Speaker
Description
Recent developments in the scaling and training of large language models (LLMs) have led to a dramatic change in how the public views Artificial Intelligence. No longer the vaguely aspirational preserve of science fiction stories, AI is now expected to work, and not just in the laboratory but in a wide range of consumer products. Yet as AI outperforms people on tasks that were once considered yardsticks of human intelligence, one area of yhuman experience still holds out, for now at least: our very human sense of humour. This is not for want of trying, as this talk will show. There is good reason for computer science to take humour seriously. 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. The writer Clive James once pronounced that one should not trust anyone lacking a sense of humour, even, indeed, to post a letter, for what is humour but our sense of equanimity and poise in the face of the unpredictable when common sense has been pushed to the brink? My talk will describe where researchers are on this road to more humorous machines, and explore how we might go further towards giving LLMs a robustly human funny bone. The talk will also cover related issues such as acceptability and value alignment in LLMs, since humour often pushes the bounds of what is socially acceptable in polite company.
