OpenAI, Meta and Mistral describe their AI systems in increasingly human terms, and a University of Münster study argues that this language can shape how much trust and authority people give the machines. The important point is not that chatbots use friendly words. It is that calling software a thinker, expert or colleague can quietly change what people expect it to know and what they are prepared to let it do.
Watch Desk analysis
What happened
Anastasia Glawatzki analysed 35 corporate documents for a study published in AI & Society. As summarised by Devdiscourse, the research traces three recurring visions: AI as a tool with human-like capabilities, AI as an empowering assistant or collaborator, and AI as an increasingly autonomous superintelligence.
The study says corporate descriptions attribute reasoning, understanding, communication skills, social abilities and even personality to systems that still generate outputs through computational models. Words such as “expert”, “assistant” and “thought partner” carry assumptions about judgement and reliability before a reader has tested what the system can actually do.
Why it matters
Language is not merely decorative when it is attached to software used in workplaces, public services and professional decisions. If AI is presented as a dependable colleague, people may move from functional trust, such as asking it to draft something, towards epistemic trust, such as accepting its judgement. That is a sizeable leap hidden inside a rather pleasant noun.
Glawatzki’s argument also turns the usual accountability question around. Describing a system as independently reasoning or deciding can make the organisations behind it less visible, even though people still choose the training data, model design, safety measures, deployment rules and commercial priorities.
The paper says productivity and democratisation claims can make adoption feel compulsory. If AI-assisted workers are framed as faster or more competitive, choosing not to use it may begin to look like falling behind. The study also warns that such narratives can obscure labour displacement, unequal access and the human work required to train and maintain these systems.
Our read
This is a useful study because it asks a question that benchmarks generally dodge: what happens when the description of a machine becomes part of the machine’s social power? It does not prove that every use of “reasoning” is manipulative, nor does it settle whether AI can perform any particular cognitive task. It offers a sharper test for claims about capability.
Readers assessing an AI product should separate the label from the evidence. What can the system demonstrably do, under what conditions, with what error rate and who remains responsible when it gets the answer wrong? “Helpful colleague” is a marketing phrase until someone explains the colleague’s supervision arrangements.
What to watch
- Whether companies publish task-level evidence alongside human-sounding capability claims.
- Whether procurement rules distinguish conversational presentation from demonstrated competence.
- How AI language changes across non-Western companies and cultures, a gap the study says it does not address.
- Whether regulators examine anthropomorphic claims as part of AI accountability and consumer protection. The specific WittyWires relevance is the study’s documented analysis of how major AI companies frame systems as thinkers, helpers and autonomous actors, with direct implications for trust, delegation and governance.
Discussion spark: Should AI companies be required to describe systems in technically precise terms, or would banning human-like language make useful technology harder for ordinary people to understand?
Sources and evidence
- How OpenAI, Meta and Mistral Frame AI as Thinker, Helper and Future Superintelligence – Devdiscourse (23 September 2026, 17:05 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.