The Creative AI meetup is designed to bring together artists, developers, designers, technologists and industry professionals to discuss applications of artificial intelligence in the creative industries.
January’s talk discussed existential risk and computational creativity.
Shahar Avin of Centre for the Study of Existential Risk presented his work on a superintelligence modification (mod) for Sid Meier’s Civilization® V, a popular turn-based strategy game. The aim of the mod is to concretise some of the issues surrounding catastrophic AI risks, and to put individuals in a situation that makes both the risks and possible solutions accessible.
The mod allows the player to pursue victory through technological superiority, via developing a safe superintelligence, while introducing associated risks from rogue superintelligence, which could lead to human extinction (and game loss). Players can allocate resources to AI research and to AI safety research, negotiate AI treaties with other civilisations, all while balancing the demands of all the other interlocking systems of the game, including trade, diplomacy and warfare. The mod was made available to a closed group of testers and the responses were mixed, highlighting some of the difficulties of concretising abstract concepts in this area, while also suggesting certain key characteristics of the domain are amenable to treatment through the medium of a video game.
An interesting discussion at the end included one audience member mentioning versions of Civilization® that have been played solely by AI, all of which have ended in destruction and the end of the game. The lesson seemingly being that the modelling suggests that time spent on developing safe superintelligences could avert catastrophic risks and ultimately human extinction.
Simon Colton explored ways in which generative systems can evolve into genuinely creative, autonomous systems, drawing on 20 years of Computational Creativity research.
Simon’s talk was entitled “From Creative AI to Computational Creativity and Back Again”.
One of the maxims emerging from the Creative AI movement, fuelled by developments in generative deep learning models, is the notion of producing the highest quality output possible from creative systems. If we accept that how and why an artwork was produced is often taken into consideration when value judgements are made, then the academic field of Computational Creativity has much to offer to help Creative AI practitioners. Simon explored ways in which generative systems can evolve into genuinely creative, autonomous systems, drawing on 20 years of Computational Creativity research. Conversely, the remarkable power of generative networks to hallucinate images, music and text represents a real boon for Computational Creativity researchers interested in the simulation of imaginative behaviour, and also upon the ways in which we are currently (and could in future) harness this power to explore practical and philosophical aspects surrounding the idea that software can be creative.
The most interesting aspects of the talk were his assertion that there is no agreed definition of creativity, and the idea that the context of a piece of creative work counts for much. If we know an artwork is made solely by AI, we judge it very differently, and usually negatively.
Simon Colton is a Professor of Digital Games Technologies at Falmouth University and a part-time Professor of Computational Creativity at Goldsmiths College. He is an AI researcher specialising in questions of Computational Creativity — getting software to autonomously create artefacts of real value in interesting ways. He has published nearly 200 papers and his research has won national and international prizes. He is most well known for the software he has written and co-written to make mathematical discoveries; paint pictures; make games and generate fictional ideas, including The Painting Fool. He’s also known for his philosophical and theoretical contributions to Computational Creativity, in particular driving forward the assessment of creative software via what it does, rather than what it produces.