Can AI support creativity? The technology is often framed as threatening creative work either by automating it or by encouraging imitation. But Professor Rebecca Fiebrink’s work in creative machine learning suggests a more useful way to think about this relationship. In our March research seminar, she showed how machine learning can help people work with meaningful data, communicate ideas through examples, and build new kinds of creative projects.

Our current seminar series focuses on teaching applied AI and how educators of subjects beyond computing can make AI and machine learning relevant in their classroom. We were delighted to have Rebecca join us to share insights about the place of machine learning in artistic creation. In her talk, Rebecca explored three connected questions:

• How machine learning can be valuable to musicians, artists, and other creators
• What machine learning tools for creators should look like
• What creators need to know about machine learning in order to use it effectively

Using movement, sound, and image data to teach about machine learning

One of the seminar’s key ideas was that machine learning can help creators work with forms of data that already matter to them. Rebecca showed that useful data can come from many sources, including microphones, webcams, phones, wearables, sensors, and body movement. She argued that collecting data is often relatively easy, while interpreting and using it is much harder. 

This suggests a different starting point for AI education. Instead of beginning with a large dataset prepared by somebody else, learners can start with data that is meaningful in their own context. For instance, data about hand gestures can be linked to different musical rhythms, colours, or game actions.

What counts as input?

The seminar also points to a broader shift in how we think about input if we consider creative work. Traditional computing often treats input as something abstract and controlled: a click, a typed command, or a button press. But many creative practices do not work like that. They depend on timing, gesture, rhythm, touch, sound, and movement.

Instead of asking learners to translate everything into words or code first, Fiebrink suggested that educators can use machine learning to allow learners to begin with movement, demonstration, or sound. This is especially relevant in art forms shaped by flow and physical expression, such as music, dance, performance, and interactive media.

Educators can use machine learning to allow learners to begin with movement, demonstration, or sound [instead of with code].

That creates interesting possibilities for teaching. AI does not have to be explored only through screens, prompts, and abstract models. It can also be approached through embodied activities, where learners use gestures, performance, and experimentation to see how an AI system responds. This can make machine learning feel more connected to forms of making that young people already understand.

Teaching machine learning through examples

A second important theme in the seminar was that machine learning allows people to instruct computers through data and examples. Rebecca suggested that this can be especially valuable in creative and embodied work, where what a person wants to express may be difficult to describe in words, maths, or code alone.

One of the strongest examples in the seminar was ‘Wekinator‘, a tool Rebecca has been developing since 2008. She described the tool’s approach as ‘interactive machine learning’: users demonstrate training examples, train a model, test it in real time, then modify their examples and repeat the process.

This is a useful example for the classroom because it shows that training a machine learning model is not a single event, after which the model is trained and finished. Instead it is an iterative process. With Wekinator, learners can try something out, observe the result, and improve the system by changing the examples they provide. That makes ideas such as testing, evaluation, and bias much easier to discuss.

Supporting creativity and learner agency

Rebecca also argued that machine learning can help more people become creators. She contrasted large, one-size-fits-all systems that encourage users to imitate existing styles with smaller, more personal systems that can be trained on new data for specific purposes. She captured this contrast clearly, from prompts such as ‘Write music like Bach!’ to examples of personalised tools and interfaces.

This is an important distinction in teaching and learning. If learners only use AI tools to reproduce familiar outputs, then creative work can become narrow and formulaic. But if they can build or train systems around their own interests, intentions, and materials, then machine learning can support experimentation and authorship.

If [learners] can build or train systems around their own interests, intentions, and materials, then machine learning can support experimentation and authorship.

Teaching AI without turning it into a black box

In the final part of the seminar, Rebecca moved from examples to teaching principles. One of the clearest was that machine learning should be taught at a high level with minimal maths, but not as a black box.

Learners do not need advanced mathematics to start exploring machine learning meaningfully, but they do need to understand that:

• Machine learning models are built from data
• Models make predictions based on patterns
• People can inspect, test, and improve models

Rebecca also argued that small data and interactive machine learning can be highly effective. She highlighted quick experimentation, creative usefulness, and the opportunity to build intuition about ideas such as outliers, features, regularisation, and bias in data. Small-scale activities can make technical ideas more visible and manageable for learners.

Why this matters for teaching

Rebecca ended on an inspiring note: she argued that learning and teaching creative machine learning is both worth doing and possible. She pointed to a growing set of tools that support experimentation and original creative work without much maths or coding, including Wekinator, Teachable Machine, Micro:bit CreateAI, and more.

The seminar also addressed some important limitations. Rebecca warned that commercial tools are not always good at supporting learning or genuine creative work. She also discussed the difficulty of making generative AI tools safe for children, noting the need for built-in filters, moderation, prompt design, and extensive testing. Therefore, what’s important is to think about what learners are actually learning, and to make space for experimentation without losing sight of safety and critical thinking.

Join our next seminar

Our research seminars brings together educators and researchers to explore key questions in computing education.

Next in our series on applied AI, Prof. Gianfranco Polizzi (University of Birmingham, UK) will talk about media literacy in the age of AI. Sign up now to join the seminar on 16 June, 17:00 BST: