"I can do that too!" The artificial intelligence says to humans

Hello, I am Sai Bot

"Look straight ahead and relax your shoulders" a voice speaks to me as I sit opposite Sai Bot to be drawn. Sai Bot's face is a caricature of white characters on the purple background of the screen. Sai Bot chats and chants blithely, "Is it all right if I change the colour of your lips?". Meanwhile, on a second screen in front of me, colorful strokes gradually appear, which in the end result in my abstract portrait. The result is reduced, but very interesting: my strand of hair running diagonally across my forehead is recognisable as a green line, my eyes are two small purple squares, my mouth a minimalist pink shape made of two lines.

Did an artificial intelligence actually just create a portrait of me? Sai Bot is the brainchild of Swiss artist Tobias Gutmann, who, in a poetic gesture, has taught an artificial intelligence to draw. The AI is fed with thousands of portrait drawings that Tobias Gutmann has created around the world since 2012 with his Face-o-mat project. From this data set, Sai Bot learns the formal language and repertoire of the portraits and can create its own new combinations from them.

Narrow AI

Sai Bot moves in the highly topical field of tension of the question of whether an artificial intelligence can independently produce new content and what consequences this entails. There now are AI´s that talk to us, that can write texts, compose songs and - like Sai Bot - produce images. Until a few years ago, all of these activities were still considered fields reserved for human creativity that could not be automated. Since spring 2022, several companies have released their AI image generators for testing, the best known of which is Open AI with Dall-E. In this type of software, a text called a "prompt" is used to control the output of the image. The AI draws from an extremely large pool of images from the internet and becomes more and more precise as training progresses. As impressive as the images generated in this way are, because they are new and strange, they are ultimately based on the huge input of already existing images on the internet and the precision and creativity of the prompt, both of which come from the human hand or brain. Copyright issues regarding the rights of use of these images (from the training data pool as well as those generated by the software) are as yet unresolved. The AI´s we are familiar with, which we have already integrated into our lives, such as Siri, chatbots on websites, search engines and facial recognition from cameras, are based on "Narrow AI". These programmes are designed to perform a specific task. They are capable of becoming more precise when exposed to a learning process with large amounts of data, but they cannot suddenly perform a new task.

Creativity, intuition, intention

Sai Bot's Narrow AI is based on face recognition and translates what it "sees" into the existing repertoire of shapes from Tobias' drawings. Sai Bot is, in a sense, an automated alter ego of the artist. Sai Bot draws like Tobias, but how exactly are the parallels in the working process and what are the differences from bot to human?

Unlike Tobias, Sai Bot shows no signs of fatigue. The output is always of the same quality, whereas the artist's output is highly dependent on his physical and mental state. On a good day, his portraits are better than Sai Bot's, on a bad day, Sai Bot may be more imaginative, Tobias says in conversation.

Sai Bot's creativity is based on the various possible combinations. Tobias' creativity is based on his own experiences and observations, his artistic practice and his thought processes. But how exactly do people come up with new ideas? Innovation is seen as the essence of a creative process, but this often does not happen by itself - contrary to the widespread belief of the almost spiritual inspiration of the Muses´ Kiss. Tobias mentions the morphological box of the Swiss astrophysicist Fritz Zwicky in this context. According to this method of finding ideas, each task is divided into individual sub-problems that can be combined in different ways. In this case, for example, various face shapes, eye colours, hair colours, hair lengths, lip shapes, glasses, etc., are then combined differently to produce an individual face. This breaking down of the individual features and the recombination of the individual elements brings up the question - does the human intuitively do what the machine does automatically? In this way, the two processes may not be so dissimilar. After all, Tobias' drawing process is also automated to a certain extent. There is only one attempt at the portrait, there is no new beginning, no failure, the time limit for each individual portrait is 5 minutes; if it takes longer, doubts about one's own work arise, which act as a damper. "Automatism leads to more interesting results," says the artist.

But what else do humans have over the machine? The difference lies in intentionality. Even if Sai Bot, with a lot of training, can suddenly produce new combinations in the future, there will be no awareness of the reason for one's actions. There is something intimate about portraying someone. Tobias says that he captures less the look, but rather the sound, the basic tone of a person. "It goes deeper" says a sitter after meeting Sai Bot in the Face-o-mat. To recognise and draw a face is to capture a personality. Sai Bot has yet to learn this repertoire of emotional intelligence.

Facial recognition

Tobias Gutmann's portraits are not photorealistic, but freer, more abstract interpretations of the face in its own formal language. That's what makes them so desirable, charming, exciting - you want to have your portrait taken by Tobias because you don't know exactly what will come out of it. With the portrait, you also become part of a conglomerate of images that belong together, you become part of the Face-o-mat project. And Sai Bot works in the same way. People queue up to have their portraits taken. The fascination with one's own image is great - not least due to the selfie culture, the face as the ultimate identifier, as a profile picture, as an emoji, as Face ID and thumbnail, as the ultimate "I was here" of the digital present. One's own image is something that has a tradition as long as it has absolute topicality. And so, it is not surprising that we are desperate to know how an AI sees us, that we want to recognise our fellow human beings in the portraits and that a game emerges from this, that we're desperate to come back for a second portrait with a different hairstyle or without glasses to see if Sai Bot draws a similar picture of us. In short, Sai Bot and his entire "Saiversum” puts us under its ultimate magical, artificial intelligent spell.