Power And Prediction
The Disruptive Economics of Artificial Intelligence
byAjay Agrawal, Joshua Gans, Avi Goldfarb
Book Edition Details
Summary
AI looms on the horizon as both a formidable disruptor and an opportunity goldmine for industries worldwide. In "Power and Prediction," acclaimed economists Ajay Agrawal, Joshua Gans, and Avi Goldfarb unveil the imminent upheaval set to redefine the business landscape. This compelling narrative traverses the twilight zone they call "the Between Times," where AI's transformative potential stands poised to revolutionize decision-making before it's fully integrated. With precision and clarity, the authors guide business leaders through the labyrinth of AI innovation, highlighting both the promise and peril of these technological advances. This essential read not only demystifies AI's economic implications but also empowers stakeholders to seize the helm amid the coming storm, ensuring they emerge as victors rather than casualties.
Introduction
Organizations across industries are grappling with a fundamental question: why does artificial intelligence, despite its impressive capabilities, remain largely confined to narrow applications while failing to deliver the transformative impact many predicted? The answer lies not in the limitations of AI technology itself, but in our misunderstanding of how prediction machines reshape entire systems of decision-making. This book presents a comprehensive economic framework that reframes AI as prediction technology and examines how it disrupts traditional organizational structures through what the authors term "the great decoupling" of prediction from human judgment. The theoretical foundation rests on distinguishing between three types of AI solutions: point solutions that improve existing decisions, application solutions that enable new decisions, and system solutions that require fundamental reorganization of interconnected decisions. The framework reveals why most organizations remain trapped in "The Between Times" - the period between witnessing AI's potential and achieving widespread adoption. The authors demonstrate how successful AI implementation often requires not merely technical integration, but systematic redesign of how organizations coordinate decisions, allocate power, and manage uncertainty. This systematic approach provides leaders with tools to anticipate disruption, design reliable AI systems, and navigate the complex transition from rules-based to prediction-enabled decision-making.
AI as Prediction Technology and Decision Systems
At its core, artificial intelligence represents an advancement in prediction technology rather than the general intelligence often portrayed in popular culture. This foundational insight reframes how we should approach AI implementation across organizations. Modern AI systems excel at taking information we have and generating information we need, whether identifying objects in images, translating languages, or forecasting market demand. The economic value of these predictions lies not in their technical sophistication, but in their ability to enhance human decision-making processes. The prediction-centric view reveals three essential components of any decision: prediction about likely outcomes, judgment about the relative value of different outcomes, and data to inform both prediction and judgment. While AI dramatically reduces the cost of prediction, it cannot replace human judgment about what matters and how much different outcomes are worth. This distinction becomes crucial when organizations attempt to implement AI systems, as they must carefully consider not just the technical capabilities of their prediction machines, but how these predictions will integrate with existing decision-making structures. Consider how Netflix transformed from a DVD-by-mail service to a streaming platform powered by recommendation algorithms. The company's success stemmed not from having the most sophisticated AI, but from redesigning its entire business model around prediction-enhanced decision-making. Instead of customers browsing physical inventories, Netflix uses AI to predict viewing preferences and automatically delivers personalized content recommendations. This transformation required fundamental changes in how the company gathered data, made programming decisions, and measured success - illustrating how prediction technology becomes valuable only when embedded within appropriately designed decision systems.
From Rules to Decisions: System-Level Transformation
Organizations rely heavily on rules and standard operating procedures to manage complexity and ensure reliability across distributed operations. These rules represent decisions made in advance to avoid the cognitive costs and coordination challenges of real-time decision-making. However, rules also embed assumptions about uncertainty and hide opportunities for AI-powered improvement. When prediction technology becomes available, organizations face the choice of whether to convert rules into decisions that can respond dynamically to new information. The transformation from rules to decisions creates what the authors call "system-level" challenges because individual decisions within organizations are interconnected. Changing how one decision gets made can have cascading effects throughout the organization, potentially disrupting carefully calibrated workflows and coordination mechanisms. Rules serve as "glue" that holds systems together by creating predictable, reliable behavior that other parts of the organization can depend upon. When AI predictions enable more responsive decision-making, this reliability may be compromised unless the entire system is redesigned to accommodate the increased variability. The COVID-19 pandemic provides a compelling example of this transformation challenge. Public health authorities initially relied on rules-based approaches like universal social distancing because they lacked sufficient information to make targeted decisions. When rapid testing technology became available to predict individual infection status, it became theoretically possible to replace the blunt rule of "everyone stays home" with more nuanced decisions about who should isolate. However, implementing this prediction-based approach required coordinating changes across multiple interconnected systems - from workplace policies and sick leave arrangements to data privacy protections and regulatory frameworks. The difficulty of achieving this coordination explains why many organizations continued following rules-based approaches even when better prediction technology became available.
Power Redistribution and the Great Decoupling
The introduction of AI prediction into organizational decision-making creates what the authors term "the great decoupling" - the separation of prediction from judgment that were previously bundled together in human decision-makers. This decoupling fundamentally alters power relationships within organizations because it changes who gets to make decisions and how those decisions are made. When prediction and judgment were inseparable, decision-making authority naturally resided with individuals who possessed both predictive expertise and good judgment. AI prediction disrupts this arrangement by making high-quality predictions available to anyone, potentially shifting decision-making power to those with superior judgment rather than predictive ability. The redistribution of power through decoupling manifests in several ways. First, it can centralize decision-making by enabling judgment to be codified once and applied at scale across many similar situations, reducing the need for local decision-makers. Second, it can democratize decision-making by making expert-level predictions available to non-experts who can then apply their own judgment. Third, it can create entirely new categories of decision-makers who specialize in translating predictions into appropriate actions for specific contexts or stakeholder groups. Consider how AI-powered credit scoring has transformed lending decisions. Traditional loan officers combined prediction about default risk with judgment about the borrower's character and circumstances. Modern AI systems can generate more accurate default predictions than most human underwriters, but someone still needs to exercise judgment about risk tolerance, regulatory compliance, and business strategy. This decoupling has shifted power away from branch-level loan officers toward central risk management teams who design automated decision rules, while also enabling new fintech companies to enter lending markets without traditional banking expertise. The transformation illustrates how AI doesn't simply automate existing roles but reconstructs the entire architecture of organizational decision-making and the power relationships embedded within it.
Designing AI Systems for the Future
Successfully implementing AI requires moving beyond point solutions toward comprehensive system design that accounts for the interconnected nature of organizational decisions. The authors present frameworks for designing reliable AI systems that can handle the uncertainty and variability introduced when rules are converted to prediction-based decisions. This design challenge is particularly complex because organizations must balance the benefits of responsive decision-making against the need for coordination and reliability across multiple interconnected processes. Two primary approaches emerge for managing this challenge: modularity and coordination. Modularity involves designing systems with clear boundaries so that AI-enhanced decisions in one area don't create unpredictable effects elsewhere. This approach allows organizations to implement AI incrementally while maintaining system stability. Coordination involves developing communication and control mechanisms that enable different parts of the organization to adapt together when AI predictions change decision patterns. Both approaches require careful attention to feedback loops, information flows, and incentive alignment. The future of AI implementation lies in what the authors call "system solutions" that reimagine entire workflows around the capabilities of prediction machines. Rather than simply inserting AI into existing processes, successful organizations will design new systems from the ground up, optimizing for the unique characteristics of prediction-enhanced decision-making. This might involve creating new organizational roles focused on judgment rather than prediction, developing real-time coordination mechanisms that can handle increased decision variability, or building entirely new business models that leverage prediction capabilities in ways that weren't previously possible. The ultimate goal is creating "oiled" systems that can adapt fluidly to new information while maintaining the reliability and coordination that complex organizations require.
Summary
The transformative potential of artificial intelligence lies not in replacing human intelligence, but in fundamentally restructuring how organizations coordinate prediction and judgment within interconnected decision systems. The path from current AI implementations to true transformation requires moving beyond point solutions toward comprehensive system redesign that accounts for the complex interplay between prediction technology, human judgment, and organizational power structures. Organizations that successfully navigate this transition will create sustainable competitive advantages by building decision-making systems optimized for a world where high-quality predictions are abundant and cheap, while those that fail to adapt risk being disrupted by competitors who better understand the systemic nature of AI-driven change.
Related Books
Download PDF & EPUB
To save this Black List summary for later, download the free PDF and EPUB. You can print it out, or read offline at your convenience.
By Ajay Agrawal
