Start free trial
Trending New Genres Authors
EnglishEnglish
EspañolSpanish
简体中文Chinese
繁體中文Chinese (Traditional)
FrançaisFrench
DeutschGerman
日本語Japanese
PortuguêsPortuguese
ItalianoItalian
한국어Korean
РусскийRussian
NederlandsDutch
العربيةArabic
PolskiPolish
हिन्दीHindi
Tiếng ViệtVietnamese
SvenskaSwedish
ΕλληνικάGreek
TürkçeTurkish
ไทยThai
ČeštinaCzech
RomânăRomanian
MagyarHungarian
УкраїнськаUkrainian
IndonesiaIndonesian
DanskDanish
SuomiFinnish
БългарскиBulgarian
עבריתHebrew
NorskNorwegian
HrvatskiCroatian
CatalàCatalan
SlovenčinaSlovak
LietuviųLithuanian
SlovenščinaSlovenian
СрпскиSerbian
EestiEstonian
LatviešuLatvian
فارسیPersian
മലയാളംMalayalam
தமிழ்Tamil
اردوUrdu
Searching...
SoBrief
Power And Prediction

Power And Prediction

The Disruptive Economics of Artificial Intelligence
AI creates lasting advantage only when organizations rebuild how decisions are made around it.
by Ajay Agrawal 2022 256 pages
4.22
7k+ ratings
Listen
Immersive
V2.3
Amazon Kindle Audible
Try Full Access for 3 Days
Unlock listening & more!
Continue

Key Takeaways

Treat AI as a redesign of decisions, not a cheaper prediction plug-in.

Prediction is the starting point. The authors define today’s AI as technology that makes predictions from data, not as a machine that independently thinks or decides. Its immediate use is often simple: replace an existing prediction with a better one. Canadian fraud-detection company Verafin did this for banks, where predictions were already central to approving transactions and the workflow could use improved fraud detection without major changes.

Systems unlock the larger prize. Electricity offers the historical parallel. Two decades after its public demonstration, only about 3 percent of US households had electricity. Its broad productivity gains arrived when factories were redesigned around electric power, not merely when motors replaced steam engines. The authors call AI’s equivalent delay The Between Times. In a 2020 survey, only 11 percent of organizations reported significant financial benefits from AI.

Analysis

The electricity analogy clarifies why a powerful invention can coexist with disappointing productivity figures: organizations need complementary investments, skills, and routines before a technology changes output. This resembles the broader economic idea that general-purpose technologies reshape many sectors only when paired with innovation around them. The analogy is useful, but not a timetable. Electricity and AI differ in cost, infrastructure, and pace of diffusion, so history supports patience without guaranteeing that AI’s gains will arrive in the same form or distribution.

Choose AI solutions by how much of the surrounding system must change.

Three routes to value. A point solution improves an existing procedure and works independently of other changes. An application solution enables a new procedure that can also operate independently. A system solution improves or enables procedures only when related procedures change too. The dividing line is dependence: can the new prediction create value on its own, or must the organization alter other parts of its operation?

Electricity makes the distinction visible. Replacing a steam engine with an electric motor at a factory’s power source was a point solution. Giving each machine its own electric motor enabled a new application, but the factory still had to accommodate the old layout. The larger gain came from redesigning the factory floor and workflow around flexible electric power. AI has the same range, from sharper fraud predictions to new products and redesigned operations.

Analysis

This classification is valuable because it makes implementation costs part of the strategy rather than an afterthought. It also discourages treating every failed pilot as evidence that the model is weak. A prediction may be sound while the surrounding process makes it unusable. The categories are contextual, however: a tool can be a point solution in one organization and a system solution in another, depending on existing capabilities. Leaders should map dependencies before assigning a project label or promising a return.

A prediction estimates what may happen; judgment decides which outcomes matter.

Three inputs shape a decision. Data supplies information about past cases. Prediction uses that information to estimate what is likely in a new case. Judgment assigns value to the possible outcomes. In a bank, AI might estimate the chance that a transaction is fraudulent. People still have to weigh the cost of approving fraud against the cost of wrongly declining a legitimate purchase. Better prediction cannot settle that trade-off by itself.

Correlation is not a decision rule. Toy sales and advertising both rise around Christmas. That pattern alone cannot show whether advertising drives sales or the holiday drives both. Predicting what happened under familiar conditions is different from predicting what would happen after changing them. To establish cause, researchers may need experiments, such as randomly showing different customers different advertising strategies.

Analysis

Separating likelihood from value makes hidden assumptions easier to inspect. A model may be accurate on its chosen target while the target itself, such as spending rather than health, encodes a poor proxy for what society cares about. This is why decision quality cannot be reduced to predictive accuracy. The distinction also helps organizations assign responsibility: model builders estimate outcomes, while accountable decision-makers set the trade-offs. Those roles can collaborate, but neither should be mistaken for the other.

Use AI to expose uncertainty that old rules and comforts conceal.

Rules are precommitted decisions. People and organizations often follow a rule because gathering information and deciding each time is costly. A rule can also make a larger operation reliable. But it may hide uncertainty and create waste. Many travelers leave early for airports because traffic, security, and flight delays are hard to predict. Airports have responded by making long waits more comfortable, even though their stated purpose is to move passengers efficiently.

Better information can change the rule. Navigation apps can estimate travel times and help passengers choose when to leave instead of relying on a fixed buffer. In greenhouses, AI can predict pest pressure, potentially freeing growers from practices such as keeping facilities small for easier inspection. The opportunity is not simply to automate the old routine, but to ask which costly protections become unnecessary when uncertainty is better measured.

Analysis

The airport story shows how institutions can adapt to the behavior a rule creates, rather than remove the source of the behavior. Similar feedback appears in risk management: a guardrail or safety procedure may look wasteful precisely because it prevents the bad outcome it was built to prevent. Better prediction can sharpen these trade-offs, but it should not be treated as permission to discard safeguards automatically. Good redesign tests whether a protection is still valuable and who bears the cost if it is removed.

Build for new value, not merely cheaper labor or isolated task automation.

Task counts miss system change. A job can contain many activities, yet automating one activity may save no money if the employee is still needed for the rest. The authors argue that the more consequential opportunity is to reorganize work so better predictions create value that the old arrangement could not deliver. An AI recommendation system, for example, has limited effect if the rest of a company cannot respond to what customers choose.

Health care shows the coordination problem. An AI may predict disease more accurately, but hospitals still need suitable treatments, incentives to use the prediction, clear liability, and a workflow that connects diagnosis to care. Cardiopad brought remote heart tests to patients in Cameroon, yet a shortage of cardiologists still constrained diagnosis. A useful device did not, by itself, redesign the wider care system.

Analysis

This is a powerful correction to automation debates that focus only on which jobs are exposed. It shifts attention toward the design of work, incentives, and new services. Organizational research likewise finds that technology changes performance through routines and authority, not through equipment alone. Still, value creation and labor replacement are not mutually exclusive. A system redesign can improve service while also reducing demand for some roles, so evaluating the opportunity should include both the gains created and the people who bear transition costs.

Machines do not hold power; whoever sets their judgment rules does.

Prediction does not choose the goal. An AI can estimate the chance of an outcome, but people determine what action follows and how the possible outcomes are valued. Those choices may be made at the moment, or codified in advance and applied automatically. In either case, humans supply the judgment, even when the machine’s role makes the decision look impersonal.

Flint shows how judgment shifts power. Researchers predicted which homes were likely to have lead water pipes, and their predictions were correct about 80 percent of the time. Political pressure pushed the city toward digging more evenly across neighborhoods, reducing the success rate to 15 percent. A court settlement then required the city to use the predictions, and the success rate rose to about 70 percent. The model did not settle the dispute. Authority over how to use it did.

Analysis

The Flint case makes clear that technical performance and legitimate decision-making are separate questions. A prediction can improve efficiency while raising concerns about fairness, representation, and who gets to set priorities. Calling the machine powerful can obscure the officials, managers, or firms that choose its objective and thresholds. At the same time, locating human judgment does not automatically create accountability. Responsibility can be spread across developers, executives, and regulators, so systems need explicit ownership of both the rule and its consequences.

Deploy early when feedback is fast, because learning can compound into market power.

Models improve through use. AI needs initial training data, but it may also learn from feedback: information about whether its predictions proved useful or correct. When each prediction quickly produces an observable outcome, early use can create a reinforcing loop. Better predictions attract users, whose activity generates more feedback, which can improve the model further.

Search illustrates the advantage. Google sees users’ searches and clicks within seconds. That rapid feedback helps it update results across an enormous volume of activity. Microsoft invested heavily in Bing, yet the authors argue that Google’s longer-running feedback loop and related investments helped preserve its lead. A head start alone is not enough: data has diminishing returns, and markets can support different products when their users value different things. The competitive edge comes from useful learning, not simply owning the largest data pile.

Analysis

Feedback loops connect machine learning to network effects: each user can improve a service for other users, potentially concentrating a market. That makes early deployment strategically important, but it also makes data governance important. More feedback is not automatically better if it is unrepresentative, manipulated, or collected without meaningful consent. The strength of a learning advantage also depends on how quickly outcomes arrive and whether rivals can access comparable data. Competition policy should therefore examine the feedback system, not only the model’s technical quality.

Pair adaptive decisions with coordination or modularity to keep systems reliable.

A better local choice can unsettle others. The authors call this the AI bullwhip: one organization uses prediction to vary its decisions, and that variability disrupts connected decisions elsewhere. A restaurant that orders ingredients according to shifting AI forecasts may make its supplier’s demand harder to predict, which then complicates the supplier’s own orders and planning.

Two designs can contain the disruption. Coordination shares information and aligns decisions across the system, as a rowing team uses a coxswain to adjust its pace together. Modularity isolates a decision so it can change without forcing every connected process to change at once. Each trades off against the other: coordination can capture shared gains but is costly, while modularity limits disruption but can sacrifice useful synergies. Simulation and digital twins can help organizations test these arrangements before changing real operations.

Analysis

The bullwhip is a reminder that prediction changes not only what an organization knows, but also how predictable it is to its partners. This connects to supply-chain research, where small swings in retail demand can amplify upstream. Coordination is most useful when decisions have strong shared effects; modularity is attractive when information and timing are difficult to synchronize. Neither is universally superior. A digital model can reveal likely interactions, but its value depends on whether it captures human behavior and incentives as well as physical flows.

Map your mission into core decisions before buying another AI tool.

Start from the purpose. The AI Systems Discovery Canvas is a planning framework for imagining an organization with powerful prediction tools, rather than beginning with its existing rules and departments. It asks leaders to identify the mission, reduce the work to its fewest essential decisions, and specify the predictions and trade-offs each decision needs.

Home insurance reveals a new possibility. Its core decisions include acquiring customers, setting premiums, and paying claims. More precise predictions could help an insurer identify a home’s specific risks, such as fire from faulty wiring or flooding from a leak. The insurer might then recommend or subsidize prevention devices, shifting its service from paying after damage toward helping prevent it. That opportunity could require new incentives, responsibilities, and coordination across departments.

Analysis

The canvas is useful because it loosens the grip of inherited org charts and routines. It asks what the organization is trying to accomplish before debating which team owns a tool. Its simplicity is also a constraint: reducing a complex institution to a few decisions may hide informal work, political obligations, or values that resist neat measurement. The best use is as a structured starting point, followed by consultation with the people affected and explicit testing of who gains, who loses, and which risks the simplified map leaves out.

Make AI auditable, then redesign decisions so bias is detectable and fixable.

Bias can enter through the target. A health-care algorithm once used spending as a proxy for illness. Because the system spent less on Black patients, the model understated their health needs. Changing the target to health risk reduced the disparity. In another study, an AI examining knee images predicted 43 percent of racial differences in reported pain, while radiologists’ assessments explained only 9 percent, suggesting that the model detected relevant signs clinicians had missed.

Auditability creates leverage. Researchers can test an algorithm repeatedly with controlled changes and inspect how its outputs differ. That made it possible to identify and correct some forms of algorithmic bias more directly than hidden human preferences. Yet Amazon abandoned a recruiting model that learned from male-heavy historical data and downgraded women-related signals. Better tools do not guarantee fair systems; the data, objective, and deployment choices still matter.

Analysis

The strongest claim here is comparative: algorithms can be interrogated systematically, while human motives are difficult to observe. That makes measurement and correction more feasible, not fairness automatic. An audit can miss discrimination if it tests the wrong outcome or ignores unequal access to opportunity before the decision. The deeper implication is institutional: organizations need authority, incentives, and recurring review to act on audit findings. Transparency is valuable only when someone has both the power and the obligation to change the system.

Analysis

The book’s central contribution is to move AI strategy from the model to the organization. Treating AI as cheaper prediction is analytically clarifying, but prediction has economic value only when it changes a decision, and a changed decision can unsettle procedures built around predictable behavior. This sequence explains why technically impressive systems often produce modest returns: the organization may be optimized for the old information environment.

The electricity analogy gives the argument historical depth. Its force is not that AI will follow the same timetable, but that a general-purpose technology can require costly complementary redesign before productivity statistics register its contribution. The three solution types make this insight usable. Leaders can ask whether a model improves one existing step, enables a new independent action, or requires interdependent procedures to change together.

The book is especially strong where it connects this design problem to authority. Once a machine supplies a prediction, humans still decide what outcomes matter, where thresholds sit, and who may act. Therefore, AI adoption redistributes judgment and economic power. Resistance is not simply fear of new software. It may be a rational defense of roles, income, or discretion that a new system would diminish.

One limitation is that system redesign can sound like a technical optimization problem, when its objectives are political and ethical as well as operational. A system can be reliable, profitable, and still distribute risk unfairly. The discussion of bias recognizes this, but auditability alone cannot decide what fairness means. Institutions need contestable objectives, meaningful appeal, and clear responsibility for harm.

The practical lesson is not to wait for a perfect model or assume disruption is inevitable. Map the decisions, their dependencies, and their error costs. Then test whether coordination, modularity, or a new organization can make the prediction useful. The technology matters, but the system determines what it is allowed to accomplish.

Summary in 30 Seconds
AI creates value when better forecasts change the surrounding decision process, not when they merely replace one estimate. Separate likelihood from judgment about which outcomes matter. Fast feedback can improve models and strengthen market position, while isolated automation often produces little gain. Coordination limits disruption across connected operations, and modularity contains it. Fair deployment requires testing objectives, data, outputs, and responsibility for harm.
Contains spoilers

Last updated:

Report Issue

Review Summary

4.22 out of 5
Average of 7k+ ratings from Goodreads and Amazon.

Power and Prediction receives mostly positive reviews for its insightful analysis of AI's economic impact. Readers appreciate its accessible explanations, real-world examples, and thought-provoking ideas about AI's future role in decision-making and system-level changes. The book's framework of point, application, and system solutions is praised for its clarity. Some criticize repetitiveness and dated content, particularly regarding generative AI. Overall, it's considered a valuable resource for understanding AI's potential in business and society, though some find it lacking in technical depth.

Your rating:
4.56
679 ratings
Want to read the full book?

Glossary

The Between Times

Technology before widespread adoption

The period after a technology demonstrates major capability but before organizations redesign their products, workflows, and institutions to use it at scale. The authors use the term for AI’s current economic moment and compare it with the decades between electricity’s demonstration and its broad productivity effects.

Point solution

Independent process improvement

A point solution improves an existing procedure and can create value without changes to other parts of the system. In AI, it uses a prediction to improve an existing decision that can be made independently, such as improving a bank’s fraud prediction within an established transaction-approval process.

Application solution

Independent new procedure

An application solution enables a new procedure that can be adopted independently of other system changes. In AI, a prediction supports a new decision or changes how a decision is made, while the surrounding procedures can remain largely intact. Its feasibility depends on whether that new action is valuable on its own.

System solution

Interdependent process redesign

A system solution creates value by changing multiple dependent procedures together. In AI, a prediction improves existing decisions or enables new ones, but those benefits require changes to other decisions and processes. System solutions can unlock larger gains than isolated tools, while demanding more coordination and often shifting organizational power.

Hidden uncertainty

Risk concealed by routines

Uncertainty that people or organizations manage indirectly through rules, habits, and protective arrangements rather than by making informed decisions in each situation. For example, travelers may arrive early because airport delays are uncertain, while airport amenities make the resulting wait less unpleasant. Better prediction can reveal opportunities to replace such rules.

AI bullwhip

Uncertainty passed downstream

The disruption that occurs when AI improves one decision but makes that decision less predictable for others in the system. A restaurant that changes ingredient orders in response to forecasts may make demand harder for its supplier to manage, prompting further variability upstream. Coordination or modularity can help contain these effects.

AI Systems Discovery Canvas

Map decisions from mission

A framework for identifying how AI might reshape an organization. It begins with the organization’s mission, identifies the smallest set of core decisions needed to achieve it, and maps the predictions and error trade-offs associated with each decision. The canvas helps reveal dependencies and potential system-level changes that existing rules may obscure.

judgment by exception

Human review of uncertain cases

A way to combine automated predictions and human judgment. A system acts automatically when its prediction is sufficiently confident and the situation fits judgments already specified in advance. Cases with low confidence or unfamiliar circumstances are referred to a person, who decides what to do.

feedback data

Outcomes used to improve predictions

Information that connects a model’s original inputs and predictions to what actually happened. AI systems can use this evidence to update their predictions as users and conditions change. When outcomes arrive quickly, feedback data can create a competitive advantage because greater use produces faster learning.

FAQ

What's Power and Prediction: The Disruptive Economics of Artificial Intelligence about?

  • AI's Transformative Impact: The book explores how AI is revolutionizing decision-making across industries, akin to the historical impact of electricity.
  • System-Level Change: It emphasizes the need for organizations to adopt system-level innovations to fully leverage AI's capabilities.
  • Economic Disruption: The authors discuss how AI can disrupt existing power dynamics and create new opportunities for businesses.

Why should I read Power and Prediction by Ajay Agrawal?

  • Comprehensive Understanding: The book provides a deep dive into AI's transformative potential and its implications for various sectors.
  • Strategic Insights: It offers valuable strategies for integrating AI into organizational structures to stay competitive.
  • Real-World Examples: Numerous case studies make complex concepts more relatable and easier to understand.

What are the key takeaways of Power and Prediction?

  • AI as Prediction Technology: AI is primarily a tool for enhancing decision-making through improved predictions.
  • System Solutions Over Point Solutions: Organizations should focus on system-level changes rather than isolated AI applications.
  • Disruption of Power Dynamics: AI can shift traditional power structures, necessitating an understanding of these changes for future navigation.

What are the best quotes from Power and Prediction and what do they mean?

  • "AI may be to the twenty-first century what electricity was to the twentieth.": Highlights AI's potential to fundamentally change our lives.
  • "The path to large productivity increases lies in understanding what a new technology offers.": Stresses the importance of recognizing AI's unique advantages.
  • "AI only has value if it leads to better decision-making.": Emphasizes that AI's true power lies in enhancing human decision-making processes.

How does Power and Prediction define AI's role in decision-making?

  • AI as a Prediction Tool: AI improves predictions, which are crucial for effective decision-making.
  • Decoupling Prediction from Judgment: AI allows for more efficient decision-making by separating prediction from human judgment.
  • System-Level Innovations: Organizations must integrate AI into their decision-making frameworks to utilize its predictive capabilities fully.

What are point solutions versus system solutions in the context of AI according to Power and Prediction?

  • Point Solutions: Specific AI applications that enhance existing processes without major system changes.
  • System Solutions: Redesigning entire processes to fully leverage AI's capabilities, requiring broader organizational change.
  • Importance of Transition: The book argues that the real value of AI lies in developing system solutions for broader innovation.

How does Power and Prediction address the economic implications of AI?

  • Shifts in Economic Power: AI can disrupt existing power structures, creating new economic dynamics.
  • Value Creation vs. Cost Savings: Successful AI adoption should focus on creating value, not just reducing costs.
  • Feedback Loops: Continuous learning from data enhances predictions, providing a competitive advantage for early adopters.

What challenges do organizations face when implementing AI according to Power and Prediction?

  • Resistance to Change: Established rules and power dynamics can hinder AI adoption.
  • Need for System Redesign: Implementing AI often requires a complete overhaul of existing systems and processes.
  • Balancing Reliability and Flexibility: Organizations must maintain reliable operations while adapting to AI-driven processes.

How does Power and Prediction suggest organizations can overcome these challenges?

  • Adopting a System Mindset: Recognizing the interconnectedness of decisions and processes can aid AI integration.
  • Fostering a Culture of Innovation: Encouraging experimentation and innovation helps adapt to AI technologies.
  • Engaging Stakeholders: Involving key stakeholders can address concerns and facilitate smoother transitions to AI systems.

What role does feedback play in AI systems according to Power and Prediction?

  • Continuous Learning: Feedback allows AI systems to learn from past predictions and improve over time.
  • Data Collection: Implementing strategies to collect feedback data is crucial for refining AI models.
  • First-Mover Advantage: Fast feedback loops provide a competitive edge, leading to sustained success.

What is the AI Systems Discovery Canvas mentioned in Power and Prediction?

  • Framework for System Design: A tool to help organizations identify key decisions and necessary information for AI integration.
  • Three Steps: Involves articulating the mission, identifying core decisions, and specifying predictions and judgments.
  • Facilitating Change: Encourages a systematic approach to integrating AI, considering broader implications beyond individual applications.

What is the significance of the "winner-take-most" dynamic in AI according to Power and Prediction?

  • First-Mover Advantage: Early adopters of AI can establish significant advantages over competitors.
  • Feedback Loop Effects: Better predictions attract more customers, creating a self-reinforcing cycle.
  • Implications for Competition: This dynamic can lead to monopolistic behaviors, requiring strategic planning to remain competitive.

About the Author

Ajay Agrawal is a distinguished professor at the University of Toronto's Rotman School of Management, where he holds the Geoffrey Taber Chair in Entrepreneurship and Innovation and serves as Professor of Strategic Management. His expertise spans strategic management, entrepreneurship, innovation, artificial intelligence, and healthcare. Agrawal's contributions extend beyond academia; he co-founded NEXT Canada, an organization dedicated to fostering entrepreneurship and innovation. As an author, Agrawal has written influential books on artificial intelligence's economic impact, including "Power and Prediction." His work bridges theoretical concepts with practical applications, offering insights into how AI is reshaping various industries and decision-making processes.

Download PDF

To save this Power And Prediction summary for later, download the free PDF. You can print it out, or read offline at your convenience.
Download PDF
File size: 0.35 MB     Pages: 21

Download EPUB

To read this Power And Prediction summary on your e-reader device or app, download the free EPUB. The .epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.
Download EPUB
File size: 1.53 MB     Pages: 25
Want to read the full book?
Follow
Listen9 mins
Home
Swipe
Library
Get App
Try Full Access for 3 Days
Listen, bookmark, and more
Compare Features Free Pro
📖 Read Summaries
Read unlimited summaries. Free users get 3 per month
🎧 Listen to Summaries
Listen to unlimited summaries in 40 languages
—
❤️ Unlimited Bookmarks
Free users are limited to 4
—
📜 Unlimited History
Free users are limited to 4
—
📥 Unlimited Downloads
Free users are limited to 1
—
Risk-Free Timeline
Today: Get Instant Access
Listen to full summaries of 26,000+ books. That's 12,000+ hours of audio!
Day 2: Trial Reminder
We'll send you a notification that your trial is ending soon.
Day 3: Your subscription begins
You'll be charged on Nov 25,
cancel anytime before.
Consume 2.8× More Books
2.8× more books Listening Reading
Our users love us
1M+ readers
Trustpilot Rating
TrustPilot
4.6 Excellent
This site is a total game-changer. I've been flying through book summaries like never before. Highly, highly recommend.
— Dave G
Worth my money and time, and really well made. I've never seen this quality of summaries on other websites. Very helpful!
— Em
Highly recommended!! Fantastic service. Perfect for those that want a little more than a teaser but not all the intricate details of a full audio book.
— Greg M
Save 62%
Yearly
$119.88 $44.99/year/yr
$3.75/mo
Monthly
$9.99/mo
Start a 3-Day Free Trial
3 days free, then $44.99/year. Cancel anytime.
Unlock a world of fiction & nonfiction books
26,000+ books for the price of 2 books
Read any book in 10 minutes
Discover new books like Tinder
Request any book if it's not summarized
Read more books than anyone you know
#1 app for book lovers
Lifelike & immersive summaries
30-day money-back guarantee
Download summaries in EPUBs or PDFs
Cancel anytime in a few clicks

Why one summary?

Scanner
Find a barcode to scan

⭐
⭐
⭐
We have a special gift for you
Open
38% off
DISCOUNT FOR YOU
⭐
⭐
⭐
$79.99
$49.99/year
Only $4.16 per month
Continue
2 taps to start, super easy to cancel
Settings
General
Widget
Loading...