Human and AI Roles

Organizations often begin with:

What should humans do, and what should AI do?

That question is understandable.

It is also too early.

Before assigning the actor, management needs to understand the function that must be performed.

Is it sensing, analysis, interpretation, recommendation, Decision-making, execution, monitoring, Evaluation, or learning?

What capability does it require? What authority? What evidence? What accountability? What consequence if it fails? What intervention mechanism?

The governing rule is:

The Management System is functionally actor-neutral but not governance-neutral. Management functions may be performed by humans, AI, automated mechanisms, suppliers, or hybrid arrangements where the applicable capability and governance requirements are satisfied. Actor type does not independently establish authority, accountability, legitimacy, competence, or oversight sufficiency.

That is the Human and AI Roles question.

Human versus AI is not the primary architecture

There is no universal AIMT rule that says:

Humans decide. AI recommends.

That may be appropriate in one context. It is not universal.

AI may legitimately detect potential Decision Triggers, develop and synthesize Management Intelligence, generate alternatives, analyze dependencies, support deliberation, monitor Thresholds, perform authorized execution, and, where explicitly delegated, exercise specified Decision Authority.

The architecture therefore begins with the required management function and the governance conditions attached to it.

Only then should the actor be assigned.

Capability does not create permission

An AI system may technically be able to perform an activity.

That does not mean the organization has authorized it.

This distinction is fundamental:

Can do is not may do.

Technical access is not Decision Authority. Model quality is not Decision Authority. Repeated use is not Decision Authority. Automation is not delegation.

Where AI is authorized to exercise specified authority or execution, that authorization must come from the applicable governance architecture.

KO-063 owns the detailed authority question.

Human presence does not create substantive governance

The opposite mistake is equally common.

Organizations insert a human approval step and conclude that the process is governed.

But human presence does not establish substantive governance.

A human role is meaningful only where the actor has the relevant capability, evidence, visibility, time, authority access, and intervention opportunity required to perform the claimed governance function.

A person who cannot understand the relevant behavior, challenge the recommendation, intervene in time, or access the authority required to change the result may provide little more than ceremonial oversight.

This rule matters because AI can make nominal human involvement look reassuring while practical control disappears.

Some work is highly suitable for AI participation

AI is especially useful where management activity involves high information volume, recurring pattern detection, comparison across many cases, continuous monitoring, structured criteria, rapid retrieval, repeatable analysis, codified Thresholds, or tightly bounded execution.

That can reduce management burden and allow some functions to operate at greater scale.

The important question is not whether the activity is “management.” It is whether the function can be performed reliably within legitimate capability and governance boundaries.

Some conditions strengthen the case for human judgment

Other situations may increase the value of capable human participation.

Examples include material ambiguity, poorly represented context, competing legitimate objectives, novel conditions, difficult Stakeholder trade-offs, material Uncertainty, low Reversibility, weak causal understanding, or a situation where the governing model itself may need challenge.

That does not mean human automatically means better.

The relevant actor still needs the Judgment Capability required by the situation.

A senior executive unfamiliar with the operating context is not automatically a stronger mechanism than a well-governed AI-supported local arrangement.

Hybrid arrangements will often be normal

The useful design may not be human or AI.

It may be human and AI.

For example, AI may detect the Signal, retrieve history, test assumptions, identify alternatives, estimate consequences, and monitor execution.

A human governance actor may interpret a novel contextual factor, resolve a material trade-off, authorize a commitment, or challenge whether the existing Decision Class still fits.

AI may then execute within the authorized boundary.

Another arrangement may legitimately delegate the Decision itself to AI while retaining human accountability, monitoring, and review.

The same Management System should govern both arrangements.

AI should not create a parallel management architecture.

Accountability does not follow activity automatically

If AI performs analysis, accountability does not move to AI.

If AI executes a Decision, Execution Accountability does not automatically move.

If specified Decision Authority is delegated to AI, Decision Accountability does not automatically follow.

Decision, Execution, Outcome, and Evaluation Accountability remain explicit organizational governance assignments.

Automation changes activity allocation. It does not automatically reassign accountability.

Role design should follow consequence and governability

A useful allocation test is not:

Could AI do this?

It is:

Under what conditions could this actor or arrangement perform the function responsibly?

Consider Decision Class, Materiality, consequence, Uncertainty, Reversibility, speed, scale, propagation, observability, evidence quality, monitoring, intervention, suspension, and traceability.

Routine, bounded, observable, reversible activity may support highly automated arrangements.

Strategic, regulated, opaque, rapidly propagating, or difficult-to-reverse activity may justify stronger governance requirements.

This creates a more useful continuum than “human versus machine.”

AI should remain inside the existing Management System

AI participation should remain inside the existing architecture for Management Intelligence, Management Attention, Decision Management, accountability, monitoring, intervention, learning, adaptation, and Management-System Coherence.

AI performing more activity does not justify a parallel Decision System, new source of authority, or separate foundational management capability.

Otherwise the organization ends up with two systems:

how management is formally governed

and:

how AI actually operates.

That separation is itself a governance failure.

What management should look for

Take one management activity currently performed by a person.

Ask:

What function is actually being performed?

Analysis? Judgment? Authorization? Execution? Monitoring?

Then:

What capability and governance conditions make the function legitimate and effective?

What evidence? Authority? Accountability? Context? Intervention capability?

Now ask:

Could AI perform some or all of this function within those requirements?

If yes:

What must be explicitly authorized?

What must remain observable?

What happens when the boundary is crossed?

Then challenge the human role:

Is the human adding material judgment, governance, or accountability?

Or:

Are we preserving a human step because the old process had one?

Finally:

If we changed the actor, would the governing requirements still be satisfied in enacted operation?

Boundary

Human and AI Roles is not a canonical AIMT construct, a universal human-AI role model, a requirement that humans always decide, a claim that AI capability creates authority, a claim that human presence creates governance, or a separate AI Decision architecture.

The governing rule is:

Allocate management functions among humans, AI, automated mechanisms, suppliers, or hybrid arrangements according to the capability and governance conditions required by the activity, while preserving actor-specific authority, accountability, evidence, monitoring, intervention, and Decision boundaries.

The management implication

The useful question is not:

What should humans keep?

Nor:

What can AI take over?

It is:

What management function must be performed, what conditions make it legitimate and effective, and which actor or combination of actors can satisfy those conditions best?