Design under constraints~45 min

Design Under Constraints: Design the Wrong Answer

Early preview draft — not yet practitioner or accessibility reviewed.
Goal
Practice the interface work probabilistic features demand: an AI feature that is sometimes wrong is WORKING AS DESIGNED — so the wrongness needs designing too.
Scenario
An app you know adds an AI feature (summarize this thread, tag these photos, suggest a reply — pick one). Design not the happy path but the other four: low confidence, flat-out wrong, slow, and unavailable.
Audience
An engineer who needs to know exactly what to build for each state, and a user who must never be silently misled.
Constraints
Four states minimum: uncertain (how does it communicate less-than-confident?), wrong (how does a user notice, correct, and recover?), slow (what happens at 10 seconds?), unavailable (what's the manual path?). The happy path may appear only as context. Two responsibilities run through all four: privacy (say what happens to the user's data and corrections — collected, retained, or deliberately not) and accountability (a human can always tell what the system did on their behalf and undo it). Work from a real product you use — never feed anyone's confidential material into an AI tool as part of this lab.
Deliverable
The four states, sketched or described concretely, plus for each: how the user discovers the state, what they can do about it, and what the system learns (or deliberately doesn't) from the correction. One line on which state you'd ship most carefully and why.
Counts on your evidence profile as
A completed Skill Lab with its rubric and your deliverable
Accessibility requirement
Every state's signal must work without color alone and be announced to assistive tech — an uncertainty indicator only sighted users perceive fails the state it exists for.
AI policy
AI use is allowed for this lab. Using an AI tool to explore how competitors handle these states is fine; the four state designs are the deliverable.
Rubric
  • Uncertainty is legible, not decorative: The uncertain state changes what the user sees or must confirm — not just a "results may vary" disclaimer nobody reads.
  • Correction is a first-class path: A user can fix a wrong output in the flow itself, and the design says where the correction goes — including if the answer is "nowhere, deliberately."
  • There is always a way through: The unavailable state leaves the task completable manually — the AI failing never strands the user.
  • Data and accountability are answered: The design states what user data each state collects or retains (or deliberately doesn't) and how a person can see and undo what the system did — reasoning about responsibility, not interface polish.
Resources

Skills: Problem Framing, Interaction Design, Accessibility

Relevant to: Product Designer, UX Designer, Interaction Designer

Your work

Sign in to start this lab.