AI Snake Oil: What Artificial Intelligence Can Do, What It Can’t, and How to Tell the Difference

Two computer scientists, both admitted AI enthusiasts, shine a very sharp light on the sea of misinformation, over-enthusiasm, confirmation bias, and misleading claims about AI. The book is written for the layman, and begins with possibly the most important piece of information a consumer or business exec needs to understand: "AI" is not one thing. There are as many different kinds of AI as there are different modes of transportation. If the only word a media report or a marketer ever used was "vehicle," then how could you tell if they were talking about a bicycle, a car, an airliner, or a ship? Yet they are very different things.From that foundation, Narayanan and Kapoor survey the different applications that go by the name "AI," and show why some are valuable and some are as the title describes, snake oil. Of all the current literature on the subject, this is the most informative, least biased, and most useful for anyone contemplating an AI application for vetting job applications (pure snake oil), medical diagnostics (choose carefully, much snake oil), or as a research aid (potentially valuable).With this book read and understood, anyone is armed with the knowledge they need to sort the hype from the good information. I would argue that without this book, unless you are already trained and experienced in the field, you have no chance of figuring that out.

An important and valuable look at a technology that is getting a lot of attention, but not very much of it thoughtful or critical. This book applies a more clear-eyed vision to the promise and the problems of AI.

This book does a great job of explaining AI tools and teaches you how not to fall for the hype.

This book describes precisely what we knew but didn't dare to say. I recommend it to everyone who is wondering what AI is and why it suddenly appeared. And yes, I have a partial degree in AI and am still partially involved. Congratulations to the authors.

I recommend this book to anyone who is thinking about the impacts of AI on individuals and society, and how to play a role in shaping the future towards positive outcomes. The book presents a clear and convincing exposition of AI's limitations and harms and offers a constructive perspective on how to think about them. The examples used to illustrate the points are relevant and top-of-mind for any reader thinking about this space. Bravo!

A must read for anyone in the position of adopting an AI solution

I was expecting much from this, but this is essentially a summary of what has happened in this field which is open and readily available in the field in open resources.

Thoughtful, well-written, and informative.

Book goes away from hype and criti-hype, explaining in layman terms what AI is feasible to do today, identified serious harms, challenges, dangers but also opportunities that AI field practitioners, researches, enterprises, governments but also each of us as individuals must steer for a better future

This is a different book and convey a different message - AI is not a panacea! The book contains many examples of the false claims and failures of AI. When almost everyone is going in a direction the authors have chosen to go in a different direction which is a bit risky. Companies and individuals are riding on the wave / hype of AI and making windfall gains - from course sellers, marketing professionals and professional scientists. In fact, many hardware (laptop) companies have started selling AI laptops! There few points I want to make about the book. Here are those.1. Authors make a case that predictions are not possible! This is quite opposite to what hard science (physics) tells us - everything is 100 % predictable if you the initial conditions and laws of dynamics (even in case of quantum mechanics). Authors may need to change their views about prediction.2. Authors are right that due to lack of a common definition of AI anything can be called an AI and that is a big problem.3. Authors selectively give examples of cases where AI did not work. This should not consider different from the case of weight loss products or fairness creams in India.4. AI companies and individuals are not honest this is also a claim but it should not surprise us. It is hard problem to crack that why people should not be dishonest when being dishonest really helps. Look the case of research publications - there are examples of papers being retracted from the prestigious journals. Paid publications are common.5. I 100 % agree that big tech companies are trying to make profit out of the hype being created - particularly "existential risks" of AI.This book at least makes people aware that all that is yellow and shiny is not gold.I think a detail investigation is needed to look into the marketing material being sold as research work. One of its examples is technical report of GPT4. There is nothing for researchers in that 100+ pages documents. Apart from that benchmarks and performance matrices being used by researchers are cherry picked and completely useless of real-life validation of the models.

Impressive breadth of references provides reassurance that there are many streams of AI and delivers the book's promise in spades - What AI Can Do, Can't and How to Tell the Difference.

Very interesting perspective and also highlights how AI can be used and misused

A must read for anyone who is interested in AI! Not only for Computer or AI professionals, but for all of us, whose lives are impacted by AI.