Algorithm aversion: people erroneously avoid algorithms after seeing them err.

Area of Science:

• Decision-making
• Behavioral economics
• Forecasting science

Background:

• Evidence-based algorithms demonstrate superior predictive accuracy compared to human forecasters.
• Despite proven accuracy, human forecasters are frequently preferred over statistical algorithms.
• This preference, termed algorithm aversion, carries significant economic costs.

Purpose of the Study:

• To investigate the causes of algorithm aversion in forecasting decisions.
• To examine how observing algorithm performance influences trust and selection.
• To understand why individuals prefer human forecasters even when algorithms are superior.

Main Methods:

• Five studies were conducted involving participant observation of algorithm and/or human forecasting performance.
• Participants were presented with scenarios involving algorithmic or human predictions.
• Decision-making tasks assessed participants' confidence and willingness to rely on algorithmic versus human forecasts.

Main Results:

• Participants who observed algorithmic forecasters exhibited reduced confidence and were less likely to select them.
• This aversion persisted even when algorithms outperformed human forecasters.
• Mistakes made by algorithms led to a quicker decline in trust compared to similar human errors.

Conclusions:

• Observing performance, particularly errors, exacerbates algorithm aversion.
• The tendency to lose confidence in algorithms after errors drives the preference for human forecasters.
• Understanding algorithm aversion is crucial for optimizing decision-making processes that involve predictive tools.