This is the second book that I’ve recently read while trying to increase my knowledge of how AI works. The difficulty is that AI is so rapidly evolving that any book that you read is probably out of date by the time that it is published.
But you need to start somewhere.
I highly recommend this book as an introduction to the subject. The writing is mostly straightforward, and not buried in technical jargon. Professor Mitchell gives plenty of real life examples to help us understand the concepts that she is describing. You can read more about her bio at the link.
I’m not going to give a detailed review of the book. Instead, I’ll share that I am most interested in the potential ethical issues surrounding use and development of AI systems. At the end of the book Professor Mitchell provides some references for further reading of this subject. I plan on reading these later.
In your reading you might be more interested in other aspects of AI. She gives a broad enough overview of the subject that you’ll probably discover your specific area(s) of interest.
What are examples of the potential ethical issues surrounding usage and development of AI systems?
- Lots of data are required to develop and tune these systems. Sometimes the data used in your date, and sometimes it’s used without your permission or knowledge. As an example, you could be helping an AI developer every time you post a photo on Facebook or instagram, or tag someone or something in a photo, or even when you answer a captcha for a website–such as “click on all of the pictures that contain bicycles.”
- The data used to develop an AI system could contain quantitative or societal biases, and thus the AI model could propagate these same biases forward in time. A related concern is “who decides when something is biased”, and “how will they rectify it?”
- AI system learning can be modified for nefarious purposes. An example given in the book is when a medical diagnostic program is modified to alter its output in order to extract underserved payments from insurance companies. It’s easy to imagine other misuses.
- The impacts on the economy from widespread use of AI are unknown and difficult to predict. One researcher says that AI can be as important as the commercialization of electricity. The difference is that electricity was well understood before it was commercialized. AI technology at this state is not well understood.
I realize that AI development is fast moving, and this 2019 book is likely out of date in some respects, but I still found it interesting and worth my time.