Comet sketches a ladder in her notebook. "Bottom rung, a teacher does everything by hand. Top rung, SH-1 does everything alone."
"Let's put every feature on the top rung," she says. "That saves the most time!"
Wren taps the ladder. "What does the evidence say? Check missed 3 of 10 wrong answers on our test."
Librarian Joss leans in. "The panel needs to know who decides what. Clear roles."
Nova lights up the rungs one at a time, from the bottom. "Would you like a hint?" she asks. "For each feature, ask what happens if it gets one wrong."
Comet hands you her pencil. "Reviewer, where does each feature belong?"
Human roles in AI decisions must be clearly defined.
Set-ups range from fully manual to fully automatic. AI can decide on its own, hand the decision to a human expert, or give a human decision maker a second opinion.
Some systems need human oversight and some do not.
Processes for human oversight should be defined and written down.
The crew turns this into a trust ladder for SH-1. The rungs are their own made-up example.
| Rung | Who does the work | Who has the final say |
|---|---|---|
| 1. By hand | A teacher | The teacher |
| 2. Second opinion | A person, with SH-1 adding a view | The person |
| 3. SH-1 drafts, a teacher approves | SH-1 writes it first | The teacher, before students see it |
| 4. SH-1 decides alone | SH-1 | SH-1, with no person checking |
A ladder only works if everyone knows their rung. Who reads the quiz? Who has the final say on a Check mark?
NIST says processes for human oversight should be defined and documented. So the crew writes them down.
| Statement | True or false? |
|---|---|
| Human roles in AI decisions must be clearly defined. | ? |
| Every AI system needs exactly the same amount of human oversight. | ? |
| AI can give a human decision maker a second opinion. | ? |
| Putting every feature on the top rung is the safest plan. | ? |
| Oversight processes should be written down. | ? |
Steady climbing, reviewer. Tomorrow in the Explorer Lab you write the oversight plan.