AI and Humans - Together, We Create

Artificial Intelligence gathers experience, analyzes structures, then detaches itself from the past and creates something new and surprising on this basis. A creative person doesn't do it any differently.

Quote from Matthias Bethge, taken and translated from "Die Kunst der Zukunft - über den Traum von der kreativen Maschine", Hanno Rauterberg, Suhrkamp.

I often end up in discussions about the creative nature of computers. Especially that computational creativity is inferior to human creativity. I am not sure if I can subscribe to this point of view.

Human and computational creativity are different in many ways. The realm of experiences is not the same. Humans get inspiration from everything. Computers from data. But... human and computational creativity have more commonalities than differences.

In the last decade, it became clear that the creative power of Deep Neural Networks is growing extensively. What began very humbly with small networks predicting the next note for single-voice songs, has now become Large Language Models composing multi-track music. Both in symbolic music and in the audio domain. We are talking about Deep Neural Networks trained on 200K MIDI files, for example, with even bigger datasets already visible on the horizon.

200K MIDI files is a lot. And this is where computational creativity can really shine. Computers are excellent at analyzing the underlying structures of any data. Music is no exception. One might argue that the mathematical properties make it even easier for computers. Let us leave this for a future discussion. I have seen Neural Networks learn from music in a matter of hours and days. Definitely, time spans that everyone would feel comfortable leaving the computer alone and doing its thing. Still, the results were breathtaking, to say the least.

In a previous article, I mentioned the focal lens that Neural Networks can be for any dataset. A Neural Network is capable of extracting the underlying patterns and rules of data and can then be asked to do something useful with it. Composing is just one use case out of many.

Certainly, this is similar to how humans work creatively. Humans consume musical material, get inspired, replicate and change musical ideas, and recombine them into new songs. Inspired plagiarism, with just enough change to not be a copy.

For the artist, Deep Neural Networks offer another layer of creative work. From the literal beginning to the end. Gathering a musical dataset, preprocessing it into a form that a Neural Network can understand, training different network architectures, finding a very good network, integrating the network into a useful tool, and finally using the tool to compose music. What sounds like building a pyramid is something that in the near future, more people would be able to do. Especially without having a Ph.D. in Computer Science.

I feel that the creative landscape is about to change. Are we going to see Advanced Music like we saw Advanced Chess? Advanced Music in a sense that humans compose WITH computers to create music that has not existed yet? Combining the goal-driven and cross-modal motivation of humans with the cool analytics of Artificial Intelligence? I believe so.

Photo by "My Life Through A Lens" on Unsplash