( OS prefers reducing animation! )

Research Blog


Experiments with Practice Based Digital Research

In this post, Creative Informatics Director of Research Prof. Melissa Terras reflects on her involvement in the development of an artificially intelligent improvised Fringe show generator, which has been getting rave reviews locally and around the world.

Research undertaken by Creative Informatics is evolving to take various guises. From quantitively mapping the cultural and creative sectors in Edinburgh, to interviewing Fringe performers who have had to switch to digital performance due to COVID-19, to creating installation art that confronts linguistic capitalism, we have a range of projects and initiatives that all seek to investigate how data can be used to drive ground-breaking new products, businesses and experiences, and also to critically evaluate our own approach to doing so.

In my role as Director of Research for Creative Informatics there are numerous management tasks to undertake, and work packages to oversee – but occasionally it’s good to get your own hands digitally dirty, undertaking practice-based research (“an original investigation undertaken in order to gain new knowledge partly by means of practice and the outcomes of that practice”). This keeps data skills fresh, but also to push your understanding of the digital – and to do the most essential thing that fosters good research practices: play.

This year, that playful project for me has been ImprovBot: Edinburgh Festival Generated. The premise of this is that “for the first time since 1947, the Edinburgh Fringe Festival as we know it is not going ahead. But there are years of previous programmes stored in the digital vaults. The Bot, an artificial neural network, goes to work on these data, compiling the world’s first AI-generated event blurbs for a virtual arts festival of comedy, plays, musicals, and cabaret… will you lend a hand in the effort?” Exploring the junction of human creativity and comedy, ImprovBot was operational for the planned duration of the Fringe 2020, tweeting out a new show description derived from original Fringe show listings, between 2011-2019.

the whole thing stands as a temporal elegy to a Fringe that couldn’t happen in real life

It turns out that these 2 million words of descriptions from 28,000 previous shows are an ideal corpus to train an AI to mimic the language of the brochures. At 7pm every evening, we posted a live improv, attempting to respond to these shows, by the Edinburgh University Theatre Company’s improv troupe, The Improverts. The overall results of 350 new show descriptions are funny, jarring, unexpected, and poignant. The “new shows” produced by the Improverts are hilarious – but the whole thing stands as a temporal elegy to a Fringe that couldn’t happen in real life.

The Improverts, ‘Sheepleton’: Melissa’s favourite ImprovBot performance

The response online has been phenomenal – with a **** review from The Stage, the leading British weekly newspaper covering the UK’s entertainment industry, an article in the Guardian, and other news pieces from Australia to Iran to Portugal to Indonesia, and beyond.

In creating, managing, and producing the project, I’ve learnt an incredible amount – working with and learning from a small but playful team. Full credit goes to Gavin Inglis who programmed up the Recurrent Neural Network used to create The Bot’s descriptions, and Rudolf Ammann, who lovingly produced 350 different illustrations riffing on concepts of AI, to illustrate each of the new Bot created shows. In advance of the launch, its been a refresher course in generative AI, data management and analysis, and graphic and UX design. Since the launch, I’ve brushed up my social media, promotion, and chatting-to-print-media skills. I’ve remembered how roller-coaster these “live” projects can be (a month long digital project – much like a Fringe show – is exhausting). I’ve been thinking of how to capture, curate, and discuss the project, paying attention to the interactions, and thinking of methods, approach, and the questions that this type of project raises.

short-term digital installations or events can be a great way to rejuvenate the spark for research

The technological intervention here was not necessarily new – there are other projects that exist in the AI and performance space (such as Echoborg, a show created afresh each time by the audience in conversation with an artificial intelligence, and Improbotics, an “Improvised Theatre Experiment”). There are even more advanced AI methods we could have used to create ImprovBot’s content. However, the idea of using tech to imagine a festival that didn’t happen, and to bring some levity at a time when the usual seasonal events are cancelled, everywhere, has seemed to strike a chord. There are issues here about the nature of comedy, the individuality and skill of human improvisation and performance, the methods which are used to increasingly imitate (but only imitate, never creating) human communication patterns, and this unhappy relationship between Bot and Human where the success of one may see the collapse of the livelihood of another. There are questions about where best we can deploy these type of technologies in the creative industries, and how they can be resourced (most AI is funded, these days, for surveillance capitalism, not just for fun). These are the areas in which Creative Informatics sits, and should have opinions on. As ImprovBot comes to the end, this all needs carefully considered, and written up. These short-term digital installations or events can be a great way to rejuvenate the spark for research, and to appreciate, and critically evaluate, the creative technologies built and maintained by others.