You Look Like a Thing and I Love You is a great, quick read for anyone who is even remotely curious about what things like AI, machine learning or neural networks actually mean and how it affects your day-to-day life.
Whether it’s facial recognition, to self-driving cars, to the search results you get from Google, this book describes the different types of machine learning algorithms, often in a humorous way (the first few pages deal with the author trying to generate pick-up lines using a neural network — hence the title of this book).
It’s a quick read and dives into some of the limitations that machine learning algorithms are bound by and how machine learning algorithms can be tricked, return results outside the parameters of a problem, or even cheat (e.g., when one AI agent was given an objective to minimize the number of times a character died in a video game, it found out how to simply pause the game).
One particular section of the book talks about the recently released GPT-2 data model by OpenAI, which can be used to generate fairly intelligent and coherent sounding blocks of text. It prompted me to try and run the model on my own machine.
After giving the model a prompt of “machine learning can be scary,” the computer responded with the following block of text:
“The problem is, it’s easy to forget that we can’t avoid the problem with software. It’s often hard to avoid the problem, but knowing how to avoid it is not the same as knowing how to avoid it. The problem is, the problem is easy to avoid, but the problem is hard to avoid.”
Do we need to worry about robot overlords any time soon? Probably not.
This is a book that I’d recommend to both people who are tech savvy and to parents who might still call you with questions on how to turn on a computer… or at least anyone curious to how machine learning affects various aspects of our lives.