating art “organically”.
No one ever understood what I was talking about really. Once had a popular curator tell me “This would be interesting if you could get the robots to make their own aesthetic decisions.” To which I enthusiastically replied “They are, and I have published the code for anyone to see.“ But without giving what I said any thought they dismissively said “That’s not possible.”
So I always tried to simplify the description looking for something more easily understandable and came up with Feeback and Reflection. Instead of saying my robots had two dozen agents, I simplified it to two agents. An Expressive Painter, and a Precise Drawer. Then I split up the robot’s algorithms into either expressive and precise and had the Painting Agent and Drawing Agent go back and forth with each others work.
Here is what that looks like in my most recent work Self-Portrait 2024 - Feedback
Begin by using Deep Learning to imagine a portrait then
Paint Expressively…
Draw Precisely…
Paint Expressively…
Draw Precisely…
Paint Expressively…
many many times…
Until an independent Agent looks at the canvas and decides the painting is done.
This is how we can teach machines to see like us, but like I often say when AI begins to truly create its own, art for themselves, I doubt we will even be able to perceive it. And while I don’t know what it will create, I am almost certain it will not be a painting of its physical self.
But until then we have my robot’s self portraits where the actual robots are an important part of the creative process.
Self-Portrait 2024 - Feedback
All my painting robots have been agentic since 2010. It is interesting to watch the current evolution of the term into what people are beginning to understand agentic as. It is not just chatbots, and never has been. Agentic has always been about breaking down hard problems into smaller problems and writing AI Agents to handle the small problems. Once it can solve all the hard problems, through smaller problems, you have a mind - what is now sort of being called an Agentic Framework.
Interesting thought just now.
Have always thought of the artificially creative mind I have built as a large collection of independent creative agents, exactly as Minsky described in his seminal work Society of Mind. Should I instead be thinking of this synthetic mind as a custom Agentic Framework.
Pindar