Power And Prediction: The Disruptive Economics of Artificial Intelligence

by Ajay Agrawal, Joshua Gans, Avi Goldfarb

Disruption resulting from the proliferation of AI is coming. The authors of the bestselling Prediction Machines describe what you can do to prepare.

Banking and finance, pharmaceuticals, automotive, medical technology, retail. Artificial intelligence (AI) has made its way into many industries around the world. But the truth is, it has just begun its odyssey toward cheaper, better, and faster predictions to drive strategic business decisions--powering and accelerating business. When prediction is taken to the max, industries transform. The disruption that comes with such transformation is yet to be felt--but it is coming.

How do businesses prepare? In their bestselling first book, Prediction Machines, eminent economists Ajay Agrawal, Joshua Gans, and Avi Goldfarb explained the simple yet game-changing economics of AI. Now, in Power and Prediction, they go further to reveal AI as a prediction technology directly impacting decision-making and to teach businesses how to identify disruptive opportunities and threats resulting from AI. Their exhaustive study of new developments in artificial intelligence and the past history of how technologies have disrupted industries highlights the striking phase we are now in: after witnessing the power of this new technology and before its widespread adoption--what they call "the Between Times." While there continue to be important opportunities for businesses, there are also threats of disruption. As prediction machines improve, old ways of doing things will be upended. Also, the process by which AI filters into the many systems involved in application is very uneven. That process will have winners and losers. How can businesses leverage, or protect, their positions?

Filled with illuminating insights, rich examples, and practical advice, Power and Prediction is the must-read guide for any business leader or policy maker on how to make the coming AI disruptions work for you rather than against you.

Genres: Business, Artificial Intelligence, Nonfiction, Economics, Technology, Science, Finance

256 pages, Hardcover

Published November 15, 2022

About the author

Ajay Agrawal is a professor at the University of Toronto’s Rotman School of Management as the Geoffrey Taber Chair in Entrepreneurship and Innovation as well as the Professor of Strategic Management. Agrawal co-founded NEXT Canada, previously The Next 36 in 2010.

Reviews

[Narang Unnati · Rating 5 out of 5]

Loved reading this book!

Very CLEAR, well-crafted narrative and powerful ideas. I had earlier read the “Prediction Machines” by the same authors. In the original book, the main idea is that AI reduces cost of prediction. As prediction gets cheaper, we will use more AI. This book goes beyond a “point solution” that swaps out current tech with new tech (e.g., steam with electricity) or “application solution” that extends independent of the system (e.g., power in small unit) to “system solution” involving dependent parts (e.g., Ford's innovation).

My favorite part about this book is the cohesive set of ideas about how we are in the transitionary "between times" and what we need is a system-wide change to get to the next level. Dispels some myths and finds new potential avenues for the future of AI beyond cheaper predictions. Good examples from healthcare and other fields too. Very relatable and easy to grasp for a non-academic, general audience. Finally, lots of powerful storytelling in the book, e.g., David Friedberg started The Climate Corporation to sell insurance to farmers but turned out that AI could help farmers improve decisions on their farms. No one asked Friedberg: If my knowledge is no longer useful, who will need me?

[Karen Macartney oettl · Rating 5 out of 5]

This is a follow up to the team of author’s first book on the power of prediction and AI (“Prediction Machines” This one cuts to the heart of making faster, cheaper and better decisions and is a key book to read if you are fascinated by AI or if you think your business could benefit from better decisions for system-level innovation

[Rodrigo · Rating 5 out of 5]

After publishing Prediction Machines, the authors realized the necessity of considering not only the economics of artificial intelligence but also the systems in which the technology operates. At the time, most markets are not ready to benefit entirely from AI solutions.

Point solutions lower costs but don’t demand structural business change. Application solutions are broader but still don’t demand structural change. System solutions require a new vision for the whole market to deliver more value for the company and customers.

To help assess how close a market is to system solutions, they propose an AI Systems Discovery Canvas in chapter 18.

[Rads Ads · Rating 5 out of 5]

AI is becoming more and more relevant in today’s world, and it is set to influence every person on the planet in one way or another before long, if it isn’t already. Power and Prediction is a very thought-provoking and informative guide into understanding AI, how and why it is transforming the economy, and how the future of AI may unfold, disruptions and all.

If you want to prepare yourself or your business for the future of AI then this in my opinion would be a great place to start. Power and Prediction is extremely well researched and is written by collective authors with vast experience within this field. A fascinating read about a subject we have to confront, whether we want to yet or not.

[Hannah Guzman · Rating 4 out of 5]

Power and Prediction: The Disruptive Economics of Artificial Intelligence is an incredibly insightful and thought-provoking read that explores the impact of AI on the labor force and society as a whole. The authors, Ajay Agrawal, Joshua Gans, and Avi Goldfarb, have done a commendable job of explaining complex concepts in an accessible and engaging manner.

One of the most impressive aspects of this book is how well it is written. The authors have a knack for explaining complex ideas in a clear and concise way, making it easy for readers to understand even the most technical aspects of AI. The book is also well-organized, with each chapter building upon the previous one, leading to a cohesive and comprehensive understanding of the topic at hand.

Another highlight of the book is its exploration of the relationship between AI and the labor force. The authors do an excellent job of outlining the ways in which AI is already disrupting traditional labor markets and the potential implications for workers in the future. They also offer thoughtful insights on how society can adapt to these changes and ensure that the benefits of AI are shared equitably.

Overall, Power and Prediction is an excellent read for anyone interested in understanding the impact of AI on our economy and society. It is a well-written and insightful book that offers a compelling look at the "in-between times" we are currently living in as AI continues to grow and evolve. Highly recommended!

[Amanda · Rating 4 out of 5]

I try to keep up to date on technology but find AI to be very boring to learn about in general. This book kept me engaged, framed the premise in a digestible way, and helped me think about how AI could be used more powerfully.

If you are super up to speed on AI this might be too entry level but that is exactly what I found great about it. A must read if you are like me and want a foundation and framework for AI and it’s predictive capabilities.

[L L · Rating 4 out of 5]

I don't often review books for work on Goodreads, but this was so informative, I thought it'd be worth sharing here. The popular adaptation of artificial intelligence has grown exponentially in the last year, including the obvious talk about its implications for white collar work. Until I read this book, I don't think I really understood what AI was or how it worked. The authors use the example of electricity and its gradual adaptation as a technology, as a helpful metaphor to understand what we might expect from AI. When factories started adapting electricity, as a point source of power, instead of steam, the benefits were minor. It was cheaper. Then, electricity could be adapted at an application level-- changing the way it was applied in its use. Electricity could be much more easily turned on and off, allowing machines to be run as-needed, rather than all at once when the steam was being generated. But the real benefit of electricity wasn't realized until factories were redesigned at the system level-- machines no longer had to be located near the central shaft where steam was being generated. You could completely redesign the factory. And it took several decades before factories adapted to this change.
Using this metaphor, the authors then explore the possible applications of AI, especially for decision-making, and how it might reshape current systems of our economy. They distinguish between prediction (what AI does), and judgment that constitute decision-making. AI gives us much improved data about what is likely to happen. Humans still have to exercise judgment (or decide the parameters for judgments to be automated). An automated car is a good example--- automated car sensors give a percentage probability that a certain obstruction is a person. At what percentage threshold should the car actually respond accordingly? Set the percentage too low, and the car may never go anywhere, stopping mid-traffic, in the middle of the street. Set it too high, and people could die.
The book explores many different industries-- insurance, medical systems, banking, retail, and gives practical examples of how AI might be applied to redesign the system and what implication it has for who has power and who does not.
Written by academics, but still accessible, I also appreciated that this book did not have the typical starry-eyed embrace of AI that many business books do. It was much more descriptive about the implications. That being said, the book underexplores some of the risks of AI, though its chapter on AI and bias is interesting. The authors argue that because it is easier to measure and diagnose when AI is biased, it is easier to address.

[Ryan Weller · Rating 2 out of 5]

Interesting point about systematic change to truest leverage AI disruption, but shallow. Very repetitive. Could have been condensed to 100 pages or boosted beyond academic arm waving

[Kara · Rating 2 out of 5]

Very simplistic description of AI - basically the power of prediction - and the rationale of why it hasn’t yet had the impact that it will eventually have.

[Bela · Rating 1 out of 5]

Genuinely struggled to finish this one. The majority of the book is spent rehashing the same point; that the adoption of AI is best served by re-inventions of existing systems, as opposed to being inserted into existing workflows. Am I five? Not only is this same, apparently earth shattering, insight made so often I feel like I’m doing a Duolingo Spanish lesson, but it’s completely obvious!

This leads into my real problem with the book. It takes this step (painfully many times) yet again and again fails to take the equally obvious next step. How does the system of free-market capitalism function in a world where the only asset a wage-earning person has, their labour, is worthless? Is this not the most important system-design question this issue has to offer? The only mention actually, is when they say “currently AI hasn’t replaced many jobs yet, so we have some breathing room.” Lol??? The authors instead give a handful of examples of current industries that can grow their profit margins by using machine learning to replace their employees and obliterating any privacy a person may still think they have to train their model. Naturally they also cite the healthcare industry constantly, since it’s hard to argue “don’t save these people’s lives with this new technology.”

Are these three “thought leaders” consciously pushing an agenda or are they just dumb as rocks?

[Kathy (semi-hiatus) · Rating 3 out of 5]

I don’t like this book, but it’s good enough compared to similar books. For me, one criterion for assessing a book is whether it attracts me to something I normally dislike. I've never been a fan of business/tech, but I would recommend this book for the timely, thought-provoking ideas about our experience with The Between Time in AI development. It’s just a lengthy thought experiment. It's not written with the utmost clarity, yet the language is clear enough for people to grasp the idea.

AIs can be applied in three manners: point solution, application solution, and system solution. While the former two focus on enhancing the existing system in a business, the latter redesigns the business model to more efficiently take advantage of AI's potential. The central idea centers around the predictive power of AI at a lower cost, and as prediction becomes more accurate and cheaper, a potential application at scale will trigger a series of systemic changes as decisions are often interdependent across multiple divisions within an organization. This would require a system solution to check the setbacks such as the loss of reliability and uncertainty. The book also discusses the decoupling of prediction and judgement as an outcome of AI application.

Books technical as such can become overly elaborate for the general public and thus lose their educational purposes. Despite the occasional appearance of lengthy sentences, the authors attempted to write in plain language and use examples relevant to an ordinary life, making the information accessible to non-academics. The cohesive idea of The Between Time permeates throughout the book in the discussions of AI development, application, and the outcomes. Humans are not perfect, and so aren’t prediction machines. It provides arguments supported by reliable evidence that dismiss the concern about AI having too much power. On the other hand, system solutions will never be perfect: regulations and business environment of different organizations, including power dynamics and organizational interests, will prompt resistance to change. Considerations of potential obstacles further indicate the uncertainty about the direction of AI in the future. I’d only be more satisfied if the book gets down to the root of AI and distinguishes it with computational mechanics. But overall, the authors have provided a comprehensive overview of AI's prospect and what it takes to fully leverage its potential.

[Scalar42 · Rating 2 out of 5]

Perhaps I set my expectations too high, but this book offers little insight into either economics or artificial intelligence. I believe you could mostly replace the term 'AI' in the book with 'computers,' and it would essentially mean the same thing.

The authors begin by proposing a categorization method for innovations as "point solutions," "application solutions," and "system solutions," using the adoption of electricity to illustrate their points. Initially, I was intrigued, wondering how the arguments would unfold, but I soon became disappointed. The book becomes distracted, filled with pages of examples that sometimes seem far-fetched. I understand that the authors might need to include examples when necessary for a general audience, but on the other hand, the book does not delve deeply into its main topic—namely, how exactly AI models and systems have been impacting existing societies and industries, especially in the economical perspective. Show me the data, cases, interpretations, and possible solutions, not just illustrative examples that merely scratch the surface.

Also, the constant analogy between electricity and AI seems way off to me. AI models might be seen more as a new computational paradigm than as a new pipeline for energy or information flow. If an analogy is necessary, I think the internet, which enables distributed computation and transmission with real-time collaboration, offers a much better comparison.

[Susan · Rating 2 out of 5]

This book discusses AI from a business and economic angle. The discussions on AI are much more generic than I expected given the credentials of the authors, and most of the book is quite repetitive. One of the few interesting points for me is on the comparisons of point vs application vs system levels of AI applications. Most current applications look for where AI could swap out and improve on an existing task in a system. However, the benefits there are only incremental, and it isn’t where the biggest opportunities lie. The parallels with the 2-decade delayed adoption of electricity was also illuminating.

[Georgia Green · Rating 2 out of 5]

ChatGPT really said your thesis is kind of wrong …

[Stetson · Rating 3 out of 5]

In Power and Prediction: The Disruptive Economics of Artificial Intelligence, Ajay Agrawal, along with co-authors Joshua Gans and Avi Goldfarb, extend their analytical framework introduced in their prior book Prediction Machines. The thesis of this follow-up is that the transformative potential of AI lies not merely in its technical capabilities, but in the systemic and institutional changes required to fully harness those capabilities. Today, we're living through a test of this hypothesis as OpenAI, Anthropic, Google, Meta and others have democratized prediction machine access in addition to ongoing proprietary efforts. Nonetheless, it seems institutional inertia and the limitations of genAI have thus far held the promised AI revolution up.

Agrawal and company argue that we are in a “between times” phase (or at least were in 2022 - I don't know if the author's have updated their position) where AI systems have become very good at prediction, a foundational component of decision-making, but society has not yet constructed the complementary infrastructure necessary to allow AI to truly upend existing industries or social practices. AI can perform narrow tasks exceedingly well, but for that to yield wide-scale disruption, new business models, workflows, regulatory frameworks, and norms must be developed. It is unclear is the genAI boom has falsified their notions about the specific versus general capabilities of AI. There are AI critics who argue that AGI/ASI is not going anywhere and specific-problem AI is the way to go. Either way, the point about wide-scale disruption still stands. This is likely to remain true, especially as the progress from scaling training sets has plateaued and the black box/hallucination issues with genAI stubbornly remain. The technology itself is not yet powerful enough to simply outrun skeptical stakeholders, bumbling bureaucracy, and sclerotic structures.

Being economists, the authors ground their argument in economic theory, particularly the notion that the cost of prediction has dropped dramatically, much like how the cost of computing or storage once did, creating new optimization opportunities. Yet prediction alone is not equivalent to action. The value of AI, they stress repetitively, comes when prediction is leveraged by decision-making systems/agents, which themselves are embedded in bureaucratic institutions with rigid rules, entrenched interests, and often outdated assumptions. They introduce the concept of the “judgment gap,” which refers to the human component still needed to make decisions based on AI predictions. Three years later, progress on bypassing this "judgement gap" doesn't appear to have materialized, though AI proponent (doomers and e/acc types) will definitely argue that progress in autonomous agents may be bypassing the "judgement gap."

The book provides a variety of real-world examples, from healthcare and law to autonomous vehicles, to show how friction arises when institutions are not yet ready to cede decision authority or restructure processes around AI. The authors advocate for what they call "systemic invention," a deliberate rethinking of how decisions are made and institutions are designed, in order to realize AI's full potential.

There isn't much particularly gripping or objectionable in this sometimes dry text. I'm a bit underwhelmed by the thesis given that it follows linearly from those with some sense of the history of Schumpeterian innovation. I appreciated that this contained serious thinking about AI without the usual Bay Area philosophizing about AI ethics and the prospects of AGI.

*Another book that could definitely have been an article and/or essay and probably is in the literature.

[-Anne- · Rating 3 out of 5]

Talking about the system-level disruptions to industries that AI will cause but not mentioning graphic design, art, or creative industries is a bold choice...

Nevertheless, the machine bias and judgment sections were insightful.

[Reader · Rating 3 out of 5]

Learnt quite a bit, rather a long and repetitive at parts however

[Nag Jayaraman · Rating 2 out of 5]

Light introductory level content on trending topic. Typical book of professors on speaking circuit, collating bunch of anecdotes and material they read recently.

For the topic the content is too lighter to be rated high

[Dónal Murray · Rating 3 out of 5]

Great content but very repetitive and seemed to labour the point in places

[Mae · Rating 3 out of 5]

The major topic is to discuss decoupling prediction and judgement and it extends to more than 300 unnecessary pages.

[Jason Friedlander · Rating 4 out of 5]

This is a clearly written book about the ways the development of Artificial Intelligence capabilities can affect how we approach different fields. It treats AI as a powerful tool for prediction and explores the importance of adopting system-level changes in order to best utilize it. This is in contrast to point-level solutions, which can be helpful in small doses, but are limited in long-term effectiveness. It argues that the most important effect of AI is its ability to decouple prediction from judgment. The perspective of the book mostly considers business opportunities and ways to better improve health care, but its lessons can theoretically be applied to other fields as well.

A lot of what the book offers seems to me unconcerned with privacy and AI safety. The effectiveness of AI on a mass scale for different businesses and industries appears to require allowing for even more data to be taken from us and used to conduct assessments of our behavior, with which the possibilities of our future actions will thereby be constrained. It also worries me that the reliance on so-called superhuman AI predictors transfers the culpability of unethical business practices from organizations to “algorithms” that may occasionally carry “unforeseen errors and biases”. The worst part is that without transparency we won’t even have the ability to protest the specifics of their deployment.

Worth reading but I’m not a fan of the entrepreneurial perspectives suggested (especially with the insurance industry).

[Omar Shokry · Rating 3 out of 5]

Some say that instead of reading about what AI is and or what potential does it have as an emerging technology that could change our entire reality is less useful than actually learning and working with AI tools so that we can get ahead before its too late and we all lose our jobs haha. I humbly disagree. How could you possibly find the drive and motivation needed to spend all the energy needed to learn emerging tools if you dont even know the “Why” of it all? How could you possibly approach a technology that you are probably (irrationally) afraid of if you believe all the jargon about it replacing you and your job? Whats worse, people get emotional and hyped up because they either dont want to miss an investment opportunity or they want to brag to others of how AI “made them 10x more productive” or “makes them x amount of $/day without lifting a finger”. Im sure you read such things on twitter,instagram,tiktok,etc. Anyway, if you are the kind of person who likes to detach from all the hype, stop for a second, think about everything you have seen and read, decided that you dont know anything (just like i did) and you actually have a genuine curiosity of what AI TRULY is and what kind of potential it may or may not have, then this book is for you! I didnt give it more than 3 stars because its a bit abstract at times and keeps repeating the same point over and over for some elements of the framework used to analyze the technology and its connection with entrepreneurship. But I get it. Also this book has one of the best definitions of AI that I have encountered so far (may be the best actually). Enjoy!

[alexander · Rating 3 out of 5]

Nice writeup with some very good perspectives, even if it came out the week of ChatGPT’s release(!)

However the real world examples are just a bit too long and meandering, occasionally even naive,
detracting from the often highly informative insight that precedes them.

[Teo Hong · Rating 4 out of 5]

Provides a broad analysis into AI and it's applications.

Just like any other technology, AI is neutral at its best, and the adoption merely mirrors what humans want and see through it's lens.

I'll recommend this to anyone wanting to know more about the nature and non-technical aspects of AI

[Afsheen · Rating 4 out of 5]

Very interesting and well written narrative of powerful ideas! I like how this book explains how we've been in transitionary periods before, and we're in one again. The beginning of the book was more interesting to me than the rest of the book.
AI is all about prediction, human prediction.
Good examples from other industries as well, such as healthcare and insurance.

[Zoe Pacalin · Rating 5 out of 5]

Finally!! Next time someone wants to shoot the shit on ai taking over the world i will send them this book 🙏 sound thinking with (technical) knowledge and perspective (on other major historic shifts). I especially like the bits about fewer rules:

Al doesnt take decision power from human But it might change which humans make the decisions

The place ai taregts well is where there are rules
...
Rules tolerate but do not embrace and leverage uncertainty in the ways ai can

Decouple prediction and judgement to calc expectation, then decision is clear
Rather than avoiding situations we need to make decisions by abiding by rules

Point solution- no sys change
Application solution- new procedure no system change
System- improves existing by changing dependent procedures

[Dennis Murphy · Rating 5 out of 5]

Power and Prediction: The Disruptive Economics of Artificial Intelligence by Ajay Agrawal, Joshua Gans, and Avi Goldfarb is a pretty powerful and useful book, but its a little out of date. The book is mostly about machine learning and the power of prediction, and what that means for the tradeoffs between judgment and prediction. We take a tour down multiple illustrative examples, and we see where these types of systems work and where they fail. We further go on to design best practices as a how to in order to incorporate them. However, the narrow range of machine learning originally described has since widened in the outside world, and it is unclear whether the lessons the three authors proposed remains as accurate for general purpose models. I, in particular, am skeptical about the verification and data problems.

Still one of the better books out there.

[Soniaa · Rating 5 out of 5]

The Author immerses you into a think tank of the extensiveness Artifical Intelligence has to offer in our modern day. The Author cleverly reminds you of the importance of our adaptability to change, and how this shapes our quality of life.
This book has been written with great reflection and research. It truly takes you on a journey, that naturally makes you appreciate that the world we live in has endless potential for the way we live our lives, for the here and now and in the future.
Artifical Intelligence is all around us, it makes decisions on our credibility, our characteristics, our expectations, quality of life and much more!
A very enjoyable read, that has been written extremely well. A real page turner

[Juan Rivera · Rating 4 out of 5]

Reading this book felt like digging into an ancient text—like those old medical books that talked about balancing humors or bleeding patients to cure them. Or like a philosopher explaining the stars through divine will, long before telescopes and science.

That’s how I felt… except this book isn’t actually that old—it was written just five years ago.

The concepts are still relevant, but the author couldn’t have foreseen the last link in the chain: generative artificial intelligence.

I considered putting it down halfway… but the examples are still so timely that I ended up reading it through to the end.

[Raymundo · Rating 3 out of 5]

2.5/5. Un libro interesante sobre cómo adaptar la IA a nuestras vidas. El libro remarca la importancia de pensar en el sistema, no en la sustitución de procesos. El libro es aburrido, no hay casi diagramas, su lectura es complicada. Lo leí para ver qué libros recomendar a estudiantes, y creo qué hay libros mejores para introducirlos en el tema.