Review of Illingworth, S. (2026).
The emergence of generative AI in the shape of ChatGPT, Claude, Gemini, and more, has resulted in a plethora of books dealing either with how to use these systems effectively, or devoted to warnings of the doomed future of humankind. Happily, Sam Illingworth’s self-published little book fits neither of these categories, rather it deals with the question of whether to use these tools at all, and when it may be better to put them aside. Illingworth comes from a background of atmospheric physics and is now Professor of Critical AI Literacy and Napier University, Edinburgh. As noted, his book is self-published through Amazon, and was available initially at the bargain price of £0.99, which is when I bought it. It is now still a bargain at £2.99 for the Kindle version, or £7.99 as a paperback. Illingworth runs a Website on Substack at
Illingworth is not opposed to AI; in fact he acknowledges its usefulness and uses it himself. However, his concern is not with such things as how to write better prompts, but how to judge when it is useful. To aid decision-making in these terms (in Chapter 2) he suggests a four-part test for any given task: whether it is bounded, i.e., not open-ended; whether the output from the system can be checked for accuracy, etc., by a human evaluator; whether the comparison with not using AI is honest; and whether the thinking behind the performance of the task is the user’s own. The failure to observe such rules has already had consequences for companies. For example, the financial firm Klarna in Sweden sacked many of its customer-service people when adopting AI systems as alternatives. They then found that the AI system could not handle complex issues demanding human judgement and had to rehire many of those previously sacked.
This is a short book of only seven chapters (each chapter followed by a relevant poem by the author): after the first two introductory chapters the remainder deal with aspects of AI that affect whether and how we use it. Chapter 3 deals with the genuine time-saving that AI can achieve, but there is then the hidden cost that the advocates of AI often fail to mention, that is, the time needed to check the output. At times, this can exceed the time saved by using the AI.
Generative AI systems such as Claude and the rest are designed to keep the user engaged with the system, to the extent that some users become depended on the system, regardless of the costs they are incurring. In Chapter 1 the author noted that, ‘Your AI tool is not your friend. It simulates empathy to keep you prompting.’ This is dealt with further in Chapter 4, which also deals with ‘AI slop’, that is, ‘content with no agency, no depth, and no purpose’, which is automatically generated by AI systems at the behest of commercial organizations. The author notes that both OpenAI and Anthropic generated slop products in the form of a continuous video feed. However, it was quickly discovered that these products were losing more money than they generated simply because there was no human interaction and the human viewers quickly lost interest.
Chapter 5 may be the chapter of greatest interest to readers of this journal, since it deals with the problem of detecting the difference between generative AI output and human writing. This problem has led to a number of problems in universities, with students accused of plagiarism when, in fact, the detection systems were incapable of accurately identifying AI writing and human writing. Basically, the detection systems currently available do not work: in a study of fourteen systems, ‘The best detectors on the market were wrong at least one time in five, and some of those errors accuse innocent students.’ In another study of essays written by non-native English speakers, which were written under circumstances where AI could not have been used, 61% was flagged as AI written. Some universities have abandoned these systems and are relying on working with students to ensure honest, effective use of generative AI, sometimes associated with oral examinations instead of essays.
We are beginning to realise the ecological cost of these AI systems, with citizens in some places campaigning against the building of data centres which consume vast amounts of electricity and water (for cooling and electricity generating purposes). Chapter 6 deals with this and the author notes: ‘training GPT-3 consumed about 5.4 million litres [of water] in total, roughly 700,000 of them evaporated on site cooling servers and the rest used off site generating the electricity.’ The fairly obvious solution to this problem is for governments to require those building such centres to create their own generating stations, rather than using the national grid and potentially depriving the ordinary citizen of effective electricity supply, and to discover their own sources of water supply. Such a solution, however, is unlikely to be adopted since the rhetoric of both politicians and business leaders refers to the potential for employment – although once the centres are built they actually employ very few people.
For a scientist like the author to reach the conclusion he does in Chapter 7 is, perhaps, surprising. He argues that the skills required for the effective use of generative AI systems, ‘are the interpretive skills, and for two and a half thousand years the place they have been taught is the humanities. Which would be a comfortable conclusion for a professor and a poet to reach, except for one detail. We have spent the last twenty years defunding them.’ And that defunding of the arts and humanities continues as governments are sold the idea that higher education should deliver a trained workforce, rather than developing imaginative, problem-seeking individuals who can change technology and society.
This is a short book but, as the author notes, as it is self-published it can be revised and extended as the technology develops and, until a second edition, you can follow his thoughts on the website.