MIT Press Announces:
"The AI Conundrum"

MIT's April 2023 Book Announcement

How well do you know AI? Take the self-assessment and read chapters 1 and 2.

Self-Assessment Read Chapter 1 and 2

Why We Wrote This Book

  1. Did you take the self-assessment? If you didn't score above 100, you could benefit from leveling up your understanding of how AI works. We wrote this book for you.
  2. This book has plenty of case studies of how to use AI effectively to gain massive savings in time and cost, but we also address why AI can be risky in certain applications. This book provides a framework to access risk, and design governance systems.
  3. Business decision makers and government policy makers will be well served by investing time to understand AI and to have a framework that supports responsible use. Some in the field of artificial intelligence today, especially those with an economic interest in AI, tend to gloss over AI’s weaknesses. But no one can afford to wear blinders when it comes to artificial intelligence. We all need to study the limits of AI in order to help us avoid risky overestimation of AI now, while potentially building better AI for the future.
  4. AI is an industrial revolution on par with electricity. We wrote this so you will draw lessons from previous waves of industrial revolution such as looking beyond the mere replacement of current practices toward more transformative applications. After discussing the broader arc of artificial intelligence’s significance as a new wave of industrial revolution, we focus on use cases to demonstrate the value that can be generated with AI.
  5. Amidst the use case that demonstrate AI strengths, we weave together use cases where the manifestation of AI’s weaknesses has cost business billions in losses. It is important to understand how AI can be misapplied. In some cases, AI applications have cost lives and freedoms that are unconscionable. We explain how to avoid the risks. We also note that some risk are easier to offset along with the more challenging use cases where AI’s weaknesses may not be solvable or easily mitigated.
  6. No book on the strengths and weaknesses of AI is complete without a clear eyed discussion of AI bias. By the end of Part 3 of the book, you will will understand why bias emerges in AI systems, and countermeasures to help mitigate it.
  7. In sum, we wrote the book so that you can know more about the inner workings of AI and have a framework for accessing cost, benefit and risks. A 25 question checklist in the appendix provides a practical guide that can help ensure you gain the maximum value from AI while managing the risks.


Advances in artificial intelligence are accelerating so quickly, and its presence in our daily lives has become so pervasive, that it is becoming essential for people to develop a solid understanding of how it works. Take our 12-question self-assessment at to see how you score. If you are at the point of most business people today in terms of your understanding of artificial intelligence, this book can help you level-up with a foundational understanding of artificial intelligence in a way that is easy to understand yet deep enough to be able to apply and implement AI effectively and responsibly. This volume is not for technical experts — we aren’t teaching you to write code. There are plenty of books available on coding and systems design and the detailed mathematics of AI. Nor is it for those looking for philosophical discussions of AI — such discussions without a grounding on how AI really works can be counterproductive. Our approach is to explain the mathematical underpinnings of AI in a way that is accessible to business people and policy makers, and to address the philosophical and ethical implications in a way that is concrete. Students will also find this a useful entry point into the world of AI. The book is filled with practical applications.

At the core of the AI conundrum is the observation that AI’s greatest strengths can also be its greatest weaknesses. Mathematicians call artificial intelligence a universal approximator because it has the ability to recognize a pattern from any set of data it is given, making it an extraordinarily powerful tool. But AI is not perfect and in the process of finding patterns, it can find spurious as well as authentic patterns. Biases and imprecision can creep into AI’s output, and it can be very difficult to spot them. This can lead to disastrous results for businesses and government entities that rely on the power of artificial intelligence. Hence the AI conundrum: how do we make use of the strengths of AI while avoiding the weaknesses? How do we identify the risks so we can effectively mitigate them?

Understanding AI’s strengths and weaknesses requires a basic comprehension of how AI works. Part 1 covers these fundamentals in an accessible manner to provide insight into the workings of AI and develops a framework to analyze the risks inherent in various types of AI applications. This level of understanding will equip readers with the necessary foundations in order to have a grasp on why results can sometimes go badly awry.

Part 2 provides an abundance of examples that demonstrate the vast array of useful applications for artificial intelligence but also highlights the situations where AI goes wrong. The consequences may be trivial — but they may also cost lives and livelihoods. The risk framework helps to elucidate when AI’s failures are relatively inconsequential, such as in marketing and language translations, versus situations where the risk of AI failures can be more serious, such as in certain types of commodities trading, self-driving vehicles, and social media.

Part 3 shows why we so often find bias in the output of artificial intelligence, how systems can be tested for bias, and measures to help mitigate bias. Part 4 discusses future directions for AI, such as explainable AI and the integration of AI into robotics to further expand the sensory input for AI. We conclude with discussion of the risks of artificial intelligence and a discussion of why the industry should move to avoid anonymous and autonomous use of AI and instead embed AI with accountability by design.

Because advances in artificial intelligence are accelerating, we recognize that AI has the power to reshape our world, and any book can become dated quickly. Therefore, we have focused on a foundational explanation of how AI works and a framework for assessing risk that we hope readers will find practical and usable for the foreseeable future.