What Can Machine Learning Do? Workforce Implications

Artificial Intelligence and Innovation and Economic Growth

Article Snapshot

Author(s)

Erik Brynjolfsson and Tom Mitchell

Source

Science, Vol. 358, No. 6370, pp. 1530-1534, 2017

Summary

Some predict that machine learning (ML) will replace human workers. However, ML’s economic effects will be more complex, and a new framework is needed to understand how ML will affect incomes, demand, and wages.

Policy Relevance

ML will cause demand for some skills, workers, and products to rise.

Main Points

  • ML is a general-purpose technology like the steam engine and electricity, and will give rise to many related innovations.
     
  • ML will replace some human workers, but the effects on employment will be far more complex than simple replacement; in the past, new information technologies were disruptive and increased wage inequality, though also increasing living standards and incomes overall.
     
  • Labor demand will fall for tasks for which ML can substitute for a human worker, but will rise for tasks that require human workers and complement ML systems.
     
  • ML and widely available training data reduce the cost of training computers to perform tasks such as fraud detection more accurately than manually programmed systems.
     
  • Automation can be automated by apprenticing an ML system to a group of human workers for training; if the computer can access independent data to identify the optimal decision ("ground truth"), the ML system can improve human performance.
     
  • ML systems will perform well when these eight key criteria are met:
     
    • A well-defined input is linked to a well-defined output.
       
    • Many examples are available for training.
       
    • The goal is clearly defined.
       
    • No long chain of reasoning or common sense is needed.
       
    • No detailed explanation of how the decision was made is needed.
       
    • The system tolerates error.
       
    • The process learned does not change rapidly.
       
    • No dexterity is needed.
       
  • In the past, automated tasks were mostly routine, but ML will alter this trend.
     
    • ML performs some sales and customer service tasks well, despite their emotional nature.
       
    • Some creative tasks will be automated, such as device design.
       
  • The effect of an ML system on the workforce will depend on economic factors, including:
     
    • Whether the ML can substitute for a human worker.
       
    • Whether demand for a task increases when ML lowers the cost of performing the task.
       
    • Whether a task complements a task for which demand is rising because of ML.
       
    • Whether workers can easily acquire new skills that complement AI.
       

Get The Article

Find the full article online

Search for Full Article

Share