What are the limits of AI?

ORIGINAL INTRODUCTION

The computing pioneer Alan Turing predicted that, by the twenty-first century, ‘one will be able to speak of machines thinking without expecting to be contradicted’. If anything, his prediction now seems rather conservative. One tech entrepreneur – who has been appointed by the UK government to chair the Frontier AI Taskforce, a body established to ‘develop the safe and reliable use’ of AI – has described the AI of the future as potentially not just human-like but God-like (with a capital ‘G’) and ‘capable of infinite self-improvement’.

This prospect is presented as either inspiring or terrifying – often both at once. As well as ploughing £100million of public funding into its new Taskforce, the government has also announced a Global Summit on AI Safety, and will convene with tech giants at Bletchley Park in November this year ‘to ensure this technology is developed and adopted safely and responsibly’. The choice of Bletchley Park is meant to evoke the urgency of the Second World War, while also reminding us of AI’s origins in Alan Turing’s work, which established the basis for all modern computing.

Meanwhile, most of us struggle to make sense of successive headlines which tell us that ‘generative’ AI – including text generators and chatbots like ChatGPT and Bard, and image generators such as Stable Diffusion, Midjourney and DALL-E – has either made some astonishing new breakthrough, or failed to live up to the initial hype. When major technical mishaps continue to disrupt our daily lives – from the UK’s air-traffic control being brought to a standstill by a single piece of wrongly inputted data, to a security breach at the Electoral Commission exposing the personal data of 40million UK voters – how seriously should we take the proposition that today’s tech, or tomorrow’s, might have the power of God?

Is ‘infinite self-improvement’ a genuine possibility with AI, or might a more thorough assessment reveal some fundamental limits? If we delve into the rich history of computing, going all the way back to the nineteenth century, could we find the key to a more rational understanding of today’s fast-evolving technology?

SPEAKERS
Dr Stuart Derbyshire
associate professor in psychology, National University of Singapore and the Clinical Imaging Research Centre

Professor Anders C Hansen
professor of mathematics, University of Cambridge; author, Compressive Imaging: structure, sampling, learning

Timandra Harkness
journalist, writer and broadcaster; presenter, Radio 4’s FutureProofing and How to Disagree; author, Big Data: does size matter?

Andrew Orlowski
writer and critic; business columnist, Daily Telegraph

Dr Kathleen Stock
columnist, UnHerd; co-director, The Lesbian Project; author, Material Girls: why reality matters for feminism

CHAIR
Sandy Starr
deputy director, Progress Educational Trust; author, AI: Separating Man from Machine