Prediction, Judgment, and Complexity: A Theory of Decision Making and Artificial Intelligence

Innovation and Economic Growth

Article Snapshot

Author(s)

Ajay Agrawal, Joshua Gans and Avi Goldfarb

Source

in The Economics of Artificial Intelligence: An Agenda, Ajay K. Agrawal, Joshua Gans, and Avi Goldfarb, eds., University of Chicago Press, 2019 (forthcoming)

Summary

Artificial intelligence (AI) raises the possibility that machines will substitute for humans. AI improves the accuracy of predictions, but when a prediction cannot be made with absolute certainty, judgment is needed to choose the best course of action. Judgment can sometimes be automated.

Policy Relevance

The affects of AI on employment are uncertain. AI will affect contracting behavior and the boundaries between firms.

Main Points

  • Formerly, machines enhanced human productivity, as the human performed the mental aspect of a task while a machine performed the physical aspect; AI blurs the line between physical and mental tasks.
     
  • Advances in AI improve the accuracy of predictions; however, predictions differ from exercising judgment, which requires comparing the payoffs from different actions.
     
    • If a bank is uncertain whether a credit card charge is fraudulent, the bank compares the payoff of refusing a legitimate charge to the payoff from approving a fraudulent charge.
       
    • Judgment is the process of determining these payoffs.
       
  • Some suggest that radiologists have no future because machines can now diagnose diseases and bone fractures; the future of radiology will depend on the need for human judgment in deciding what to do when a diagnosis is uncertain.
     
  • When no predictions can be made, decision-makers must use judgment.
     
    • In simple cases, one choice dominates; in complex cases, one choice rarely dominates.
       
    • In a complex case when no predictions can be made, decision-makers cannot use judgment and will choose the action that gives the best result on average.
       
  • In complex cases, sometimes AI will make predictions with absolute certainty and no judgment is needed, but at other times AI will make predictions with less certainty; if judgment is needed, judgment generates more benefits as predictions gain in accuracy.
     
  • Some tasks that require judgment (such as switching railway trains) can be automated; as automation proceeds, machines can substitute for humans.
     
  • In complex situations without prediction, some assume that humans perform better than machines, but human decision-makers tend to choose the action that gives the best result on average, a course of action that is not obviously advantageous and which can be automated.
     
  • It is uncertain when and where AI will substitute for human labor; when judgment is needed, more humans might be employed, but AI will increase opportunities to study judgment to discover when it can be automated.
     
  • AI will reduce the “incompleteness” of contracts by enabling the parties negotiating a contract to anticipate many contingencies; employment contracts are an exception because it is not usually possible to describe all the tasks that will be required of an employee.
     
  • AI will affect the boundaries between firms.
     
    • If complex decisions such as the scheduling of airline flights can be automated, more vertical integration is likely to result.
       
    • Vertical integration will be costly if complex decisions cannot be automated because of the high costs of judgment.
       

 

Get The Article

Find the full article online

Search for Full Article

Share