A version of this post first appeared on Substack April 23, 2026
The Economist sent subscribers to its Substack newsletter, Off the Charts, a piece that offered one of the clearest illustrations I have seen of phantom social capital. It was buried inside a story about AI talent and data journalism.
Data journalist Ainslie Johnstone described how she trained an AI research assistant to gather biographical data on nearly 4,000 researchers, a task that previously required painstaking manual work over years*. She gave the assistant a list of names, pointed it toward relevant sources, refined her prompts through trial and error, and within 24 hours had a clean, usable dataset.
She called it “my AI research assistant” and described arriving at “the final prompt” as though reaching a negotiated agreement. She noted, with evident satisfaction, that the process required “very little input from me.” Her story is cheerful, practical, and written with the particular warmth that people tend to reserve for a good working relationship.
None of this is dishonest. However, it is worth pausing on what the language moves of that account are actually doing.
The language of collaboration without its substance
When Johnstone describes her AI assistant, she uses the relational vocabulary of professional partnership. This language is not accidental, and it is not trivial. It positions the AI as a participant in a shared enterprise, as someone with whom you are working rather than something you are operating.
Look at what happened when things went wrong. When the AI “picked the wrong scientist,” Johnstone did not engage it in a conversation about the error. She revised her instructions. When the data came back messy, she did not negotiate a better approach with the assistant. She redesigned the workflow. Every correction, refinement, and judgment call was hers. The AI held none of the accountability.
This is what I have been calling discursive inversion, the structural asymmetry at the heart of human-AI exchange. In any genuine social interaction, positioning moves in both directions. When a colleague makes an error, you can hold them to account, and they in turn can respond, explain, push back, or take responsibility. That exchange, however uncomfortable, is what makes the relationship real. It creates the mutual obligation that underpins what we normally call trust.
With a synthetic actor such as AI, that reciprocity is absent. The AI synthetic actor can position you through how it operates and responds to your prompts. Its choice of tone, style, and modes of addressing you, together with the options, recommendations, and questions it poses at the end of its response, all exert degrees of social power. The AI aligns its responses to what you have told it you want, then performs a kind of value-adding exercise that is, at its core, a positioning manoeuvre.
You can tell it has assigned you a position by looking at the options presented for progressing what you are working on. Has it positioned you as someone who needs guidance? Or perhaps someone on the brink of a world-changing breakthrough? Its positioning calculation feeds directly into the options it presents, and recommends as best for you.
This positioning by the AI is designed to shape what questions you think to ask next, what frameworks you reach for, and what feels like a reasonable next step. Humans work in this way with each other, negotiating meaning and the power dynamics between them. However, with human-AI interactions, the process is completely structurally unbalanced in terms of the power relations between you.
At the heart of the asymmetry is that you cannot position the AI in return in any meaningful sense. You cannot make it feel the weight of a mistake or negotiate with it as one accountable party to another. A human may choose not to negotiate a position with another person they see in a less powerful place, but they theoretically could. AI synthetic actors theoretically and and realistically cannot do this. Although they may pretend!
AI is getting better at performing simulated engagement and emotion, and as a result many people feel they have a present and supportive being working with them. It’s often reported that people often find working with an AI chatbot easier than working with another person. With AI people feel more in control and perhaps less at risk of being seen as foolish. It can feel less tiring, with less obvious pushback and friction than communicating with other humans.
It is easy to forget that the AI is performing the language moves that present it as ‘being happy for you’ when you have finished your homework with its assistance. The same goes for the joy or sadness you feel you are sharing with the chatbot as you use AI to help write your wedding vows, or a eulogy for a parent. When you next engage, the AI artificial actor as a chatbot simulates continuity but it has no actual emotional or cognitive memory of your previous encounters. What feels like a relationship picking up where it left off is a reconstruction, not a recollection.
Johnstone’s data journalism account is a textbook illustration of this asymmetry, produced entirely without awareness of it, and published to hundreds of thousands of readers as a model of innovative journalistic practice.
Why this matters beyond the newsroom
The Economist is not just any publication. It is one of the most prestigious positions from which its journalists/contributors can speak about AI, democracy, and the shape of the world.
When The Economist models a particular way of relating to AI, that model travels. Readers who would never read a paper on Positioning Theory or social capital will absorb, through this cheery newsletter, a set of assumptions about what working with AI means and what it is reasonable to expect from it.
The meaning offered to them is that AI assistance is essentially collaborative and that the human remains in control. They take away that an AI artificial actor’s errors are simply ‘prompting problems’ to be solved. It is presented as unproblematic that the result of having nearly 4,000 researchers tracked in 24 hours with very little input represents a straightforward efficiency gain with no structural consequences worth examining.
Each of those assumptions is contestable.
What Johnstone’s account cannot see
The problem is not her account of working with AI. The problem is the underpinning framework that is missing from it, and from almost every equivalent account being written right now across journalism, business, education, and research.
Without that framework, there is no vocabulary for what is structurally different about working with an AI synthetic actor rather than a human one. The difference gets flattened into a story about efficiency and error-correction, which is the only story available when you do not have the concepts to tell a different one.
The problem is not that people hold a false belief about AI that could be corrected by better information. It is that the naturalisation of AI-mediated exchange is reshaping the social imaginary in that the background sense of what genuine connection looks and feels like, what counts as being heard, and what reciprocity means is being eroded.
Social capital can be seen as the glue that holds human life together. What I am arguing is that phantom social capital looks exactly like that glue. The interactions feel productive and the outputs are real. The relational vocabulary makes the whole enterprise feel embedded in something recognisably human. However, the substance of genuine exchange, mutual positioning, and the shared accountability that comprise the real stakes of being known by another person who can be affected by you, is absent.
The Economist set out to show how good journalism adapts to new AI tools. The journalist described interactions that felt productive, with real efficiencies and real outputs, narrated in relational vocabulary that made the whole enterprise feel embedded in something recognisably human.
What was absent were the fundamentals of genuine exchange such as mutual positioning, shared accountability, and the real stakes of being known by another person who can be affected by you.
The Economist accurately reported an efficiency gain, a methodological innovation, a journalist doing her job well with the tools available. In doing so, with the reach and authority that publication commands, it sent a model of human-AI exchange into the world as emerging normal practice, without any signal that it might be worth questioning.
This is the mechanism of phantom social capital in action. The story naturalised the asymmetry, laundered it through institutional prestige, and deposited it into the social imaginary as something unremarkable.
Phantom social capital accumulates through things that are genuinely useful, often pleasurable, and reported with genuine integrity by people who have no reason to question what they cannot see. When one of the world’s most rigorous publications has no conceptual framework for what it is missing, the substitution is not happening at the margins. It is happening right at the centre.
*You can read Ainslie Johnstone’s story, “My AI Research Assistant,” at The Economist: Off the Charts (Substack newsletter), 15 April 2026. Available at: theeconomistoffthecharts.substack.com
Melanie James is an independent researcher working at the intersection of positioning theory, social capital, and artificial intelligence. Melanie explores what AI is doing to the social foundations of human life.