Interaction Design for AI
Latency and streaming affordances, error recovery, steering controls, and designing for probabilistic output.
12concepts
102flashcards
86minutes of reading
- 01 Accessibility in Streaming and Generated Interfaces Why token streaming fights the assistive-technology mechanism meant to announce it, what a runtime-generated interface does to the possibility of an accessibility review, and where AI output helps and costs disabled users their own voice.
- 02 Citations and Source Attribution in AI Interfaces Why a citation marker in a generated answer is a claim about support that is wrong a measurable fraction of the time, how citation recall and precision quantify that, and why the marker raises trust whether or not it is accurate.
- 03 Designing for Probabilistic Output Why interfaces built on the assumption of correct output fail when output is usually correct, the design moves that make errors survivable, and the cost of hiding uncertainty.
- 04 Latency, Streaming and Perceived Speed Why time to first token dominates perceived speed, how streaming changes what users tolerate, and the interaction costs that come with it.
- 05 Mixed-Initiative Interaction and When to Interrupt The expected-utility rule that decides whether a system should act on its guess, ask the user, or stay silent, why reversibility moves that threshold further than better accuracy does, and what an interruption actually costs.
- 06 Setting Expectations Before First Use Why the direction of a system's errors moves perceived accuracy more than its error rate does, what makes an error boundary learnable, and why onboarding is the cheapest intervention available and the first one to go stale.
- 07 When Chat Is the Wrong Interface Why a text box is the most expressive and least specifiable interface ever shipped, the expressiveness-versus-adjustability tradeoff that explains when to replace it with controls, and what generated UI costs in learnability.
- 08 Branching, Comparison and the Linear Transcript Why a chat transcript makes exploration destructive and comparison expensive, what the research interfaces that break linearity actually provide, and the quadratic comparison cost that stops users exploring long before the model runs out of ideas.
- 09 Feedback Collection That Is Worth Having Why thumbs-up and thumbs-down produce almost no usable signal, which implicit behaviours carry more information, and how to collect explicit feedback that can actually train something.
- 10 Interfaces for Multi-Step Agents What changes when the system acts over minutes rather than responds in seconds, how to make a long trajectory legible without demanding constant attention, and where the intervention points belong.
- 11 Reasoning Traces as an Interface Surface Why the thinking text shown next to an answer is a progress indicator and a steering handle rather than an explanation, how providers split the trace into reasoning and progress updates, and what the faithfulness results mean for anyone rendering it.
- 12 Steering, Correction and Repair What a user does when the output is nearly right, why regeneration is the wrong primary affordance, and the controls that let someone converge rather than resample.