Book review: 'You Look Like a Thing and I Love You' by Janelle Shane
Image credit: Dreamstime
This book about how AI works and why it is “making the world a weirder place” is charming, side-splittingly funny, and contains just the right amount of informative.
When we hear about AI, it is usually characterised as a faceless, mysterious, all-powerful entity that is going to steal all our jobs/create a utopia/render us disposable and perhaps kill us all. Really, AI is already here and changing our world. This book is a short and sweet guide to what AI is, how it works, what it can do and – importantly – what it cannot do.
Janelle Shane is a scientist and AI researcher known for her blog ‘AI Weirdness’, which mostly features her projects teaching AIs to do pseudo-creative work (such as naming guinea pigs, creating recipes, or writing Harry Potter fanfiction), with surreal and almost Pythonesque results.
‘You Look Like a Thing and I Love You: How Artificial Intelligence Works and Why It's Making the World a Weirder Place’ (Voracious, £20, ISBN 9781472270405), which is named after a chat-up line suggested by an AI, continues in the spirit of her blog. Shane explains complex concepts with humour, memorable examples, and just the right amount of detail such that you put down the book feeling as though you really get it. This book explains from basic principles how the AIs you may have heard of work, such as how Google’s Deep Dream program transforms images into psychedelic works of art.
Much of the book is dedicated to the problems associated with AI’s expanding presence in our lives, particularly due to its replication of human bias. AIs simply makes predictions based on past (often biased) behaviour. An AI used to recommend which prisoners should be paroled copied racist behaviours from training, while an AI being trained to select promising job applications found new and creative ways to be sexist, despite persistent efforts to hide all indications of gender; pretending that AI is the perfect impartial decision-maker is an example of “bias laundering”.
Shane also discusses the difficulty of getting an AI to achieve a goal in a way that is relevant in the real world; AI has a tendency for breaking things in an effort to reach their goal (such as by abusing video game glitches) if not constrained sufficiently. An illustrated discussion about the ridiculous techniques machine learning models will adopt as researchers encourage them to learn to ‘walk’ is one of the highlights of the book. The book also explains that humans can also be susceptible to unusual behaviour in pursuit of a ‘reward’. For instance, the YouTube algorithm was previously found to be encouraging creators to put a disproportionate amount of time into creating attractive thumbnails, which would attract viewers and boost their content. When its reward function was tweaked to reward longer viewing times instead, controversial content (including conspiracy theories) started to be boosted by the algorithm. “As more of our daily lives are governed by algorithms, the quirks of AI are beginning to have consequences far beyond the merely inconvenient,” Shane warns.
With so many books being published about AI over the past few years, it is worth asking: what makes this one distinct? ‘You look like a thing and I love you’ stands out for Shane’s madcap sense of humour and affection for the subject. She anthropomorphises AI without implying it has human-level general intelligence (“It declared Darth Vader was a tree and then proceeded to argue with me about it”), and includes nuggets from her blog which genuinely enhance understanding; teaching an AI a knock-knock joke is a simple and very funny way of showing the readers how an AI discovers the rules a developer would otherwise have to program in themselves.
Shane could have easily repurposed her blog content into a stocking-filler type ‘humour’ book - and it would still be good. However, she has taken the time to write a book which – though it does not introduce any revolutionary new ideas – is an informative and enjoyable read for all non-technical readers, no matter their familiarity with the subject.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.