Creative Machine Learning Using Figment

Using machine learning for non-computer science practitioners can feel daunting. In recent years, the situation has improved through libraries like ml5.js, which make machine learning available to anybody with basic coding skills. But linking all these different components together is still a lot of work. Figment's aim is to be the easiest tool to get started using AI for creative purposes. Through visual coding you can easily integrate pretrained models into a live setup, or prepare images for training.

This two-day workshop introduces you to a new experimental tool used for real-time machine learning and data preparation. Figment was used during the LAbO summer in 2020 to train a virtual AI-driven dancer. It was also used during a number of workshops where in a couple of hours students, some with no prior coding skills, were able to create projects that employed machine learning to do pose detection, collage generation and more.

Figment is part of "The Algorithmic Gaze" — a research project initiated by Dr. Frederik De Bleser and Lieven Menschaert in which they explore the creative applications of AI by gaining insight into the artificial thinking process of the computer. How can we, by "deciphering" the algorithms, get a view of "The Algorithmic Gaze" — the image that the computer has of the world, and how does this view influence our own view of reality? How do our own biases permeate those of the machine? Can a computer algorithm guide a creative process?

DAY 1 — Thursday, February 24th

10 a.m. - 12 p.m. | Introduction to generative artificial intelligence. Show cutting edge work, and talk about the possibilities and limitations of AI generation techniques.

2 - 5 p.m. | Introduction to Figment; learning the basics of the software, applying real-time machine learning models, controlling parameters interactively using the web camera or external devices using OSC.

DAY 2 — Friday, February 25th

10 a.m. - 12 p.m. | Using Figment for batch training. Using collected data, we can prepare the raw images into a unified setup.

2 - 4 p.m. | Train the algorithm

4 - 5 p.m. | Results and wrap-up