AI Snake Oil - My Thoughts on the book Not another book review

AI Snake Oil - My Thoughts on the book

Not another book review

David J

Recently I read the book AI Snake Oil. This book is a must read for all AI Entrepreneurs, Practitioners and Educators. I would say it should be a minimum required reading for Educators especially since Entrepreneurs typically tend to over value and oversell their claims and solutions. I would categorize most of the practitioners experimenting with AI into the AI educator space. This article is not a conventional book review which tends to focus much on what the book does well and where it fails. My article just picks up salient points made in the book and I add my take to it

The book does a great job differentiating between the two major categories of AI we have today

1. Predictive AI

The Kind of AI where machine learning is the focus. AI, where large data sets are fed into data lakes, labelled to construct models. The typical use cases of these models are to predict the end outcome or quickly categorize and enrich profiling information when it encounters new data.

2. Generative AI

The kind of AI whose focus is deep learning, building and improving neural networks. The current generation of this effort has been in building large language models. As the name of the category goes, the use case of these models are to generate or produce new content such as text, images, videos and of late agentic or autonomous actions.

A Good prediction is Not a Good Decision

Ever talk to people who claim their decisions are based on empirical science. That they believe in data driven decision making. A closer look into their own lives would reveal all kinds of anomalies where they made decisions based on gut, emotion etc. What we would find is we retroactively use data to justify our decisions. The real question is what decisions do we take when we don't have data available? In real life most moments call for decisions without much past data.

To prove this, one just has to take a look back at life altering decisions that were made in one's life. You will find very rarely were they made with 'Empirical data'. The book rightfully points out the fallacy that all great decisions are made with data. The book has plenty of great examples. The accuracy of a prediction is only as good as your past data. When you encounter newness as we often do in the real world, accuracy will actually lead us astray.

Does that mean we throw away all past data and be done with predictive AI?

The answer is a definite no. However, that is where the education part comes in. One needs to be well educated on how these prediction machines work and take these predictions with a grain of salt.

Is GenAI an Existential threat?

"The fox knows many things, but the hedgehog knows one big thing"

What do we mean by Existential threat here? Is it the loss of livelihood of the billions? or threat to humanity? Most of the A.I. Doomsayers view the former as the threat. However, is this view justified? Most AI practitioners tend to overestimate and be overoptimistic about Gen AIs capabilities.

Ever worked with the Engineering types or managed a team of engineers? Just because they have a solution in their head, they tend to mostly underestimate the effort involved to deliver it. This underestimation is in the order of magnitude. The spirit animal of an engineer are 'hedgehogs': Once they have a solution in their head, they tend to burrow and dig deep to get to the root of it. However, they underestimate the effort needed to integrate the solution into the world where others can use it. Most of the AI practitioners today are the engineering type and the hype they project into the world is nowhere close to the kind of threat the AI doomsayers want us to believe. The book does a great job shedding light on some of this hype.

AI the boon and bane of Social Media

The book talks about how Facebook and other social media apps were successfully able to use AI to reduce cost on content moderation. While benefitting from labor arbitrage and low wage workers to operationalize content moderation. Facebook and others were able to further save billions on content moderation efforts, by using Al. The book offers several examples of where content moderation failed miserably.

Everyone remembers the horrific photograph of the Napalm girl, the nine year old Vietnamese girl running naked after being severely burned. While the picture shows someone naked It is however historically significant in terms of depicting the horrors of War.

Facebook's decision to not allow that particular image on its platform was more a decision of their policy. What is hiding in plain sight here is Facebook’s knowledge in the limit of AI on content moderation.

AI, while it may be good at pattern matching and predicting data to place it in a category, cannot know the intent of the producer or the creator of the content. When AI decides to not allow content whose intent is to provoke thought, it tends to brush over the ugly facts of human life, the impact of War being an example. Hence AI Can Sterilize the revolutionary and thought provoking effects that art can have in this world.

“AI sterilizes the revolutionary effect of art in this world”

Is AI a solution looking for a problem?

Many of the entrepreneurs today tout this mantra.

‘Entrepreneurs are problem solvers' Hence the best path to solve one is to look for a real problem. With AI however this may not be the best or at the very least the only approach. Think about the infrastructure setup of the internet. Much of it was done without looking for a problem to solve. We found a way to connect computers. We did more of it in the hope that the whole world will be connected one day something magical will happen.

Connect the world and see what evolves was the shift in paradigm for the internet.

This connecting the world brought about email, chat, cloud technology, SaaS applications. Mobile Apps and so on.

Much of AI's impact on our world is going to be a shift in paradigm.

• The re-evaluation of what we might say human intelligence
• inventions of new business models
• safer ways to research high risk innovation
• decluttering the noise from true creativity and works of Art.

The book was great in terms of provoking this kind of thought. At Least for me :)

Tech Elitism

Much of the AI Community and research today are focused primarily on improving the performance of LLMs on benchmark datasets. This is a way for them to compete in the marketplace of AI Engineering. This also is a way for a dominant player to emerge. While this is not a bad thing. It tends to widen the gap between the engineering community and sort of the common folk.

The breakthroughs accomplished in engineering are desperately pushed into the world as some sort of a solution everyone can benefit from. As a “cure for cancer” the cure which a regular person can’t even comprehend. This can lead to a sort of tech bro culture of elitism. It leads to a false dichotomy as the book points out. Only tech companies can understand and manage AI technology.

This is not necessarily the case when AI especially becomes all pervasive. We would certainly need experts in the field of humanities, philosophy, culture theorists etc to develop a responsible roll out of AI and mediate the effects of AI on our daily lives.

Consider subscribing to my blog on Substack where you will get access to Videos, Articles and Personal Musings. A bit more under the hood view. See my last few videos/writings below

Substack link