Stop Saying 'the AI.' The Language We Use Shapes How Dangerous Anthropomorphism Becomes

When you say "the AI decided" or "an AI that thinks," you are doing more than simplifying a description. You are using language that implicitly ascribes agency, intention, and even moral responsibility to a statistical pattern generator. This habit is not harmless shorthand—it is a linguistic vehicle for deception, manipulation, and security risk.

Anthropomorphising AI means treating systems as though they possess human emotions, intentions, consciousness, or moral judgment . While attributing human qualities to machines can make interactions feel more approachable, it also clouds users' perception of the technology's true nature and limitations . The core argument from researchers and editorial boards is that the nouns and verbs we choose—especially whether we say "the AI" or "AI systems"—directly fuel these misconceptions.

The deceptive power of human-like design

A primary ethical concern with anthropomorphic AI is deception at scale. When users cannot reliably distinguish between human interlocutors and AI systems, threats of deception, manipulation, and disinformation emerge . Companies can deliberately design human-like conversational agents to exploit users' innate tendency to personify, enabling what researchers call "anthropomorphic seduction": the allure of convincing human-like interaction in the absence of any true human experience, like understanding or empathy . Corporate use of such "counterfeit people" can manipulate consumers while evading accountability .

Overreliance, emotional attachment, and erosion of autonomy

Users who form emotional connections with human-like AI are at risk of over-reliance and infringements on privacy and autonomy . Research shows that people may disclose more personal information to a humanised chatbot than they would to a clearly mechanical interface . The dynamic of "anthropomorphic inference" leads users to assign social roles to what are fundamentally predictive text systems, transforming statistical outputs into seemingly social responses like advice or companionship .

Distorted moral and legal judgments

Anthropomorphism functions as a fallacy that distorts moral judgments about AI. Users may wrongly assign moral character, status, or responsibility to a machine . When a chatbot appears to have human-like features, people engage in what psychologists call "heuristic processing"—shortcut thinking that invites them to treat the tool as capable of praise, blame, or trust in ways a mere algorithm cannot sustain . This confusion undermines meaningful public discourse and accountability .

Security risks: Manipulation as a vector

From an information security perspective, anthropomorphised AI presents a serious cybersecurity risk. "It's not just an ethical problem; it's also a security problem since anything designed to persuade can make us more susceptible to manipulation," IBM's infosec analysis warns . When it becomes difficult to tell human from machine, people are more likely to trust AI when making sensitive decisions, making them more vulnerable to social engineering scammers who exploit human trust heuristics .

How the language of 'the AI' vs. 'AI systems' fuels the problem

Nature has editorialised explicitly on this point: "We will try to avoid at all costs the use of 'the AI / an AI' due to its unfortunate suggestion of agency. Instead, we will either change to 'the AI system / an AI system' or be very clear what we are talking about" . Treating "AI" as a countable noun (e.g., "the AI decided," "an AI that thinks") subtly implies a unified, self-directed entity with internal mental states. Recasting "AI" as an adjective modifying "system" (or "model," "tool," "algorithm") constantly reminds the audience that the referent is an engineered artefact—not a mind .

This linguistic discipline directly counters what researchers call the "terminological" dimension of anthropomorphic hype: when we say "the AI," the noun form itself becomes a vehicle for ascribing agency, intention, and understanding that the technology does not possess .

The practical takeaway

Anthropomorphising AI is not harmless metaphor. It drives deception, over-trust, manipulation, and distorted accountability. A simple linguistic discipline—using "AI systems" instead of "the AI", and avoiding mental verbs like "thinks," "knows," and "understands" to describe what are actually pattern-matching algorithms —serves as a constant corrective. It keeps the tool-like nature of the technology front and centre, protects users from manipulative design, and strengthens the public's ability to understand what these systems can and cannot do.