Tutorials

Why So Serious? Building Creative Systems with a Sense of Humour

For much of its history, AI was a scientific discipline defined more by its portrayal in science fiction than by its actual technical achievements. Real AI systems, built upon very large language models (LLMs), are now catching up to their fictional counterparts and are as likely to be seen in news headlines as on the big screen. Yet as AI outperforms people on tasks that were once considered yardsticks of human intelligence, one area of human experience still holds out, for now at least: our very human sense of humour.

This is not for want of trying, as this tutorial will show. The true nature of humour has intrigued scholars for millennia, but AI researchers can go one step further than philosophers, linguists and psychologists ever could. By building computer systems with a sense of humour, capable of appreciating the jokes of human users or even of generating jokes of their own, we can turn academic theories into practical realities that amuse, explain, provoke, and delight.

This tutorial will use the concepts and achievements of AI to explore what it means to have a sense of humour, and moreover, to understand what it is to not have a sense of humour. It will challenge the archetype of the rigidly humorless machine in popular culture, to celebrate what the archetype gets right and to learn from what it frequently gets wrong. In doing so, it makes a case for the necessity of a truly computational understanding of humour.

Computers have so long been a foil for humorists that the notion of a humorous computer has remained the stuff of speculative fiction. This tutorial provides a grounded introduction to the research field dedicated to this marriage of opposites, computational humour. By exploring what it will take to give our machines a sense of humour, the tutorial will also explore the hidden mental and social computations that make us truly human.

Topics to be covered by the tutorial include: theories of humour; early symbolic models; joke analysis (e.g., of Reddit joke sets); recent developments in the field, as informed by (LLMs); practical uses of computational humour in AI; moral dimensions and dilemmas; irony, sarcasm and sentiment; and the capacity (and otherwise) of LLMs to be truly humorous.

Organizers

• Tony Veale (University College Dublin)