The Leadership Loop Defined: Five Functions, Delegation Ladder and Answerability

All courses / AI Leader / Module 1 / Chapter 4

Learning objectives

After the lesson the participant can:

• Define the five leadership functions and distinguish them from the management functions M1 to M5 of AI Boss.
• Classify leadership decisions into the seven decision classes and set a ceiling for each by the four questions.
• Apply the four Level 4 prohibitions and the tests Name, Explain and Reach to a concrete decision.
• Draft a delegation profile for one leader's decision classes with answerer, independent sources and stop path.

Key concepts

The leadership loop

Five functions that run as a loop: L1 Orient, L2 Decide, L3 Organise, L4 Enable and L5 Answer. AI systems take part only through a policy point that sets a ceiling per decision class and sits outside the AI system, so a system cannot raise its own ceiling.

Seven decision classes

Strategic, resource, structural, people, risk, external and crisis. Each has a recognisable answerer and a recognisable way of going wrong with AI. A decision in more than one class takes the stricter ceiling, and a decision not listed in the profile is undelegated.

The lowest of four answers

The ceiling for a class is the lowest level allowed by four questions: reversibility, reach, who is affected and evidence. The evidence answer can only lower a ceiling the first three allow, never raise it. Lowering is immediate; raising is a named authority's decision.

The Level 4 prohibitions

Level 4 Autonomous is never used for a decision about a named person's employment, pay, discipline or access to work; for a decision law assigns to the management body, a named office or an accountable official; for a decision setting an AI system's own mandate, budget, oversight or stop path; or for an irreversible decision beyond the system's mandate.

The answerability principle

AI may inform, draft, prepare and, within a mandate, execute; it may not answer. A named leader passes Name (identifiable to those affected), Explain (can reconstruct the reasons without the system) and Reach (has the authority, time and information to change or stop the decision outside the system).

Interactive exercises

Do them here and get feedback at once. Choose a card and then where it belongs, or drag it.

Teaching plan

Discussion

1. The two reviews of automation bias differ on how far training helps, and the chapter concludes that one should rely on design. What would a design that makes a leader engage with the question before accepting the answer look like in your board or management group?
2. An organisation that names a junior approver and gives that person neither time nor sources has built a crumple zone. Where in your organisation is the named approver furthest from the time, authority and information that Reach requires?
3. Mode 0 is described as the blind spot in a leader's calendar and mail. What is reordering what your leaders see, and did anyone choose it?

Written exercise

Self-test

Choose an answer to see whether it is right, or open the answer below each question.

1. How did the Nordvik briefing incident of April 2029 fare against the three tests of the answerability principle?

   1. Name failed, Explain was passed and Reach was passed by design
   2. Name was passed, Explain failed and Reach existed only by luck
   3. All three tests were passed because the figure was corrected in time
   4. Name and Explain were passed, and Reach failed because the council voted
   Show the answer

   B. Name was passed, Explain failed and Reach existed only by luck
   The regional director signed the pack, so Name was passed. Nobody could reconstruct where the figure came from, so Explain failed. Reach existed by luck: a unit manager read a paper nobody was required to read.

2. Why does the chapter treat a second language model from another provider as a weak independent source for a leader?

   1. Because a second provider is not covered by the organisation's mandate manifest
   2. Because the AI Act forbids the use of two models on the same document
   3. Because the second summary takes longer to read than the original paper
   4. Because a large study found substantial correlation in the errors of language models, even across providers
   Show the answer

   D. Because a large study found substantial correlation in the errors of language models, even across providers
   The chapter cites a study of over 350 models in which models agreed 60 per cent of the time when both erred on one dataset. A source must not share the failure modes of the briefing; the raw paper or a line manager's report is stronger.

3. A reorganisation includes the dismissal of a unit's manager. How does the chapter say the ceiling is set?

   1. The decision is structural and people at once, and the stricter ceiling governs
   2. The decision is structural only, so the structural ceiling of Level 1 applies
   3. The decision is treated as a resource decision and may reach Level 2 Supervised
   4. The decision falls outside the classes and may be taken at any level by the board
   Show the answer

   A. The decision is structural and people at once, and the stricter ceiling governs
   A decision can belong to more than one class, and the stricter ceiling governs. The people prohibition applies to any decision about a named person's employment, whatever other class it also belongs to.