The author does such a good job making this book accessible to people who may have only a limited working knowledge of how programming works. She also illustrates her points with funny little comics and illustrations. She says that training AI is like training a dog: you have to be specific and make sure you do the right kinds of trainings; otherwise, the AI will take the path of least resistance. One of the examples was a horse betting AI who decided that the most profitable option was to not bet at all.
She also cautions that sometimes the path the AI chooses as the most efficient may not be the most beneficial to humans. An example for this was an AI playing a flight simulator who found out that if they crashed the plane at high velocity, it overloaded the systems that measured force, which then rewarded the AI for a soft landing at 0 (even though if a landing like this was attempted IRL, it would kill everyone on the plane). Similar problems occurred when a self-driving car AI was only trained on what trucks looked like from behind and not the side (it decided the sideways truck was am overhead billboard sign and collided with it).
If you're interested in AI and tech, you'll love this book. It's very cute and very funny. It gets a little repetitive after a while but I really enjoyed reading it nonetheless.
3 to 3.5 out of 5 stars