4 stars

I really enjoyed this, especially as someone who works in data science! While not written by someone who works in the field, the writer addresses important issues & relevant pitfalls that are occurring with the sudden increase of commercial & social reliance on “black box” AI & big data.

[What I liked:]

•Although mostly anecdotal examples are used to illustrate the thesis of the book (that quantitative data divorced from human expertise & experience is not as superior a metric as some people think), the writer approaches the topic from a variety of angles so you get a good overview of current controversial issues in the field.

•The writing style is engaging, & I found the book to be solid research- & information-wise, while still written at a level that a general audience can understand without much technical background.

•I think this book covers very relevant & important topics that society at large would do well to be more educated on. They include how:

1) AI can produce efficient & effective results, but since algorithms are a “black box” (developers cannot examine how the algorithm arrived at certain predictions), & it is problematic to rely solely on such data outputs without human oversight because it easily leads to detrimental (& illegal!) discrimination.

2) The usefulness & reliability of quantitative analysis heavily depends on the accuracy of the data inputs. And to draw applicable conclusions from the outputs, you need people knowledgeable in the data field to set up the analysis parameters, properly apply methodology, & be able to recognize weak points & reasonable applications of the results (“garbage in, garbage out”).

3) Very high quality human expertise (qualitative data & analysis) is often just as good, or better, at predicting outcomes than solely relying on quantitative data.

4) It is often better to use a mix of human expertise/domain knowledge + quantitative data to evaluate decision making, rather than relying only on one or the other.

(For the record, I pretty much agree on all these points! In particular, points 1) & 2) are being studied & debated in the academic machine learning & data science fields right now.)

[What I didn’t like as much:]

•I pretty well liked this book! Though, as I mentioned before, most of the conclusions are anecdotal examples &/or backed up with few presented studies (probably for reader engagement/keeping the book to a reasonable length). I should also mention that points 3) & 4) above are conclusions I’m less informed about, so I’m taking them with a grain of salt until I see them corroborated in the academic literature. (Although I do find them plausible.)

[I received an ARC ebook copy from NetGalley in exchange for my honest review. Thank you for the book!]