The AI Paradox - Virginia Dignum ****

This is a really important book in the way that Virginia Dignum highlights various ways we can misunderstand AI and its abilities using a series of paradoxes. However, I need to say up front that I'm giving it four stars for the ideas: unfortunately the writing is not great. It reads more like a government report than anything vaguely readable - it really should have co-authored with a professional writer to make it accessible. Even so, I'm recommending it: like some government reports it's significant enough to make it necessary to wade through the bureaucrat speak.

Why paradoxes? Dignum identifies two ways we can think about paradoxes (oddly I wrote about paradoxes recently, but with three definitions): a logical paradox such as 'this statement is false', or a paradoxical truth such as 'less is more' - the second of which seems a better to fit to the use here. 

We are then presented with eight paradoxes, each of which gives some insights into aspects of the overriding first, which Dignum calls 'the AI paradox' - 'The more AI can do, the more it highlights the irreplaceable nature of human intelligence.' Because of potential misunderstandings of what AI is and what it can do, many of the paradoxes arise from the assumption that it can replace human intelligence, or indeed is intelligent at all in the same sense of humans. As Dignum points out, there is no doubt AI is better at some things than humans, just as a calculator is better at doing arithmetic. But because it is purely data driven and lacks any understanding or empathy (even it tries to fake it), it will never be a substitute for our abilities. Used properly it's a great asset, but we need to understand its limitations to use it well.

I won't go through all the other paradoxes here, but to give a flavour, we get the Agreement paradox 'The more we explore AI, the harder it becomes to agree on a definition', the beautifully paradoxical intelligence paradox 'AI is what AI cannot do' and the Regulation paradox (which I can't really see as a paradox at all) 'Responsible innovation need regulation'. For each paradox we get a chapter explaining why the paradox exists, what it means and in some cases (like regulation) why it's difficult to do anything about it.

A couple of small moans - Dignum refers to Ada King (Countess of Lovelace) as 'the world's first programmer' where in reality she was second as Babbage wrote several algorithms for the Analytical Engine (they weren't really programs in the modern sense) before her contribution. And the selection of examples of AI LLMs etc. was strangely limited - Anthropic got one mention of Claude (though not of the company), and X/Grok was not mentioned at all, which is odd given how often the latter gets picked up in the media when it hits problems.

The main weakness, apart from the writing style was that the book is far stronger on the issues than on solutions, which tend to be strongly oriented to relying on international bodies that seem incapable of much action. The two come together in the Solution paradox chapter where we get text like 'To move beyond AI-solutionism, a critical, multidisciplinary perspective is necessary... Aligning technological advances with democratic and human rights principles is crucial for ensuring just and equitable outcomes... A balanced approach, combining top-down and bottom-up strategies, is essential. Ethical frameworks must be adaptable to different cultural contexts, supporting a variety of interpretations and values.' A practical roadmap it isn't.

However, if you come to this book without great expectations for readability, it does provide genuine insights into the benefits and limitations of AI and how we need a better understanding of what it can do and how to use it safely and effectively, even if it can't sensibly get us from here to that desired position. As such it's valuable contribution to the AI debate.

Hardback:   

Kindle:  

Using these links earns us commission at no cost to you

These articles will always be free - but if you'd like to support my online work, consider buying a virtual coffee or taking out a membership:

Review by Brian Clegg - See all Brian's online articles or subscribe to a weekly email free here