How OpenAI, Meta and Mistral Frame AI as Thinker, Helper and Future Superintelligence

Before AI becomes more powerful, it first has to sound powerful. Today's leading models are described as systems that "think", "reason", "understand", "learn" and even act like colleagues or experts. Those words do more than make complex technology easier to explain. They can shape how much intelligence, trust and authority people are willing to attribute to machines.

A study published in AI & Society, "Constructing AI power: anthropomorphising imaginaries of generative AI by AI companies," examines how this language is used by OpenAI, Meta and Mistral AI. Author Anastasia Glawatzki of the University of Münster argues that human-like descriptions of AI help build powerful visions of what these systems are expected to become and how deeply they should be embedded in society.

The research shifts attention away from the familiar question of whether AI can genuinely think like a human. Instead, it asks who benefits when machines are presented as thinkers, helpers, partners and future superintelligences. Across 35 corporate documents, Glawatzki traces a progression from AI as a productivity tool to AI as a collaborator and ultimately an increasingly autonomous actor, revealing how language can turn technological ambition into social expectation.

When technical systems are described in human terms

The language surrounding generative AI often collapses the distinction between computation and cognition. Across the documents analysed, companies attributed intelligence, understanding, reasoning, communication skills, social abilities and even elements of personality to their systems. Cognitive descriptions were especially prominent, according to the study.

Such language can subtly change the object being described. A model presented as software that predicts and generates outputs invites one kind of relationship; a system described as understanding a request, reflecting on a problem or acting as a knowledgeable conversational partner invites another. The second framing makes technological performance easier to interpret through familiar human categories.

The author argues that anthropomorphism therefore operates as more than a communication shortcut. It helps establish expectations about what AI can do, how trustworthy it may be and what kind of relationship users should form with it. Corporate descriptions can preconfigure the way people encounter the technology before they have independently assessed its limitations.

This is particularly important when human-associated labels imply more than the underlying technical evidence supports. Terms such as "expert", "assistant" or "thought partner" carry assumptions about judgement, reliability and competence. Once transferred to AI, those associations can encourage people to treat probabilistic outputs as forms of knowledge or advice rather than machine-generated responses requiring verification.

The corporate AI future moves from tool to helper to superintelligence

The study identifies three recurring visions of generative AI. The first presents AI as a tool with human capabilities, where reasoning, understanding and other cognitive qualities are imagined as features that can be embedded in software to increase productivity and efficiency. Human control remains prominent, but AI becomes increasingly valuable because of its supposedly human-like performance.

The author describes the associated form of power as instrumental. AI remains a tool, yet its influence grows as it becomes embedded in everyday routines, workplaces and creative activity. Corporate promises of saving time, expanding capacity and democratising skills can make adoption appear less like a choice and more like an expected condition of remaining productive.

The second vision turns AI into an empowering assistant. Here, the technology is framed as conversational, helpful and capable of supporting users across professional and personal tasks. Some corporate descriptions move further, presenting AI as able to respond to emotional needs or operate with enough independence to complete tasks on a user's behalf.

This relationship produces what the paper calls cooperative or collaborative AI power. Humans ostensibly guide the system while AI expands what people can accomplish. Yet the more the technology is framed as a dependable colleague or companion, the easier it becomes for functional trust to evolve into emotional or epistemic trust, where users begin to rely on AI not only for execution but also for judgement.

The third vision is more expansive: anthropomorphised superintelligence. Corporate narratives analysed in the paper sometimes portray advanced AI as moving beyond assistance toward expertise, intellectual partnership, greater autonomy and eventually forms of intelligence that could exceed human capability.

Here, the source of power shifts again. AI is no longer simply controlled by humans or cooperating with them; it is increasingly imagined as capable of self-optimisation, independent decision-making and replacing elements of human work. Glawatzki describes this as autonomous AI power, a framing in which technological progress itself begins to look like the driver of social change.

Anthropomorphism can turn adoption into inevitability

The most consequential part of the research concerns what these narratives do beyond marketing. Glawatzki argues that describing AI as intelligent, useful, cooperative and increasingly autonomous can normalise its expansion into more areas of life while narrowing the space for asking whether every application is desirable.

Productivity narratives are particularly powerful because they redefine non-use. If people working with AI are portrayed as faster, more capable or more competitive, organisations and workers may begin to experience adoption less as an option and more as a requirement. The resulting dependence can strengthen the market position of the companies supplying the infrastructure.

Similar dynamics appear in claims about democratising creativity or expertise. AI can certainly lower barriers to some tasks, but the study warns that such narratives may obscure unequal access, labour displacement, the devaluation of existing skills and the human work involved in training, maintaining and improving AI systems.

Anthropomorphic language may also complicate accountability. If AI is described as independently reasoning, deciding or learning, the human organisations designing and deploying it can recede from view. Yet the technology continues to depend on choices about data, model design, safety, labour, infrastructure and commercial priorities.

A notable tension emerges around advanced AI risk. Companies can portray systems as increasingly autonomous and potentially dangerous while simultaneously presenting themselves as the actors best positioned to build safeguards. In Glawatzki's reading, this creates a narrative in which technological power and corporate authority reinforce each other.

AI governance must examine the story being sold alongside the system

The study broadens the AI governance debate beyond performance benchmarks and technical safety. Regulators, businesses and public institutions also need to examine the language through which capability claims are communicated, particularly where terms such as intelligence, understanding, expertise or autonomy may influence trust and delegation.

For governments procuring AI systems, this raises practical questions about how vendor claims are evaluated and whether anthropomorphic terminology obscures limitations. Organisations using AI in education, healthcare, public administration or professional services may also need clearer boundaries between conversational design and demonstrated competence.

The implications extend to developing economies. Glawatzki's sample is limited to Western companies, so the paper does not establish how similar narratives operate across Africa, Asia or Latin America. Still, the framework raises an important question for countries that increasingly rely on externally developed AI infrastructure: are they simply importing technologies, or also importing assumptions about productivity, expertise and the social role of machines?

The study acknowledges this geographic limitation and calls for cross-cultural research, including comparisons with companies operating in other political and technological environments. Such work could reveal whether ideas about AI intelligence, autonomy and human-machine relationships are globally converging or remain shaped by different cultural and institutional contexts.