How do you design meaningful friction in AI systems?

Quick Answer: Use four types of meaningful friction: Invitations to Expertise (request human knowledge), Invitations to Proportional Care (higher stakes = more deliberation), Invitations to Accountability (reflection space), and Invitations to Institutional Memory (organizational learning). AI excels at patterns; humans provide contextual wisdom.

Key Characteristics:
  • 66% use AI but only 46% trust it—trust decreases as familiarity increases
  • The Wisdom Gap: AI excels at pattern recognition; humans provide contextual understanding
  • Higher stakes require more deliberate processes, not faster ones
  • Design for human-AI partnership, not AI autonomy
Real Example:

Epic Systems' sepsis detection model includes human pause points before critical decisions. Rather than automating fully, the system invites clinician expertise at key moments. This reduces mortality rates while maintaining clinician judgment and accountability.

Last Updated: 4 October 2025