Interface

Explainability

  • Explain the benefit, not the technology
  • Use simple, direct language to describe each explicit feedback option and its consequences
  • Optimize for understanding
  • Note special cases of absent or comprehensive explanation
  • Include explanation via interaction
  • Use example-based explanations
  • Explain what’s important
  • Tie explanations to user actions
  • In general, avoid technical or statistical jargon
  • Avoid being too specific or too general
  • Show contextually relevant information

Confidence

  • Model confidence displays
  • Decide how best to show model confidence
  • Categorical
  • N-best alternatives
  • Numeric
  • Determine if you should show confidence levels
  • When you know that confidence values correspond to result quality, you generally want to avoid showing results when confidence is low.
  • Consider changing how you present results based on different confidence thresholds
  • In general, translate confidence values into concepts that people already understand.
  • Know what your confidence values mean before you decide how to present them
  • In scenarios where people expect statistical or numerical information, display confidence values that help them interpret the results.
  • Confirm success

Expectations / Mental Models

  • Onboard in stages
  • Help users calibrate their trust
  • Introduce and set expectations for AI
  • Set expectations for AI improvements
  • Account for timing in the user journey
  • Keep track of user needs
  • Identify existing mental models
  • Clearly communicate AI limits and capabilities
  • Set expectations for adaptation
  • Describe the system or explain the output
  • Account for user expectations of human-like interaction
  • Consider using attributions to help people distinguish among results
  • Keep attributions factual and based on objective analysis
  • Help people establish realistic expectations
  • Explain how limitations can cause unsatisfactory results
  • Consider telling people when limitations are resolved
  • Demonstrate how to get the best results
  • Make clear what the system can do
  • Make clear how well the system can do what it can do
  • Make clear why the system did what it did
  • Convey the consequences of user actions
  • Notify users about changes
  • Scope services when in doubt
  • Time services based on context

Calibration

  • Avoid asking people to participate in calibration more than once
  • Make calibration quick and easy
  • Make sure people know how to perform calibration successfully
  • Let people cancel calibration at any time
  • Give people a way to update or remove information they provided during calibration
  • Always secure people's calibration information

Multiple Options

  • Give users options based on categorical / N-best alternatives
  • Consider formatting
  • Use multiple shortcuts to optimize key flows
  • Whenever possible, help people make decisions by conveying confidence in terms of actionable suggestions
  • List the most likely option first
  • In situations where attributions aren't helpful, consider ranking or ordering the results in a way that implies confidence levels
  • Consider offering multiple options when requesting explicit feedback
  • In general, avoid providing too many options
  • Prefer diverse options
  • Make options easy to distinguish and choose
  • Add iconography to an option description if it helps people understand it.