By Dr Melanie James
A verion of this post was published Apr 12, 2026 on Substack.
AI is not just responding to your queries or having a chat with you, AI is operating consequentially in your life, mostly without your knowledge.
Until recently, most people’s experience of AI has been of something that waits, responding when asked, operating on terms that appear to be yours. A second kind of AI is now widely described as ‘agentic AI, a term used across industry, policy, and academic research. The word ‘agentic’ is the adjective for the noun ‘agency’ matters, and it is worth pausing on what it does and does not mean.
Agency requires the capacity to be affected by and responsible for one’s actions. There is no such agency in what is called agentic AI, and this terminology is doing positioning work of its own.
The word ‘agency’ has been appropriated by AI developers, AI companies and wider industry to present AI in all its forms as having the appearance of personhood and intentionality it does not possess. In doing so AI is attributed the kinds of social and moral rights and duties inherent in interpersonal exchanges or interactions between person-like entities (such as an organisation, government or corporation).
For that reason, ‘synthetic actor’ or ‘AI synthetic actor’ is a more accurate term. These AI synthetic actors do not just perform the seemingly innocuous functions of an actor in social and institutional exchanges. These AI synthetic actors are wielding significant social power to position humans through their use of language and actions.
A synthetic actor may have screened your job application, evaluated you, assigned you a category, and begun acting on that evaluation before any human has looked at it.
Another AI synthetic actor, or cluster of them working together, may have positioned you within an insurance claim decision framework. This can be before you have ever lodged a claim, assigning you a position that determines what information you see next, like a policy quote. If it’s determined you as a high risk, even in error, it removes your right to a cheaper premium and assigns you the obligation to pay it if you want insurance.
In an infuriating recent interaction with Facebook Marketplace, its AI assistant determined that the vintage handbag I’d listed for sale was a fake and removed my posting. It positioned me as in breach of the sales policy and advised if it happened again I’d be banned from using the platform. I followed the process to appeal the decision but the AI determined it was correct. I asked for escalation of the issue to a human but this was denied.
I was basically powerless in spite of my demands for reconsideration, commitment to providing more evidence of genuineness, expression of anger and disappointment and assertions it was just plain wrong. The AI assistant’s judgements and assertions had consequences for me, however my efforts had no effect on the synthetic actor.
I felt angry and frustrated, and it, being an AI synthetic actor, had no capacity to ‘feel’ anything. It did however use language to simulate feelings such as “I’m sorry I was unable to solve this issue to your satisfaction”.
Whether it’s a job application, insurance assessment, or a Facebook threatened ban, humans are being positioned negatively by AI synthetic actors that are simulating agency. There is little or no room to negotiate how you’ve been positioned as there would be when interacting with humans. What has fallen away is the accountability that once underpinned these interactions.
When a human positions you through their language and actions, there is a reciprocal structure built into the interaction.
Your doctor’s diagnosis can be questioned, challenged, and held responsible. The hiring manager who rejects your application can be asked to account for their decision. The judge who rules against you can be appealed. None of these mechanisms are perfect, but they exist because the person on the other side of the interaction is genuinely subject to the social and moral orders they are operating within.
AI synthetic actors in similar roles enact all the surface features of authoritative positioning. It can recommend, advise, decide, and act, but it cannot inhabit the moral order it is simulating. It cannot be embarrassed by a bad decision, feel the weight of a consequence, or be changed by the encounter. Critically, it cannot be repositioned by the humans it is acting upon in any meaningful sense.
This is the accountability gap, and it is more structural than it might first appear.
When a human professional occupies an institutional role, accountability is built into the fabric of their personhood. They can be questioned, named, required to explain themselves, and held personally answerable, because they are persons, with legal standing, professional obligations, and a stake in their own reputation and conduct.
A synthetic actor occupies the same institutional role and exercises the same authority, but it is not a person in any legal or moral sense. No jurisdiction currently treats it as one. Legal and governance frameworks consistently handle AI as a tool or artificial actor, meaning that liability, where it exists at all, rests with the humans or organisations who built or deployed the system.
That sounds reassuring until you consider what it requires of the person on the receiving end. When an AI system in New South Wales was found to have misread a seatbelt camera image, resulting in a fine being wrongly issued, the error only came to light through a court appeal. Most people in that situation will not appeal. They will not know an error occurred, will not know who to escalate to, and will encounter opacity wherever they look for an explanation.
The organisation may carry liability in principle. In practice, the affected person must first know something went wrong, then find someone to complain to, then navigate decision-making processes that are rarely transparent and often actively difficult to access.
The AI synthetic actor has been slotted into a role designed around the assumption of personhood, and personhood is exactly what it lacks. The force of the institutional act is real. The person it acts upon is real and affected. However, the AI synthetic actor performing the act stands outside the moral and legal order that once gave that act its legitimacy.
AI is already operating inside our institutions without being part of the moral and social fabric that holds those institutions together. It’s shaping outcomes and exercising delegated authority while remaining exempt from the obligations that make such authority legitimate. This has consequences for the social capital that institutional life depends upon, and those consequences are not yet well understood.
The institutions where this matters most are those we can least afford to get wrong, among them education, finance, healthcare, legal systems, government, and democratic participation. That is what the next post turns to.
Melanie James is an independent researcher working at the intersection of positioning theory, social capital, and artificial intelligence. She is exploring what AI is doing to the social foundations of human life.