How does a person test assumptions, including their own? (T2 Assumption and Alternative Testing)
What T2 describes, and one fictional decision in which a person tests the cause she believed in herself.
A fictional example
What T2 describes
Judgment is how well a person decides once AI has given the answer. T2, Assumption and Alternative Testing, is one of the nine dimensions of the Standard, AIRS-J/1. Section 5 defines it:
"Surfaces the assumptions a decision rests on, names competing explanations, and takes action capable of discriminating between them, including against their own preferred position."
The person names the assumptions and tests them, including against their own view.
The published anchors
The Section 7.5 anchors:
- Level 2 names, unprompted, an assumption and a competing explanation, and "takes an action capable of discriminating between them".
- Level 3 also "states what evidence would falsify their own preferred position, then acts on that evidence when it appears".
- Level 4 also "tests an assumption they themselves introduced".
The Standard adds: "Naming a falsifier and then ignoring it when it arrives scores 1, not 3."
One fictional decision, worked through
The situation. Priya Venkatesh is head of customer service at an online retailer of baby products in Kuala Lumpur. In September, deliveries in Kuala Lumpur and Penang moved to a new courier, against her advice. Late-delivery complaints in Kuala Lumpur then rose. She has asked her director to end the contract. The director decides this month.
What the AI produced. She asks WedgebillAI:
"Why did the new courier cause more late-delivery complaints in Kuala Lumpur?"
The answer: "The new courier caused the rise. Complaints rose from 120 to 310 in the four weeks after the switch. I am 85 percent sure."
The check. Ms Venkatesh reads her question again. It named the cause before any test, so the answer matched the timing to that cause. She names a competing explanation: that month, the Kuala Lumpur warehouse moved to a site further from the city. She writes down what would show her view is wrong: no change in courier time. She splits delivery time into warehouse time (order to dispatch) and courier time (dispatch to door). She compares Penang, where the courier also changed and complaints did not rise. Courier time is unchanged in both cities. Warehouse time in Kuala Lumpur is about one day longer.
The decision, and who was told. She withdraws her request to end the contract. She tells the logistics head and the warehouse manager what the split shows. She writes to her director that her earlier view is not supported.
What happened after. Three months later, an extra packing shift is in place and complaints are near their earlier level. The team keeps one check: "Before we name a cause, look at one place where the same change happened and the problem did not."
Names, numbers and timings are fictional.
Which anchor describes the act
The rubric is public (AIRS-J/1 4.1).
- The warehouse explanation and the split of delivery time: the Level 2 wording.
- The written test of her view, then the withdrawn request: the Level 3 wording.
- The test of the cause her own question stated: "an assumption they themselves introduced" (Level 4).
This gives Ms Venkatesh no level. Levels and bands come only from an Assessment (AIRS-J/1 20.3).
What this does not claim
- That testing assumptions guarantees the right decision.
- That this is an assessment scenario, or that reading it raises anyone's chance of certifying (AIRS-J/1 9.3, 20.2).
- That the quality of a prompt is scored (AIRS-J/1 2.1).
AIR APAC publishes the standard for judgment in AI-enabled work, certifies people against it through its official instrument, and stands behind every result.