Recently finished this one. This is a pop philosophy book on why, epistimologically, it makes sense to just use expert opinion (without checking their reasoning or reasons carefully yourself) generally. It was written just before widespread AI use, with more of a focus on vaccine denial and conspiracy theory. But I think it has a lot to do with AI use, especially in schools and conversation. Many students are turning to an AI to complete homework and study, with mixed (and often disastrous) results.
Importantly, the book only makes the moral case in a weak sense. The main good is ‘justified, true beliefs held’ with a minimum of error, time, and effort. It does argue that we have to use expert testimony for most of our beliefs (wikipedia) anyway; and that the best way to manage expert incentives and opinion is institutional (and not trying to become an expert in everything yourself).
Overall, I found it troubling to read. The rationalization is pretty compelling, and there are few glaringly obvious leaps or gaps. Worse, the rationalizations aren’t hard.
Disclaimer: did not read this book like OP did, I just read their summary/reaction)
The biggest problem with these ideas is that people now classify “AI” (next word generators) as “credible”. My boss is in full-blown AI psychosis. Trusts incorrect answers from GPT/Gemini over my 15 years experience doing my job. Can’t WAIT until GPT tells him to fire me. That’s going to be a great day.
I literally just had an AI discussion at work this morning with my boss that is suuuper pro AI and I cannot seem to get through to him that it is a fancy next word predictor, and does not actually know what it is saying.
I’m mostly in the case of calculus/pre-algebra homework, where objectively the recent models are more accurate than my typical student. They will miss a trick question sometimes, or make a calculation error. But not as much as a struggling student?
I bet they’re still trash vs Wolfram alpha.
Used to keep students away from Wolfram with story problems + extra parameters. Sadly, several LLMs have tool-calls to, for example, sage-math. Pretty sure some of them have wolfram integration.
Oh.