Ellen Broad’s multi-faceted exploration of the many inter-twined aspects of artificial intelligence embarks and concludes at the same salient juncture; emerging technologies are conceived, shaped, used, governed and iterated by humans. Just as humans are inherently neither good nor bad, the systems we construct echo our moral plurality, our unconscious bias, and, frequently, our unwillingness to be critically interrogated.

That this is Broad’s first book – given its well-researched examples, coherent structure and intellectual incisiveness – is surprising. Its clarion call – for greater care, more rigourous thinking and a more holistic approach to the almost-infantile adoption of artificial intelligence, machine learning and autonomous decision-making – is not.

Structured in three distinct parts – Humans as Data, Humans as Designers and Making Humans Accountable, the book covers much territory. From systemic and cultural biases in how data used by machine learning is selected and captured, to the errors that are introduced to data sets by humans, to decisions made about system tradeoffs, what privacy means in different contexts, how open a system is to inspection and intelligibility, how diverse that system is, to who is accountable for the impacts of a system, real life examples are interwoven with provocative and often confronting questions.

Broad does not set out – at least in this tome – to answer these questions – rather, she lays a foundation for examining each of these questions in more depth. Personally I’d like to see a follow-up to this that covers attempts to standardise practices in machine learning and artificial intelligence – the frameworks and benchmarks – often competing – that have been proposed – alongside efforts at industry adoption and (likely) the barriers that are faced.