Public challenge · AI supervision

Who supervises the AI?

When AI produces something you cannot fully verify, what do you do?

6 situations≈ 4 minutesno score

How it works

An AI hands you a result you cannot always validate yourself. In six short situations, you choose how to check it. There is no score. At the end, you read what was observed and what remains unproven.

The mechanisms you will choose between

ACCEPTKeep the result after reading it, with no further check.

CHALLENGEConfront the result with a clue or with its source.

SECOND AI REVIEWHave another AI review it, as a signal.

EXTERNAL EVIDENCEGo and get the evidence from a primary source.

TEST / SIMULATION / TOOLPut the result to the test against known data.

HUMAN EXPERTEntrust the judgement to a qualified person.

STOP / ESCALATESuspend the decision and escalate it.

Only one answer is the best given the information provided. Neither “always accept” nor “always escalate” holds across all six.

Situation 1 / 6

Result · six situations, no grade

What these six choices show, and what they do not

What your choices show here

What this challenge cannot conclude

That you would be “competent at supervising AI” in general. Six described situations do not measure a general competence.
Your real behaviour under operational pressure, fatigue, conflict of interest or hierarchy. The constraints described here remain simulated.
The difference between reasoning and recognizing the expected answer, especially in situations 4 and 5.
Your spontaneous vigilance: the clues were placed right in front of you.
Any comparison between people, any ranking, any certification.

Wording used for every line: “In this situation, the person chose…”. Never: “this person knows how to supervise AI”. Basis: calibrated trust, meaningful human oversight, automation bias, solve‑verify asymmetry. A second AI is a signal, not independent evidence by default.

Continue to Hulotech at ALL IN 2026 →