As Artificial Intelligence enters the stage in many companies and products, more and more people are confronted with judging these algorithms that are learned from data.
But how the heck would you know whether these models make sense under the hood? They automatically learn from data, and are quite intransparent as to how decisions are made. This book teaches you, in a non-technical manner, what to look for when evaluating and scrutinizing AI models. After reading, you should have a good intuition about algorithms are learned from data. You will understand why Amazon had to retract their hiring algorithm, why neural networks can be fooled with simple stickers and how to distinguish between AI snake oil and solid products.
The focus is on algorithms that are learned from data, but many concepts are also more broadly applicable to any algorithm, such as thinking about the incentives of the creators and fairness considerations.
Who is this book for?
- Auditors who are confronted with products and processes that have a learning component (AI).
- Managers who want to make good decisions about deployment of machine learning models.
- Investors who evaluate business ideas based on AI.
- Insurers who have to assess risk involving learning algorithms.
- Policy makers who have to create laws and policies around the deployment and usage of AI.
- Journalists investigating AI.
- Data scientists, machine learning engineers and statisticians, who want to take a higher level view on modeling, or have to document the data collection and modeling process.
- Concerned citizen fighting AI overlords who want to enslave humanity. It’s not the Terminator we should fear, but predictive policing, social credit scores and LinkedIN reply suggestions.
- Grumpy people. Like the ones lying on the sofa, right hand in the bag of chips, left hand holding a can of beer, watching soccer and complaining how a million dollar player missed the ball. Same but looking to do the same for AI.
Requirements for Reading this Book
None. I keep out most math and technicalities, and stick to the high level concepts. For example, to understand AI, we will be discussing mathematical optimization which is essential for learning from data.
What this book is not about
It’s not a list about different neural network architectures or anything, there are enough books and courses about that. In fact, there is almost no technical detail about specific algorithms in this book, since I believe it much more important to understand the general principles.
It’s also not about requirements for specific industries. It’s also not a formal auditing book. So it will not get you through an audit for your next medical device or get your fancy financial product im
