Unfortunately, although it covers a fair number of AI applications, it's also written more like a business book that a popular science/technology book, and as such it's pretty dull. Anyone familiar with the business book genre will recognise that deadly moment when you get to a box that's a case study. It's going to be boring. This book contains 29 case studies, one after the other, by the end of which I was quietly groaning.
However, there were definitely some insights to be gained here. In that range of case studies, there were several standouts. The main thesis that Davenport and Miller are proposing is that, despite some issues, artificial intelligence will not destroy vast swathes of jobs, but will instead improve them by taking on the boring bits, not (on the whole) displacing humans, but working alongside them. Perhaps the best example of this was the robotic weed picker, which made a farm worker's job more interesting and did something that, frankly, no human really wants to do. Admittedly, in this kind of application there would be fewer humans employed, but it feels like a genuinely beneficial change.
What was worrying, though, in the dark side orientation of this book was that there was very little consideration of some of the other potential negatives of AI - in fact, it felt the authors were almost celebrating some of these. Several case studies highlighted this approach, for example one on using AI to support a help desk, a couple on making decisions on issuing insurance policies and mortgages and one on policing.
The help desk example felt particularly insidious. The idea was that the software monitored conversations between customers and the help desk to improve the quality of interactions. But apart from a passing mention of it, the authors don't really acknowledge the Big Brother aspect of software checking your every word, rating your performance and pushing you into conformity with the required groupthink. Similarly, we heard about all the advantages for the companies using software to decide if customers should be given an insurance policy or mortgage, but not the well-documented problems raised by opaque machine learning systems using entirely unsuitable data to reject individuals. The policing example is an infamous one, and the authors had to acknowledge there have been serious problems with such systems producing racist results and making particular areas even worse than they were before, but merely say this has to be avoided, without giving any evidence that this is even possible to do.
I'm sure Davenport and Miller thought they were doing something useful in focusing on the ways that AI will not necessarily replace human workers but rather would augment their abilities. But I don't think it's possible, as was done here, to ignore some of the other dangers of AI like lack of transparency, misuse of data, surveillance and more. You have to take the view across the board.
I'd suggest this book is important reading to get a balanced picture of AI, if you can cope with the kind of mangled business-speak sentences that crop up, such as 'She works particularly at the top of the prospect funnel, trying to move leads along in the sales process and operationalize a disciplined prospecting and selling process.' The book does illustrate a few examples where having an AI helper can be genuinely beneficial to workers. And plenty more where it can benefit companies to the disadvantage of either workers or customers. This is surely valuable data, whether you side with Luke Skywalker or Darth Vader.
Hardback:   |
Review by Brian Clegg - See all Brian's online articles or subscribe to a weekly digest for free here
Comments
Post a Comment