ACT-R Cognitive Modelling

Description

Builds executable simulations of users from a validated cognitive architecture to predict task time, errors and learning on interface designs.

ACT-R is a cognitive architecture with modules for declarative memory, procedural rules, vision, motor and goals, each with parameters fitted across decades of experiments. Modelling a user means writing production rules for a task and running the model against a simulated interface; the architecture constrains timing, memory retrieval (with activation and decay), visual attention and motor movement. Predictions include task time distributions, memory errors and learning curves. CogTool made the approach accessible by generating ACT-R KLM-style models from storyboards. ACT-R is the most rigorous way to answer 'how will a skilled user perform on this design?' without testing, and to explain why.

Attributes

Loss type
Latent
Effort
Weeks

Steps

  1. Specify the task, user expertise and interface precisely, including screen layouts and timings.
  2. Use CogTool for keystroke-level prediction from storyboards, or write ACT-R productions for more complex cognition.
  3. Set architecture parameters to defaults and avoid fitting free parameters to the data you will predict.
  4. Run the model many times (e.g. 100+) to get time and error distributions.
  5. Validate against a small sample of human data; aim for predictions within about 10-20% of observed means.
  6. Use the model to compare design variants and inspect which modules (visual search, retrieval, motor) drive differences.

Sources

  • Anderson et al., An integrated theory of the mind (2004), Psychological Review
  • Byrne, ACT-R/PM and menu selection (2001), IJHCS
  • John et al., Predictive human performance modeling made easy (2004), CHI
  • Cognitive Modeling: Nine Paradigms (slide deck)

Also known as

  • ACT-R
  • ACT-R/PM
  • Cognitive architecture modelling
  • CogTool

Categories

  • User Research
  • Task Analysis & Cognitive Modelling