1. Research Question

If execution authority is migrating from humans to systems, what happens to the human?

2. Research Context

The preceding experiments progressively moved the research from the human into labour, technology, artificial intelligence, infrastructure, execution and authority.

Experiment 01 began with the human. I had started by trying to understand how humans operate and what conditions allow us to function well. The research then moved into labour, technology, artificial intelligence, infrastructure, execution, authority and risk.

By Experiment 08, I was examining what could happen when increasingly autonomous execution loops produced incorrect or misaligned outcomes.

At that point, I realised that there was a human question sitting underneath the entire body of research.

If systems could increasingly perform activities that humans had historically performed, including analysing information, making decisions and executing actions, where did this leave the human?

The question was no longer simply whether technology could perform human labour. The question was where the human would be positioned in relation to execution.

This brought the research unexpectedly close to where it had started.

The intellectual history of human–machine systems provided an important way into this question. Norbert Wiener’s work on cybernetics established a framework for understanding control and communication through feedback between humans, machines and their environments.

Later research into supervisory control and levels of automation examined how functions could be distributed between humans and automated systems. Parasuraman, Sheridan and Wickens demonstrated that automation could affect different stages of activity, including information acquisition, information analysis, decision selection and action implementation.

This meant that automation was not simply a question of whether a human or a machine performed a task. It was also a question of where responsibility, decision-making and execution were located within the system.

The concepts of human-in-the-loop, human-on-the-loop and human-out-of-the-loop provided another way of understanding this movement. These configurations describe different relationships between human participation and system autonomy, although terminology varies across the literature.

The research therefore began to connect directly with Execution Authority.

The question became whether the movement of the human within an automated system also represented a movement in authority.

3. Research Method

I examined established research in cybernetics, human supervisory control, automation and human–machine interaction to understand how previous researchers had conceptualised the changing position of humans within automated systems.

I then compared these established concepts with the emerging framework of Execution Authority.

The experiment focused particularly on the difference between human participation and human control. I examined what happens when a human is required to approve an action, when a human supervises an autonomous system and when a system can complete the execution cycle without human intervention.

I also examined the out-of-the-loop problem identified in human factors research. Endsley and Kiris demonstrated that removing operators from active control can reduce situation awareness and make effective intervention more difficult when automation fails.

This allowed the experiment to move beyond the simple question of whether a human remained somewhere within the system.

The research instead asked whether the human retained meaningful capacity to understand, intervene and exercise authority.

4. Research Findings

The research revealed that the relationship between humans and automated systems cannot be adequately understood through the language of replacement alone.

The human can move.

A human can remain directly inside the execution loop. A human can move into a supervisory position while the system executes autonomously. A human can move above the operational loop and establish objectives, constraints and permissions. A human can also move outside the operational execution cycle altogether.

Each position creates a different relationship between the human and execution.

The important finding was therefore not simply that machines can increasingly perform execution.

It was that the architecture determines where execution authority resides.

A consequential decision makes this visible.

A system can analyse information and recommend an action while requiring a human to approve it. In this configuration, the human remains at a critical point of execution.

The system can then be allowed to execute while the human monitors its behaviour and retains the ability to intervene. The human has moved from direct execution towards supervision.

The system can then be permitted to identify a situation, analyse information, select an action, execute that action, observe the outcome and continue operating without requiring human intervention. The human has moved further away from execution.

The underlying technology may not have changed substantially between these configurations.

The architecture of authority has.

The research also revealed that human presence cannot automatically be equated with human control.

A person could technically remain on the loop while lacking the information required to understand what was happening.

A person could possess formal authority to intervene while lacking sufficient time to prevent the action.

A person could be required to approve a decision while lacking the knowledge necessary to challenge the system that produced it.

A person could possess an override mechanism while lacking the practical capacity to exercise that override effectively.

These represent different failures of meaningful human control. A human may remain formally present while lacking information, time, knowledge or practical capability.

This distinction became central to the experiment.

I had spent months asking whether technology was becoming more capable.

Human-in-the-loop research made me realise that there was another question.

Where is the human?

That is not simply a philosophical question.

It is an architectural question.

The future relationship between humans and artificial intelligence therefore cannot simply be described as humans being replaced by machines.

Humans can move away from performing every task. They can move away from initiating every action. They can move away from supervising every transaction. They can move away from occupying every point in the execution chain.

Moving the human does not necessarily mean removing the human.

It changes what the human is responsible for.

This brought me back to Experiment 01.

I had started the research by asking how humans should operate.

I was now asking the same question inside a technological architecture in which machines could increasingly perform execution.

Perhaps the human does not need to compete with machines for execution.

Perhaps the human moves towards judgement, meaning, objectives, constraints, consent, oversight, responsibility and governance.

The central finding was therefore that the question is not simply whether humans remain in the system. The question is where we place them, what authority they retain and whether that position gives them meaningful control over what the system can do.

5. Research Implications

This finding changes the governance problem.

If execution authority is migrating, governance cannot be satisfied merely by requiring human oversight.

The architecture of that oversight matters.

It matters who can stop execution. It matters who establishes the objective that the system is pursuing. It matters who determines the constraints within which the system operates and who has the authority to change those constraints. It matters who receives information about what the system is doing and whether that information arrives quickly enough for meaningful intervention.

It also matters whether the human possesses the knowledge and practical capability required to challenge the system.

A human who can technically intervene but cannot understand the situation does not exercise the same authority as a human who can understand, decide and act.

A human who possesses an override but cannot activate it before the system executes does not possess meaningful control over that execution.

A human who is required to approve an outcome without sufficient information to challenge it may be performing an administrative function rather than exercising substantive authority.

The implication is therefore that human oversight must be evaluated through the conditions that make intervention meaningful.

This pushes Execution Authority beyond the question of whether humans are present.

It asks whether humans retain sufficient information, time, knowledge, capability and institutional authority to affect execution.

Contemporary AI governance frameworks increasingly recognise the importance of defining human roles and responsibilities in AI oversight. The NIST AI Risk Management Framework, for example, calls for organisations to establish roles, responsibilities and accountability for AI risk management and governance.

Execution Authority pushes this question further.

The relevant governance problem is not merely whether a human participates in the system.

It is whether the architecture preserves meaningful human authority as execution migrates into systems.

6. Conclusion

Experiment 09 brought the research back to the human.

The research had travelled from the human into labour, from labour into technology, from technology into artificial intelligence, from artificial intelligence into infrastructure, from infrastructure into execution and from execution into authority.

It had now returned to the human.

But the human I encountered at the end of this sequence occupied a different position from the human with whom I had started.

The central problem was no longer simply how humans operate.

The problem was how humans should be positioned inside systems increasingly capable of operating without them.

Human-in-the-loop research demonstrated that human involvement can be architecturally configured.

Execution Authority added another question.

Does the human’s position carry sufficient authority to matter?

This experiment therefore changed the question from whether humans remain inside technological systems to whether humans retain meaningful control over what those systems are capable of executing.

If putting a human somewhere in the system does not necessarily preserve human control, then the next question is unavoidable.

What would it actually take to keep humans meaningfully in control?

7. References

Endsley, M. R. and Kiris, E. O. (1995). ‘The Out-of-the-Loop Performance Problem and Level of Control in Automation’. Human Factors, 37(2), pp. 381–394.

National Institute of Standards and Technology. (2023). Artificial Intelligence Risk Management Framework (AI RMF 1.0). Gaithersburg, MD: National Institute of Standards and Technology.

Parasuraman, R., Sheridan, T. B. and Wickens, C. D. (2000). ‘A Model for Types and Levels of Human Interaction with Automation’. IEEE Transactions on Systems, Man, and Cybernetics – Part A: Systems and Humans, 30(3), pp. 286–297.

Sheridan, T. B. and Verplank, W. L. (1978). Human and Computer Control of Undersea Teleoperators. Cambridge, MA: MIT Man-Machine Systems Laboratory.

Wiener, N. (1948). Cybernetics: Or Control and Communication in the Animal and the Machine. Cambridge, MA: Technology Press.