Skip to content
Nornic

Nornic Labs · Research prototype

Can AI Help Us Make Better Decisions Without Making Them for Us?

This Nornic Lab prototype explores how different forms of decision support affect understanding, trust, autonomy, and the quality of consumer choices.

Enter prototypeAbout the researchMethodology

The goal is not to maximize compliance. The goal is to test whether AI can make complexity easier to understand while keeping the human in command.

The premise is augmentation, not automation: the mind stays in charge, and any computational layer exists to make complexity visible — never to decide on your behalf.

What participation involves

Six quiet steps, one honest decision

  1. Study information and consent

    You read what the prototype does, what is recorded, and how to withdraw — before anything is stored.

  2. A few baseline questions

    Short, minimal questions. No account, no email, no name — an anonymous study identifier only.

  3. An assigned scenario

    You receive a hypothetical mobile-service usage profile and are asked to decide on its behalf.

  4. One decision

    You choose one of six service plans. Depending on random assignment, you may see no assistance at all, or one form of decision support. Any recommendation can be rejected.

  5. A short survey

    How the decision process felt — understanding, confidence, autonomy. There are no right or wrong answers.

  6. Debrief

    The session ends by explaining exactly how any recommendation you saw was computed.

The experiment at a glance

One decision, four randomized forms of support

Every participant walks the same path — consent, baseline, an assigned consumer profile, one plan decision, a short survey, and a debrief. Random assignment changes only the decision support shown during the task; every screen is otherwise identical.

Experimental flow diagram: consent, baseline, assigned profile, random assignment into four parallel conditions — human alone, opaque AI, explainable AI, and reflective AI — then a decision, post-task survey, and debrief.
Fig. 01The experimental flow. Random assignment decides which one of four forms of decision support — if any — appears during the single plan decision.

Ethical commitments

The rules this prototype holds itself to

  • 01

    You stay in command

    Every recommendation is optional and clearly rejectable. Nothing is preselected, and there is no penalty for choosing differently.

  • 02

    No dark patterns

    No urgency, no false scarcity, no hidden defaults, no social-proof nudges, no emotional exploitation — inside or outside the task.

  • 03

    Deterministic, auditable AI

    Recommendations come from a versioned, deterministic scoring engine. No large language model decides anything, and explanations match the actual calculation.

  • 04

    Data minimization

    Only what the protocol requires is collected, under an anonymous identifier. Data is versioned, exportable, and deletable on request.

  • 05

    No third-party tracking

    Study routes carry no advertising pixels, no session replay, and no marketing analytics. No personal data is sent to external AI services.

  • 06

    Honest evidence

    Synthetic engineering data is always labeled as such and never presented as a human finding. Results are reported only in aggregate.

Ten minutes. One decision. Full transparency at the end.

This prototype does not provide real purchasing advice.

Enter prototype