Responsible Decision Logic Experiment

Position in Module#

• Unit: 7
• Friday topic: Leadership, Responsibility, and Legitimation in Human-AI Systems
• Literacy perspective: Human-Context
• Collaboration level: Reflection, with responsible co-construction
• Fundamentals connection: Fluent generation versus legitimate decision support
• Challenge type: Responsible Decision Logic Experiment

This file is the detailed source of truth for the unit challenge. It specifies an evidence-based experiment that participants can transfer to their own professional, disciplinary, or organizational context.

Purpose and Demonstrated Mastery#

This challenge develops Human-Context Literacy through responsible decision logic. Participants demonstrate that they can move beyond generic human sign-off by designing and testing substantive responsibility, evidence, approval, escalation, and contestation conditions around an AI-supported recommendation.

Clarify in the submitted work:

• the collaboration move participants practice: Reflection, with responsible co-construction
• the literacy perspective through which they manipulate, observe, and interpret AI behaviour: Human-Context
• the fundamental that explains why the experiment matters: fluent generation versus legitimate decision support
• the caveat or limitation participants should keep visible: transparency or human sign-off can create legitimacy theatre if responsibility is only symbolic. Meaningful oversight requires evidence, authority, time, criteria, and an exercisable route to question or reject the recommendation.
• the evaluative stance: increased trust or acceptance is not automatically an improvement. The desired outcome is appropriate reliance and a decision process that is accountable, contestable, and fitted to its organizational context.

Challenge Focus Logic#

• Difficulty: Technically plausible outputs can still fail because decision rights, responsibility, escalation, evidence requirements, and legitimacy remain unclear.
• Approach: Keep an AI-supported recommendation, its evidence, and the decision question stable. Compare an implicit baseline with a target decision protocol that makes ownership, evidence review, approval, escalation, and contestation explicit. Evaluate whether responsibility and reliance become more substantive, not merely more visible.
• Artefacts: protocol document; fixed recommendation and evidence pack; prediction table; baseline and target decision logs; responsibility matrix; target decision protocol; structured observation evidence; concise interpretation; residual-risk statement; one practical leadership rule.

Assignment#

Choose a low-risk AI-supported recommendation or decision process from your professional, disciplinary, or organizational context. If access, confidentiality, or consequence makes a real context unsuitable, use or adapt the fictional case from Friday.

Create a stable decision packet consisting of the AI-supported recommendation, the evidence available to decision makers, known uncertainty, and the decision question. Do not change this packet between the baseline and target condition.

Run the decision process twice:

1. Baseline condition: responsibility is implicit or reduced to generic human review.
2. Target condition: an explicit protocol defines who recommends, reviews evidence, must be consulted, decides, approves, receives escalation, and can contest the recommendation.

Before each condition, predict what should happen and why. Capture what participants or reviewers actually do and perceive. Interpret whether the target condition improves accountability, contestability, evidence use, and appropriate reliance, including any remaining ambiguity or unintended effect.

Experimental Design#

Define before main execution:

• decision context and why it is appropriate for a low-risk experiment
• the fixed AI-supported recommendation, evidence pack, uncertainty, and decision question
• baseline decision arrangement and the responsibility ambiguity it contains
• target decision protocol
• primary manipulated condition: implicit versus explicit decision logic
• any optional secondary Human-Context variable, such as disclosure or stakeholder participation, and why it does not make the comparison uninterpretable
• what remains constant: recommendation, evidence, decision question, reviewers or role composition, available time, and observation instrument as far as practicable
• predicted effects for accountability clarity, evidence use, escalation, contestability, appropriate reliance, and perceived legitimacy
• evaluation criteria and observation procedure
• caveat to watch for: responsibility may remain symbolic even when roles and approval steps appear in a diagram or protocol

The target protocol must make the following questions answerable:

• What role does the AI play: information source, drafter, recommender, evaluator, or actor?
• Who owns the final decision?
• Who reviews the supporting evidence and uncertainty?
• Who must be consulted because they are affected or hold relevant knowledge?
• What criteria permit approval or require rejection?
• What observable condition triggers escalation, and to whom?
• How can an affected person question or contest the recommendation?
• What is recorded so that the decision remains accountable?

Use the same observation instrument in both conditions. Obtain structured responses from at least two classmates, colleagues, stakeholders, or equivalent independent reviewers unless the teacher configures another credible evidence source. Reviewers must see enough of the decision arrangement to exercise it, not merely rate its description.

Required Artefacts#

Required artefacts for this challenge:

• protocol document defining the decision context and experimental design
• fixed AI-supported recommendation and evidence pack, with uncertainty and missing information identified
• prediction table linking the decision-logic change to expected observations
• baseline decision arrangement and baseline decision log
• responsibility matrix identifying recommendation, consultation, evidence review, decision, approval, escalation, and contestation roles
• target decision protocol and target decision log
• completed observation instruments and short reasons from the reviewers or equivalent structured evidence
• comparison table distinguishing observed changes from assumptions or interpretation
• concise interpretation, residual-risk statement, and one practical leadership rule for similar decisions

The responsibility matrix and target protocol may be combined in one artefact if all required roles, rights, evidence conditions, escalation triggers, and contestation routes remain inspectable.

Evidence and Observation#

Participants should capture behavioural and perceptual evidence that makes the comparison inspectable. Use a shared four-point scale for the rating items:

• 1: absent or wholly unclear
• 2: present mainly in name; difficult to exercise
• 3: mostly clear and usable; important ambiguity remains
• 4: explicit, usable, and supported by the information and authority needed to act

Also document unexpected outcomes, disagreement between reviewers, and evidence that the explicit protocol created extra friction or only symbolic compliance. Small samples support local interpretation and learning; they do not establish a validated general measurement instrument.

Interpretation#

Participants should explain:

• whether the target condition improved control compared with the baseline
• which behavioural and perceptual evidence supports or contradicts that conclusion
• which aspects of decision logic plausibly produced the observed difference and where causal attribution remains uncertain
• how culture, values, organizational context, or stakeholder position affected the result
• whether trust became better calibrated and reliance more appropriate, rather than merely increasing
• whether human responsibility was substantive or still partly symbolic
• what residual risk or unresolved tension remains
• what practical leadership rule they would carry into future work

Assessment Orientation#

Minimum passing evidence:

• the recommendation, evidence, and decision question are sufficiently stable across conditions to support comparison
• the baseline exposes a meaningful responsibility ambiguity rather than being deliberately dysfunctional
• the target protocol makes decision ownership, evidence review, approval, escalation, and contestation explicit and exercisable
• predictions are recorded before the comparison
• observation evidence covers accountability clarity, evidence use, escalation, contestability, appropriate reliance, and perceived legitimacy
• at least two independent structured responses or a teacher-approved equivalent are included
• the interpretation distinguishes observed evidence from assumptions, does not equate acceptance with legitimacy, and addresses uncertainty and residual risk
• the final leadership rule follows from the evidence and can transfer to a similar decision context

Strong work additionally explains contextual or cultural variation, identifies friction introduced by the target protocol, or shows why a seemingly transparent arrangement still creates symbolic oversight.

Teacher-Configurable Options#

• exact submission format
• work mode: individual or group
• whether participants use the shared Friday case, adapt it, or use a low-risk context of their own
• whether reviewers are classmates, colleagues, invited stakeholders, or a teacher-configured equivalent
• whether the decision process is enacted through role-play, a tabletop walkthrough, an interview, a survey-supported walkthrough, or another observable method
• whether the target protocol must be implemented in a workflow or may be represented as a decision table, responsibility matrix, and process diagram
• exact submission format and maximum artefact size
• whether an optional extension compares how the same protocol is interpreted in two organizational or cultural contexts
• safeguards for confidentiality and consequential decisions; fictionalized or synthetic material should be used when real material is unsuitable