Levels of Automation Analysis

Description

Rates automation at each information-processing stage (acquisition, analysis, decision, action) on a 10-point scale to evaluate workload, trust and out-of-the-loop risk.

Sheridan and Verplank's ten-level scale runs from the computer offering no assistance to the computer acting entirely autonomously and ignoring the human. Parasuraman, Sheridan and Wickens added four stages: information acquisition, information analysis, decision selection and action implementation, each of which can be automated to a different level. The analysis profiles a system as a vector of levels and evaluates primary consequences (workload, situation awareness, complacency, skill degradation) and secondary ones (reliability, cost of errors). It gives designers a precise vocabulary instead of 'AI-powered', and predicts known pathologies: high automation at decision stages with imperfect reliability produces automation bias.

Attributes

Loss type
Latent
Effort
Days

Steps

  1. Decompose the feature into the four stages: acquisition, analysis, decision, action.
  2. Assign the current and proposed level of automation (1-10) for each stage.
  3. Estimate automation reliability at each stage from model evaluations or logs.
  4. Evaluate human performance consequences: workload, situation awareness, complacency and skill loss; flag high decision automation with reliability below roughly 0.9.
  5. Prototype alternative profiles and test with users using detection of automation errors as a key measure.
  6. Choose the profile that balances efficiency with error detection, and document override and transparency mechanisms.

Sources

  • Sheridan & Verplank, Human and computer control of undersea teleoperators (1978), MIT
  • Parasuraman, Sheridan & Wickens, A model for types and levels of human interaction with automation (2000), IEEE Transactions on SMC-A

Also known as

  • LOA
  • Sheridan-Verplank scale
  • Types and levels of automation

Categories

  • Accessibility
  • Human Factors & Error