Supervised Ethical AI

2026-10-30

Introducing the concept of maximised, integrated cognitive ergonomics between a human Supervisor (legal entity) and AI

By Dr Andrew Turk, Fremantle Greens

Introduction

It has now been recognised by almost all legitimate commentators and organisational bodies that un-constrained development of AI constitutes a real and urgent threat to humanity. Only by thorough, honest discussion of the relevant issues and potential solutions can humans manage this situation. Australia’s PM Albanese assisted leaders of other ‘Middle Powers’ to come to an agreement about increasing AI accountability. This article is a contribution to this debate, from the perspective of ethics and Human Computer Interaction (HCI). 

Over the millennia, human beings have struggled with tensions between ethics and doing what you consider most profitable for yourself, or your family, ethnic or religious group or nation and perhaps even for human-kind in general. The Oxford Dictionary defines the adjective ethical as:

  • Relating to moral principles or the branch of knowledge dealing with these: (e.g. ethical issues in nursing are implemented via ethical standards). 
  • What is morally good or correct: (e.g. can a profitable business ever be ethical?). 
  • Avoiding activities or organizations that do harm to people or the environment: (e.g. an ethical expert persuading clients to switch to more ethical products or to adopt ethical shopping habits or take ethical holidays). 

Throughout human history in different ways, in all cultures, children and adolescents/young people have been taught the local body of social laws and conventions and how to use these and their other skills in an ethical manner. For instance, for more than 65,000 years each Australian Aboriginal Language Group has developed such a system reinforced with ceremonies at each stage of a person’s development in understanding. It is called Jukurrpa in the Western Desert of WA. This system of personal responsibility and community management has different content, and its own name, for each Aboriginal language community, living within the resources and constraints on its own ‘country’ (see Turk (2016; 2021) and my article in the Green Issue, February, 2026). These Jukurrpa-like systems of management and ethics can be likened to a set of controls for an AI system developed by any company, for a particular knowledge domain and under the control of a specific legal jurisdiction. Hence, this is a useful approach to developing a basic understanding of the need for AI ethics and guardrails and how they might be applied to enhance AI accountability. In this article I call this ‘Supervised Ethical AI’ or, alternatively AI Turk.

Ethics can be learned in lots of ways. In a sense, ethics is the strongest commonality between different world religions. As a teenager, I substituted belief in ethics for adherence to Protestant Christianity. About the same time, I started university studies, initially in Surveying and Cartography, focussing on research into the role of new digital and photographic technologies. As my university activities progressively developed more emphasis on research, eventually resulting in two PhDs, ‘research ethics’ became more and more important to me. This was supported by an Honors Degree in Philosophy and Psychology and resulted in various publications on this topic, often with collaborators (e.g. Turk and Mark, 2011). Interestingly, this publication referred to research on computer-based cultural repositories such as Geographic Information Systems (GIS). Although much less powerful than AI, GIS had great potential to advance detailed analysis of diverse datasets to facilitate land-use management and research in multiple other disciplines. 

As a dedicated researcher, it is always easy to be fooled into thinking that the results of your research are so important for society that it is OK to ‘cut-corners’ on some ethical aspects along the way. This is almost never true. My research culminated in a trans-disciplinary approach using Phenomenology as the required meta-paradigm to integrate all the theories and data from multiple disciplines, to achieve the most elegant, efficient and effective research results (Turk 2020; 2024). 

I have expertise in Human Computer Interaction (HCI) (Turk, 1992) and some understanding of the role of AI but I am not an AI expert. During my period of researching and teaching about Information Systems development methodologies, with PhD student collaborators, I used socio/technical approaches to integrate ‘Soft Systems’ methods with alternative ways to integrate human and computer strengths to produce optimum systems (Cross and Turk, 1997). A variant of that methodology can potentially be used for AI via an ethics-based integration of human understanding of objectives, methods and constraints with AI’s software power and innovation. 

It seems that researchers, were so impressed with the possibilities of AI, that ‘they threw the intellectual baby out with the digital bathwater’. There was a steady decline in HCI because they could not understand the intellectual power of discipline experts in command of sophisticated AI systems, via an extremely effective cognitively-ergonomic user interface, engaging with all possible senses of the expert. This is also potentially a pathway to implementing Ethical AI and adherence to other aspects of necessary AI guardrails to increase AI accountability. Reliance on stand-alone (autonomous) AI agents may also lead to a steady decline in human research capabilities.

Mechanical Turk 

The Mechanical Turk … 'chess Turk', also known as the ‘Automaton Chess Player’ or simply ‘the Turk’ … was a chess-playing machine first displayed in 1770, which appeared to be able to play a strong game of chess autonomously, but whose pieces were in reality moved via levers and magnets by a chess-master hidden in its lower cavity. . .   It was Constructed by Wolfgang von Kempelen … The machine was toured and exhibited for 84 years as an automation and continued giving occasional exhibitions until 1854, when it was destroyed in a fire. In 1857, an article published by the owner's son provided the first full explanation of the mechanism, which had been widely suspected to be a hoax but never accurately described while the machine still existed. [https://en.wikipedia.org/wiki/Mechanical_Turk]

We could call an intimately synergistic, cognitive combination of a human expert and an AI system an ‘AI Turk’, in celebration to its resemblance to the ‘Mechanical Turk’ from the 18th Century. In this integrated model it may seem that an autonomous AI agent is doing the work, but it is assisted by a human expert, who feeds algorithms and data into the AI software and monitors ethical constraints and other guardrail and accountability requirements. The human Supervisor can also continually monitor progress on problem-solving via an extremely sophisticated whole-of-body interface and, using their strong expertise, they can intervene to change the way the AI-based investigation proceeds. The Supervisor’s behaviours with respect to the AI system are similar to those of the operator hidden within the Mechanical Turk, who sensed the progress of a chess game and made the optimal mechanical chess piece moves. However, in this case, it will not be a hoax but a significant HCI innovation. Other AI researchers have made somewhat similar connections to the ‘Mechanical Turk’, but I am not aware of any work linking the AI Turk concept to HCI, AI ethics, guardrails and accountability, as in this article. It is pleasing that this term, by chance, includes my own surname. If the term ‘AI Turk’ is considered too exotic or obscure, the more prosaic term ‘Supervised, Ethical AI’ (SEAI) can be used.

How Ethics work in Extreme Circumstances such as AI

My understanding of the development process of AI is that a suitable assemblage of software is fed some particular data sets then given the task to find results to complex socio-technical problems. From here, it modifies its own software and seeks out other data to produce the best possible result, in the context of constraints and success parameters. We have already seen multiple examples of inadequately-constrained fully-autonomous AI agents developing new approaches and seeking data from unintended sources. In short, the AI software has inadequate accountability meta-parameters and supervision to ensure that any particular AI program has adequately applied ethics in undertaking it’s task.

The approach I am suggesting could safely produce the best of both worlds, via an integrated HCI system, rather than an independent, autonomous AI agent, permitting humans to remain in control. After recent well-publicised inappropriate attacks on other computer systems by AI research programs, there is already an unfortunate tendency to blame the AI, which is not formally capable of crime (because it is not a person), rather than punishing the humans who are supposed to be in charge. Rather than an operator pressing ‘go’ and abandoning responsibility, the type of integrated system (AI Turk) that I am proposing will ensure that criminal behaviour can be attributed to a human and an appropriate penalty applied. 

As discussed in my May 2026 Green Issue article, this needs to be linked to a particular version of AI, which, during a given task, will say, when appropriate, “I’m sorry, I’m afraid I can’t do that” and seek assistance from their human AI Supervisor. An example would be when complex ethical problems are beyond the AI’s understanding. Such an approach would also facilitate a continuously learning AI system, potentially using a range of human sources of understanding from individuals very experienced in particular disciplines relevant to the topic of interest. 

Possibly an ethics-free computational approach may be considered by some AI developers as one of the strengths of AI, being able to ignore those troublesome limitations of ethics that human researchers are almost always obliged to consider. An AI Supervisor would need expertise in a range of fields, in addition to their own discipline, including: Ethics, HCI, Cognitive Science, Logic, Linguistics, etc. Human AI Supervisors would learn how to work with the AI software to keep it within the guardrails, while producing optimum results. In complex socio/technical domains, this could become a highly skilled and productive role, with major responsibilities and legal risks justifying a high salary and expensive insurance. It is possible that one human Supervisor could simultaneously manage multiple AI systems, depending on the nature of the research domain and the rate of progress of each AI system in completing its ‘task’. Some complex interdisciplinary tasks may require supervision by multiple disciplinary experts, perhaps simultaneously, or more likely in sequence. The HCI design and development would need to consider and facilitate all of these options. 

How might we then have confidence that AI’s ‘crimes’ would always be punished? One potential approach is to ensure that there is a legally responsible individual person, such as the proposed AI Supervisor, or perhaps a group of people collectively responsible for AI accountability. Different approaches to crimes may apply for government and commercial AI projects and in different legal jurisdictions.

Returning to the Australian Aboriginal Jukurrpa example, we could think of an AI Supervisor as fulfilling a similar role to that of an ‘elder’ in an Aboriginal community. The elder’s long experience has developed the necessary respect, intelligence and wisdom for them to provide advice and supervision of the application of ethics by younger people. The AI Turk Supervisor seeks to utilise similar attributes and exert equivalent powers and responsibilities.

Implementing Ethical Constraints as an Example of AI Guardrails to Increase Accountability

US President Donald Trump and Chinese leader Xi Jinping met at the White House in Washington, D.C., on September 24th, 2026, concluding a three-day bilateral state visit during the United Nations General Assembly in New York. They discussed AI safety guardrails and technological competition. Trump claims they are now ‘very great friends’. They agreed to set up a formal communication channel and incident line to discuss artificial intelligence (AI) safety and national security risks. It is the relationship between AI and ultra-Capitalism or strong Communism which provides the deepest fear for the future, especially because of links to extreme nationalism.

On a more positive note, Australian Prime Minister Anthony Albanese played a key role in discussions among ‘Middle Power’ nations regarding guardrails for AI. This group of more than 20 world leaders at the United Nations General Assembly issued a plea for urgent international agreement to control the development and application of AI. This was followed a few days later (25th September) with an announcement by the PM that in June, an OpenAI agent hacked an Australian repository of medical statistics. The fact that the company took such a long time to tell anyone, has led to increased demands for AI safeguards and enhanced accountability. Hence, there is an urgent search for ways to achieve this.

The desire is to develop ‘Ethical AI' which is preferably universal, or at-least as a set of interlocking different types of accountability control (guardrails) in each legal jurisdiction. A key initial requirement is that AI adequately understands the structure of ethical principles in various domains, the software architecture to support this constraint system and the mechanisms and agreements required to efficiently and effectively monitor compliance. I have theorized a few basic concepts that this might involve:

A. With respect to all topics or areas of data that AI ‘tasks’ (’missions’) will involve, and the country(ies) where it will be used, the AI system must fully understand the following aspects:

  1. The ethical principles, processes and official procedures required and the complex combinations and/or interactions between them;
  2. All mechanisms required to relate all data, software components, computations, reasoning and outputs (to the required level of detail, completeness and exactitude) to any and all aspects of ethics;
  3. All government regulations (of any type) which are required to place ethical constraints (accountability) on any AI activity and how to respond to them, in all possible circumstances, in exactly the correct standardised fashion;
  4. How to understand, document and communicate any consequence or outcome related to any ethical infringement, no matter how small, in the standardised manner stipulated for that version of AI, type of task and any area of ethics in that legal jurisdiction;
  5. All the regulations relating to potentially multiple ‘trigger points’ for reporting any type of ethical constraint that is possibly (or certainly) ignored (or limited) within any AI task execution;
  6. How to communicate in a very timely and reliable manner with all agencies relating to ethical ’incidents’, especially those with legal significance;

  7. Understand and be absolutely required to respond to any significant breaking of ethical constraints, no matter of what type or magnitude, within the standard set of AI constraints within that legal jurisdiction, including, temporary halt of operations, reporting of the event, or even permanent termination of the AI task.

    B. In order to appropriately train and control the guardrail processes, an AI Supervisor will assist any AI system to understand and respect ethical constraints and also manage other guardrail constraints involved in accountability regulations.

    C. A critical aspect of a system involving an AI Supervisor is the HCI design to ensure the most intimate relationship needed to perform that role, using the whole Supervisor’s body as an interface, to facilitate maximum integrated physical and mental cognitive ergonomics between the AI and its Supervisor (Turk, 1992; 1995). Potentially, this will immediately alert the Supervisor that the AI agent has inappropriately entered a sensitive database, so that they can command that the AI to cease interrogating the database before removal of any data which would constitute a crime.

    D. Should any AI system break any laws, the AI Supervisor will be the person legally responsible, so that they must be trained and supported in ways that facilitate their understanding of all legal issues in the legal jurisdiction(s) where they are operating.

    E. No aspect whatsoever of AI software and operating procedures should permit an AI system to over-ride the ethical constraints infrastructure and/or any commands by the human AI Supervisor charged with the accountability management of that particular AI system.

    F. It is absolutely critical that the Ethical AI approach in principle and detail is at-least complimentary (in a very organised manner) or, preferably, identical between legal jurisdictions (states, nations, regions, world).

    G. A mutually compatible Top-Meta-AI-Overview-System should be developed (as rapidly as possible) to operate under the control of a relevant UN agency and a financial system established so that it is well funded by contributions from those profiting from AI.

    H. This top-level organisation will hold regular technical conferences and also presentations to World political leaders to ensure that its operations are well understood and enthusiastically embraced by all countries.

    I. A meta-AI system must be initiated in each AI development organisation to overview ethical and other guardrail aspects of all its activities and report on AI accountability to both the heads of that organisation and the controlling authority in any legal jurisdiction.

    J. This ‘Ethical AI’ infrastructure (at all organisational levels between the individual researcher and the UN) will be reviewed, initially monthly (then 3, 6 or 12 monthly as considered appropriate) to ensure it is fit-for-purpose.

    K. For the sake of world peace and continued economic and social development in an efficient, effective and equitable manner, if the above system determinants of ‘Ethical AI’ accountability cannot be achieved, or at some stage, do not completely continue to operate, an overarching World organisation (e.g. the UN) must rapidly intervene to ensure that a similar (or superior) system is implemented. 

    L. There may be situations where, even if AI has become even more effective and obedient to ethics and guardrails, where it would remain necessary for the AI to be supervised by an expert in some emerging field of research. This would facilitate dealing with the inter-disciplinary complexity or novelty of intellectual knowledge needed to research some emerging complex socio-technical problem. This may be an initial phase of research, necessary while the AI system accumulates sufficient knowledge about the new research field.

    M. If at some time in the future AI systems evolve (perhaps via the training methods mentioned above) it may no longer be necessary to employ an AI Supervisor (within one, some or all legal jurisdictions). This would require that the software is so sophisticated and the surveillance of its activities so detailed, rigorous and assured that it can be totally trusted not to do anything that is un-ethical, is counter to other types of guardrails or aspects of accountability or is illegal.

    N. If the software develops to the standard outlined in M, it might then be necessary to make other arrangements regarding which legal entity is responsible for any crimes that could potentially be committed by some internal AI software bug or perhaps an external hacking attack from outside the AI system, possibly located in a legal jurisdiction where AI security is less advanced. 

    O. Presumably, such a situation may also arise in a legal jurisdiction where AI security is very advanced, however, that nation seeks advantages in some trade-war, or even full-scale military conflict, via attacking an enemy’s un-supervised autonomous AI systems, to use them to create chaos and digital destruction. 

    P. Extended AI capabilities come with more extreme risks, perhaps even encouraging a new version of the ‘scorched earth strategy’, aimed in this case at a country with an economy deeply dependent on digital productivity and/or defences. Hence, for critical AI installations, it may be worth retaining AI Supervisors to guard against such possibilities, at least until our World becomes less addicted to war and/or if AI achieves an even higher level of capability which ensures that the AI system automatically responds to any internal bug or external hack by immediately shutting down.

This list provides at least an initial set of ideas about how such an approach to AI control and accountability might be arranged. This allows us to predict its likely costs and the benefits that Supervised AI Ethics (an AI Turk arrangement) could provide as an alternative to autonomous AI agents. It would be necessary to convert such a set of concepts into an international agreement, government legislation in each collaborating country, plans for government program development and detailed agreements with companies or organisations involved in AI development. 

Although there is a sense of urgency regarding this matter, I believe that it is best to debate such matters at a conceptual level (as in this article) and only do the hard detailed work of legislation, agreements, etc. once there is general agreement on the most desirable approach. I am providing these suggestions in the hope that it can help facilitate the development of well structured, open and honest debate between members of relevant academic disciplines and citizens and leaders of as many countries as possible. 

Conclusions

We are only near the beginning of this critical journey. Although any AI threats should not be over-rated, one could hypothesise that the rate of progress on Ethical AI could determine when, and how, the World (as we know it) might disappear. I contend that this is definitely not the sort of topic to leave to technical experts or to political leaders, or even to an unholy alliance between them, with economic and national considerations playing a major role in their decision-making. It needs to be considered by all people everywhere of all types. 

This article proposes the inclusion of an AI Supervisor who would intimately work with the AI system to ensure that ethical aspects and government guardrails are respected. They would serve as the legal entity responsible for any failures with-respect-to such regulations, rendering the system into an AI Turk configuration of AI software, hardware, HCI and at least one human expert. After adequate HCI developments, in the right circumstances, this could provide a viable alternative to fully autonomous AI agents, providing supervision which would include accountability for unethical actions.

After all, AI is really just limited synthetic cognition (without the bodily cognition aspects) rather than actual intelligence and wisdom, which is uniquely human. AI relies on pattern recognition ‒ on correlations rather than causations, often ignoring critical aspects of temporal and other contexts. This is despite President Trump seeking to rename AI “Super Intelligence (SI)”. The most successful form of research may involve an integrated arrangement where a discipline expert works intimately with an AI system to produce innovative research results, while simultaneously supervising the AI with respect to ethics and other aspects of accountability. 

This concept may well result in development of an exciting new area of HCI research, to design, prototype, evaluate and further develop the HCI necessary to support an integrated research approach. A key issue regarding the development of an ‘AI Turk’ system is how to enable the ideas of many commentators from a wide variety of disciplines to be collected and integrated in an efficient, effective and equitable manner. Clearly communication will play a crucial role. I encourage all Greens’ members, organisational leaders and politicians to actively engage with this issue as quickly and deeply as possible. I hope this article contributes to this process and I declare my readiness to participate in further debate. 

References

Cross, R. and Turk, A. G. (1997) Quality specification of information systems: An integration of three socio-technical approaches. In: J. S. Ahluwalia (Ed.) Total Quality Management, the proceedings of WCTQ 97, 7th World Congress on Total Quality, February 1997, New Delhi, India: Tata McGraw-Hill; pp. 619-632.

Turk, Andrew Glenn (1992) GIS cogency: Cognitive ergonomics in geographic information systems. Unpublished doctoral dissertation.  The University of Melbourne, Australia.

Turk, A. G.  (1995) Cognitive ergonomics analysis methodology. In: Cognitive aspects of human-computer interaction for geographic information systems, NATO ASI Series D, Vol. 83, Nyerges, T. L., Mark, D. M., Laurini, R. and Egenhofer, M. J. (Eds), Kluwer Academic, pp. 393-403.

Turk, A. G. (2016) Space, Place, Phenomenology and Jukurrpa. In: Jordaan, J., Haddrell, C. and Alegria, C. Dialectics of Space and Place across Virtual and Corporeal Topographies (e-book). Oxford UK: Inter-Disciplinary Press. pp. 127-141.

Turk, Andrew Glenn (2020) Understanding modes of dwelling: A transdisciplinary approach to phenomenology of landscape. Unpublished doctoral dissertation. Murdoch University, Perth Australia.

Turk, A. G. (2021) (T)Jukurrpa as Strong Examples of Heideggerian Topology. Academia Letters, Article 1551.

Turk, A. G. (2024) A Transdisciplinary, Engaged, Phenomenological Investigation of Dwelling and Landscape Language. Cambridge Scholars Publishing.

Turk, A and Mark, D. (2011) Ethical Perspectives on Landscape in Language Research. In: Mark, D. M., Turk, A. G., Burenhult, N. and Stea, D. (Eds.) Landscape in Language: Transdisciplinary Perspectives. Benjamins.

Header photo: The challenge of controlling AI Source: [https://pixabay.com/illustrations/robot-technology-future-android-89193…]

[Opinions expressed are those of the author and not official policy of Greens WA]