Prediction Machines: The Simple Economics of Artificial Intelligence

Artificial Intelligence, Innovation and Economic Growth and Privacy and Security

Article Snapshot


Ajay Agrawal, Joshua Gans and Avi Goldfarb


Harvard Business Review Press, 2018


Advances in artificial intelligence (AI) now offer improved predictions, which will reduce uncertainty and lead to the redesign of business strategies. AI will affect jobs, the concentration of corporate power, privacy, and politics around the world.

Policy Relevance

AI will require tradeoffs; more innovation might mean less competition. Demand for data might mean less privacy.

Main Points

  • The price of advanced AI systems is now falling sharply, which means that it will be deployed everywhere.
  • AI does not give computers the power to think; rather, new AI technologies will make predictions cheaper and more accurate, changing decision-making throughout the economy.
  • Prediction, the power to fill in missing information, will reshape economic activity in two ways.
    • Prediction machines reduce the cost of implementing any strategy by replacing humans that preform predictive tasks (such as inventory analysis).
    • Accurate and reliable prediction machines will change firms' choice of strategies.
  • Small improvements in the accuracy of predictions are significant; an improvement from 85 to 90 percent accuracy reduces mistakes by one-third, but an improvement from 98 percent to 99.9 percent reduces mistakes by a factor of twenty.
  • The power of improved predictions can seem magical, bringing us machines that can see (object recognition), navigate (driverless cars), and translate.
  • Machines are limited in their ability to predict human judgments when data is lacking, such as information about individual preferences or rare events; by contrast, managers must make decisions about innovation, mergers, and partnerships without much data about similar events.
  • Businesses are influenced by uncertainty in deciding whether to supply a product or service themselves, or to rely on another firm to provide it; by reducing uncertainly, AI will influence the boundaries of firms.
  • AI increases certain types of risks, including the risk of discrimination, vulnerability to hackers, the risks of massive system failure, intellectual property theft, and manipulation.
  • AI systems present certain tradeoffs:
    • AI will enhance productivity but could exacerbate inequality by lowering wages.
    • Like other software, AI has economies of scale, and might lead to monopolization.
    • AI systems perform better with more data; demand for data might create data markets in which consumers can choose to sell their data.


Get The Article

Find the full article online

Search for Full Article