Until I read Faking It by Toby Walsh, I thought that I worked in a field that was unlikely to be taken over by artificial intelligence. Wrong! Weizenbaum created a system in 1967 called ELIZA, which he intended to be a parody of a psychotherapist – others saw it as having real, clinical application. And fast-forward to 2015, when a chatbot known as Karim was used as a ‘therapeutic assistant’ to help refugees in Germany manage traumatic stress.

Walsh has put together a concise and informative summary of artificial intelligence, or AI as it is more commonly known. The book begins with a bunch of seductive stats and mind-blowing AI scene-setting. I could quote lots of it (because it is amazing stuff) but Walsh’s précis of the ever-growing ChatGPT system gives you a sense (to clarify, GPT is described as a ‘…large language model…designed to produce human-like text…’) –

ChatGPT is built on top of a series of groundbreaking natural language systems with the family name ‘GPT’. It is one of the largest neural networks ever built… GPT-1 had 117 million parameters. GPT-2 was over ten times larger, with 1.5 billion parameters. And GPT-3 was over 100 times larger again, with 175 billion parameters… GPT-4 is believed to be even larger, but OpenAI is refusing to disclose quite how much bigger it is… To give you an idea of just how much text was poured into GPT-3, the complete contents of Wikipedia made up less than 1 per cent of this input.

Walsh explores the history of AI; the meaning of ‘intelligence’; creativity within the context of AI; deception and fakes; and the future (which is essentially how we protect ourselves from deception and fakes).

It begins with the question that Alan Turing answered – ‘Can machines think?’- by establishing that AI is, as the name suggests, artificial and therefore fundamentally different to human intelligence. However, often the goal of AI is to fake human intelligence, and we are now building AI that truly deceives us (such as ChatGPT).

…many have compared the field of AI to medieval alchemy. Rather than attempting to turn base metals into gold, the ambition of artificial intelligence is to turn simple computation into intelligence.

The chapters on creativity (…can computers be creative? And, if they can, would it be real or fake creativity?) and on ethics (can AI systems be ‘moral’?) were the most interesting to me. Walsh provides lots of thought-provoking examples (including the 2020 AI Eurovision contest, which Australia won with Beautiful the World).

So what does all of this mean going forward? Walsh discusses the challenges, noting that as fast as ‘safeguards’ are put in place, they are defeated. He says that first and foremost –

…those of us working in the field have a responsibility to stop anthropomorphising the technologies that we build. Too often we talk about AI as if it were in fact human. We speak about a chatbot ‘understanding’ a sentence, a ‘self-driving’ car, the computer-vision algorithm ‘recognising’ the pedestrian, and the possibility of robot ‘rights’. In reality, chatbots don’t understand language. There is no self – no person, no sentient, self-aware intelligence – driving the car… Algorithms don’t actually recognise objects. And robots need rights about as much as your toaster does.

My sense is that Walsh had fun writing this book. The preface opens with, ‘This book is out of date’, and there are sections that are written by ChatGPT (revealed after you’ve read them). There are also lots of examples of ways to trip AI systems up, and exercises that reveal system limitations. If I’d been sitting at the computer whilst reading, I suspect I would have gone down a few internet rabbit holes as I let AI create swimming pools or croissants in the style of Van Gough and Picasso, or perused Miquela’s Instagram posts.

3/5

As part of the 20 Books of Summer reading challenge, I’m comparing the Belfast summer and Melburnian winter. The results for the day I finished this book (June 18): Belfast 7°-14° and Melbourne 3°-14°.