A great introduction on the history and challenges of AI aimed at an interested general public. Definitely some technical discussion in some chapters, but not more than an interested layperson should be able to power through. The big problem of the book is the last two chapters (out of five in total), "Artificial Intelligence Meets Natural Language" and "The Barrier of Meaning" which are very much out of date with the advent of ChatGPT and other Large Language Models (LLMs), text-to-image generation and the scientific breakthrough in protein folding coming out of Google Deepmind. From what I understand she is thinking about updating parts of the book and publishing a new edition which would be very welcome.

As it is now the two large chapters on language and meaning is way behind the state of the art right now. An example she gives is a short restaurant story and how it is translated into other languages. The story is as follows:

'A man went into a restaurant and ordered a hamburger, cooked rare. When it arrived, it was burned to a crisp. The waitress stopped by the man’s table. “Is the burger okay?” she asked. “Oh, it’s just great,” the man said, pushing back his chair and storming out of the restaurant without paying. The waitress yelled after him, "Hey, what about the bill?" She shrugged her shoulders , muttering under her breath, "Why is he so bent out of shape?"'

In the later chapters she gets the story translated into several languages and then gets some colleagues fluent in the languages to translate it back to English. The result is, as expected by anyone who used Google Translate the last few years, pretty abysmal. The meaning is completely lost. Today though, when I fed the story to ChatGPT 3.5 and asked it to translate it to Swedish, the translation, while not perfect, was actually really good and the meaning and nuances of the story is kept in the translation. For example we don't have an equivalent expression to "bent out of shape", but instead it is translated to the Swedish word for "upset". "The bill" is correctly translated and not turned into something to do with legislation.

The last paragraph of the book is:

"The impacts of AI will continue to grow for all of us. I hope that this book has helped you, as a thinking human, to get a sense of the current state of this burgeoning discipline, including its many unsolved problems, the potential risks and benefits of its technologies, and the scientific and philosophical questions it raises for understanding our own human intelligence. And if any computers are reading this, tell me what it refers to in the previous sentence and you're welcome to join in the discussion."

I asked ChatGPT about the paragraph and the last sentence.

ChatGPT
The last sentence of the quote is an invitation for computers to join in the discussion if they can identify what the pronoun "it" refers to in the previous sentence. The quote suggests a playful acknowledgment of the potential readership, including both humans and computers. It aligns with a light and inclusive tone, fostering the idea of AI actively engaging with the content and participating in the ongoing discourse about artificial intelligence.

Me: So what does 'it' refer to in the quote?

ChatGPT
In the quote, the pronoun "it" refers to the overarching subject of the sentence, which is the "burgeoning discipline" of artificial intelligence (AI). The sentence mentions that the book has provided insights into the current state of AI, its unsolved problems, potential risks and benefits, and the scientific and philosophical questions it raises about human intelligence. Therefore, "it" encompasses the entire field of artificial intelligence.

The answer ChatGPT gives is pretty good and it is beyond the technology Mitchell thought we would see by now. In the book the time line is one of the reasons she gives for not worrying about AI with values misaligned with human values. Admittedly I'm more worried about AI actually killing all humans or something equivalent than Mitchell is and I have been for a decade now, so those are the lenses I read her book through. When I read her critic about how bad AI is and how far it is from understanding natural language and understanding images I would have agreed with her when the book came out. I would have nodded along when she writes that meaning and understanding context is very hard and that computers weren't close to be able to do it. With that said, I would also have said that progress and breakthroughs can be happen very rapidly.
Today when I believe many, but certainly not all, of the challenges she describes in the book has been overcome she does seem to backtrack in her public posts and appearances. Now when ChatGPT seem to be able to both understand and answer in natural language she seem to claim it isn't very impressive. In the book she discusses a bet between Ray Kurzweil and Mitchell Kapor whether a computer program will pass a carefully constructed Turing test by 2029 and she sides with the skeptical Kapor. Given the progress with LLMs now I would, without having read the details of the bet, side with Kurzweil. A few months ago I listened to a debate she participated in where she seemed to deny there is anything like a Turing test. So even though a lot of progress have been made on the challenges and hurdles Mitchell discusses in the book, the developments seem to have made her even more skeptical than she appears in the book. It is a bit peculiar, but what is clear from the book is how tired she is of all the hype surrounding AI through the decades and even though I believe the progress the last few years has been real it is hard to deny the hype has been even bigger.

Still, while I don't agree with Mitchell concerning the progress of AI and the risk of rogue AI, her book is well worth a read so long you know it is quite dated. Keep you eyes open for a future updated edition!