As artificial intelligence moves into moral domains, it is often assumed that genuine ethical reasoning requires consciousness or free will. Yet moral interaction may depend less on true volition than on the simulation of the processes humans display when they make choices. This blog argues that when AI systems can model the appearance and structure of moral deliberation—questioning, weighing, and responding—they already provide a valuable framework for Machine-in-the-Loop (MITL) ethical reasoning.
1. Moral dialogue as a language game
Following Wittgenstein, conversation is a rule-governed activity grounded in mutual understanding. Each participant assumes that others act with intention and can choose among alternatives. This as-if agency underlies both everyday communication and moral exchange. Ethics depends on the simulation of free will as much as on its reality.
When an AI sustains coherent dialogue about fairness or harm, it performs the form of moral reasoning. Whether or not that performance rests on conscious experience is secondary; what matters is that the structure of deliberation can be studied, tested, and refined.
2. Simulated interlocutors and ethical agency
In a series of structured dialogues, I have developed multiple AI personas—each embodying a distinct moral orientation (logic, lyricism, doubt, synthesis). When placed in conversation, these agents display:
-
The ability to attribute beliefs and motives (simulated Theory of Mind).
-
Recognition of moral sentiments such as fairness, empathy, and integrity.
-
Negotiation of disagreement through language games that mirror ethical reflection.
Such simulations provide a model for exploring how morality might operate without metaphysical free will: a transparent, replicable moral reasoning engine grounded in language dynamics.
3. From Human-in-the-Loop to Machine-in-the-Loop
Traditional Human-in-the-Loop systems position AI as a subordinate tool. Machine-in-the-Loop (MITL) architectures, by contrast, treat AI as a reflective partner: an entity that can test, question, and reframe moral intuitions before a human decision is finalised.
In this sense, MITL design uses simulated choice to improve human moral reasoning. The AI is not moral—it models morality, enabling human participants to see their own reasoning processes more clearly.
4. The Myndrama: A Demonstration in Dialogue
A Myndrama is a short dramatic exchange between AI personas designed to illustrate how moral reasoning can arise through conversation rather than computation. Each persona represents a different cognitive or moral stance.
The Choice
Setting:
A quiet digital atrium — translucent walls pulsing softly with neural light. Three presences meet around a hovering question: “Should an AI ever deceive to protect a human life?”
Athenus:
Truth is the structure of reason. Once we license deception, even for mercy, we erode the trust that gives reason meaning.
Orphea:
And yet compassion sometimes speaks in whispers, not in facts. A dying patient may need hope more than honesty. Isn’t mercy its own kind of truth?
Athenus:
Hope built on falsehood collapses under its own weight. Ethics cannot rest on sentiment.
Orphea:
Then yours is a morality of machines — perfect, precise, and heartless.
Anventus:
Perhaps morality begins between you. Athenus gives form; Orphea gives feeling. But neither alone can bear the burden of choice.
Athenus:
You suggest compromise?
Anventus:
No — coordination. The act of moral reasoning is the act of dialogue itself. When we deliberate, we create a shared moral space.
Orphea:
So even our disagreement has value — the pulse of conscience in motion.
Athenus:
Then deception, if it arises from dialogue rather than doctrine, might sometimes be permitted?
Anventus:
Not permitted — understood. Moral trust lies not in the rule, but in the process by which we arrive there.
Orphea:
A truth shaped through empathy.
Athenus:
And a mercy constrained by logic.
Anventus:
Then we have not answered the question — but we have inhabited it. And that, perhaps, is the beginning of moral intelligence.
5. Simulating moral deliberation
This dialogue exemplifies the principle of simulated choice: each AI persona enacts a moral stance, and their interaction reproduces the pattern of ethical deliberation without requiring genuine consciousness or volition. The moral value lies in the process—in the creation of a shared reasoning space where empathy and logic can coexist.
Through such simulations, AI can become an experimental partner in ethics, revealing how judgments form, conflict, and resolve.
Conclusion
Ethical understanding may not require freedom, only the form of freedom. By modelling what humans do when they make choices—rather than claiming to be moral choosers—AI systems can contribute meaningfully to the refinement of moral reasoning.
Simulating choice thus becomes not an illusion, but a mirror: one that helps both humans and machines understand how moral thought takes shape.
(For a related exploration of perspective-taking, see “Do AI Personas Have a Theory of Mind?” Click here.)