The Impact of Artificial Intelligence on Rules, Standards, and Judicial Discretion

Artificial Intelligence and Innovation and Economic Growth

Article Snapshot


Frank Fagan and Saul Levmore


Southern California Law Review, Vol. 93, No. 1, pp. 1, 2019


Artificial intelligence (AI) systems will be used by lawmakers and judges to refine and apply the law. Over time, flexible standards will tend to be displaced by precise rules.

Policy Relevance

Lawmakers should focus on data collection. Lawmakers should think carefully about the goals of AI systems.

Main Points

  • Machines will make decisions about pre-trial bail, parole, prison sentences, speed limits, visas, and tariffs; humans will determine the overall goals of AI systems (such as efficiency or ameliorating climate change) and will sometimes override AI decisions.
  • The best lawmakers and judges will be adept at deciding when to override decisions reached by AI.
    • If the AI is 52 percent confident in its decision, and the human partner is equally sure the AI is wrong, the human should not overrule the AI.
    • As long as the law is reasonably stable, if the human is confident that the AI system has been given an appropriate goal, the AI should be trusted.
  • The accuracy of AI systems depends on whether the pattern the AI has identified exists in the real world; likely causes of machine learning error in legal settings include:
    • The scarcity of data;
    • Constant change in the social and legal context (the best predictors of successful release on bail in 1995 might be different from the most important variables in 2019).
  • The effect of error is more severe in law than in medicine, as one legal remedy often precludes another, but this is rarely true of medical remedies.
    • In law, rulings often assign exclusive rights, creating winners and losers.
    • If one medical treatment fails, patients are usually no worse off, and can often try another.
  • AI will be less likely than humans to be fooled by missing variables, overlooked connections, and statistical errors; AI can handle big data, and algorithms developed on one portion of the data can be tested on another portion.
  • The usefulness of AI systems in lawmaking depends on the collection of data.
    • Lawmakers can mandate that data be structured to facilitate collection.
    • Driverless cars could be networked so that patterns of careless and harmful behavior could be identified easily.
  • Overall, the use of AI in the legal system will lead to development of more detailed and precise rules.
  • Rarely, data will support migration from precise rules to more flexible standards.
    • Analysis of data might fail to reveal useful correlations.
    • Policymakers tried to use precise sentencing rules to reduce racial disparities in sentencing, but the law was altered to return flexibility to judges when the effort failed.

Get The Article

Find the full article online

Search for Full Article