Monday, 8 March 2021

Book review: Hello World

Algorithms, artificial intelligence, and machine learning are more than techno-babble, or buzzwords to use to make you seem like you are in the know. They are key terms that are necessary for understanding how our data are being used (or abused) to recommend things to us, sell us things, advertise to us, and make decisions about us. It could be argued that there is a minimum level of understanding of these terms that is necessary in order to effectively negotiate our way in the modern world. Or at least, to negotiate our way while perhaps avoiding some of the worst potential outcomes.

Hannah Fry's book Hello World is a good source of material to help you achieve that minimum level of understanding. The subtitle is "Being human in the age of algorithms". The book covers key developments in algorithms in a variety of areas including justice, medicine, crime and policing, autonomous vehicles, and even the arts. Many readers will be surprised at the sheer breadth and depth of applications to which algorithms have been applied. Even though there are whole fields in science and business that Fry doesn't explore (see for example Prediction Machines, which I reviewed here), for many readers those omissions will not be noticed.

There was little in the book that was new to me, but Fry has an easy writing style that still made it a pleasure to read. She also does an excellent job of distilling the key points in explaining otherwise dry models such as random forests, neural networks, or genetic algorithms. While those are terms that you are unlikely to encounter in everyday living, it is helpful to understand at least at a basic level what it is that those types of algorithms are doing.

Although, what exactly it is that algorithms are doing remains and always has been the problem with machine learning algorithms. From outside the 'black box', it is difficult if not impossible to determine what associations the algorithm is using or how it is using those associations. I think Fry skims lightly over this key issue, although she does pay lip service to at least one potential solution:

Thankfully, the calls are getting louder for an algorithmic regulating body to control the industry. Just as the US Food and Drug Administration does for pharmaceuticals, it would test accuracy, consistency and bias behind closed doors and have the authority to approve or deny the use of a product on real people.

The FDA hasn't exactly covered itself in glory during the coronavirus pandemic (e.g. see here), but given the potential negative consequences of algorithmic bias, it seems increasingly clear that some oversight is warranted. Fry does highlight negative consequences of algorithms throughout the book, but stops well short of Cathy O'Neil's Weapons of Math Destruction (which I reviewed here). O'Neil expresses much stronger negative views on the use of algorithms, and O'Neil's calls for reform are all the more forceful for that.

Disappointingly for me, Hello World didn't get really interesting until the concluding chapter, which was consequentially too brief. For instance, I really liked this quote, which was buried in a footnote on page 199:

There's a trick you can use to spot the junk algorithms. I like to call it the Magic Test. Whenever you see a story about an algorithm, see if you can swap out any of the buzzwords, like 'machine learning', 'artificial intelligence' and 'neural network', and swap in the word 'magic'. Does everything still make grammatical sense? Is any of the meaning lost? If not, I'd be worried that it's all nonsense. Because I'm afraid - long into the foreseeable future - we're not going to 'solve world hunger with magic' or 'use magic to write the perfect screenplay' any more than we are with AI.

Brilliant, and an antidote to the hype that constantly gets heaped on algorithms. The book could have done with more of that dose of realism in amongst the anecdotes of how algorithms are being used, for good or ill. In particular, the difference between correlation (which algorithms identify) and causality (which they cannot) is important and often ignored. All of which should lead us to conclude that algorithms are not perfect, as Fry notes (again in the conclusion):

Algorithms will make mistakes. Algorithms will be unfair. That should in no way distract us from the fight to make them more accurate and less biased wherever we can - but perhaps acknowledging that algorithms aren't perfect, any more than humans are, might just have the effect of diminishing any assumptions of their authority.

Fry concludes that perhaps the best algorithms are those than enhance, or support, human decision-making, rather than those that replace human decisions. That seems fair, at least until the algorithms can do real magic.

No comments:

Post a Comment