How do people develop AI systems?

Creative practitioners may want to work with an AI system for several reasons.

As we saw last week, the likes of Mario Klingemann and Luba Elliott are interested in utilising AI systems to push the boundaries of what we may think of as art and creativity. One reason musicians are interested in AI systems is because it allows for rapid mapping of musical parameters such as volume, pitch and rhythm, to real-time input. This allows the musician to expressively explore the musical parameter space in their performances.

But do we need to know how to code to use AI systems creatively and effectively?

Artist Rebecca Fiebrink has created The Wekinator, a tool that allows creative practitioners to utilise the power of machine learning algorithms in an interactive way without the need to manually code anything. Her approach is known as interactive machine learning (IML), and is different to the deep learning approaches you learned about previously.

_{Diagram of Wekinator instructions}

The amount of data used in IML techniques tends to be a lot smaller than deep learning. Typically, the artist or researcher will choose the algorithm they want to use. They will then collect some data and train the model using this data. The training times involved here are very quick because the amount of training data is quite small compared to deep learning. This allows the artist or researcher to quickly train a model, test the trained model, and analyse the output. If the output is not to their liking, they can quickly re-train the model, making adjustments to the training data as they see fit. This rapid, iterative approach allows for an intuitive exploration of the data and algorithms.

In the case of the Wekinator, this opens the world of machine learning to musicians and artists who don’t necessarily need or want to learn how to code. They can then create their own systems and intuitively explore their art in their own way.

For example, using a laptop camera, the musician can map pixel colour to the parameters of a software drum machine. The machine learning algorithm is given a certain number of categories, and each category is defined by the musician by recording their position with the camera. The musician then defines the output drum sample they want to associate with their position on camera. Once trained, the algorithm will then associate the performer’s position to a specific drum sample. To see this example in action, check out Rebecca Fiebrink’s Wekinator video example in the ‘See also’ section.

Tools like the Wekinator aim to provide a bridge into the world of creative AI for artists and musicians whose main concern may not be learning how to code. This democratisation of creative AI could have a major impact on the way non-coding artists approach their work.

You can hear more about this in the next step, where we speak to Rebecca Fiebrink about her work and her approach.

Have your say

Based on what you have learned so far about the development of AI systems, can you think of any other barriers that might prevent artists exploring AI systems?

Share your thoughts with other learners in the Comments section.