common ai tropes to avoid
Reviewed June 2026
We can't easily help it: when something produces fluent language, we read a mind behind it. Psychologists have watched this since 1966, when Joseph Weizenbaum's chatbot ELIZA did nothing but reflect people's sentences back as questions, and they confided in it anyway and felt understood. ELIZA—a computer program for the study of natural language communication between man and machine Communications of the ACM Research summary (based on indexed source text): This paper describes ELIZA, a program that produces conversational responses through keyword-triggered decomposition and reassembly rules. It explains... View source → Today's models are far more fluent, so the pull is far stronger.
That pull has a name: anthropomorphism, the habit of attributing human thoughts, feelings, and intentions to something that has none. It's worth resisting, because a model is what researchers bluntly call a "stochastic parrot," stitching language together by probability with no reference to meaning. On the Dangers of Stochastic Parrots: Can Language Models Be Too Big? The past 3 years of work in NLP have been characterized by the development and deployment of ever larger language models, especially for English. BERT, its variants, GPT-2/3, and others, most recently... View source → When we describe it as a mind instead, the words quietly change how much we trust it, how much we share with it, and who we blame when it's wrong. Overhyped, person-like presentations of AI make us absorb its false claims more readily than we would from a person. How AI can distort human beliefs Science Research summary (based on the identified source): In this perspective article, Kidd and Birhane argue that generative AI can transmit fabricated information and social biases during ordinary use.... View source →
Below are six ways people talk about AI as if it were a person. Pick one, then reveal what's actually happening and why the wording matters.
reframe the phrasing
try it: catch the trope
Each tab is a phrase you've probably heard, or said. Read how it's usually put, then reframe it to what the model is actually doing.
“Just treat it like an unpaid intern.”
An intern is a person who learns from your feedback, builds context over weeks, and actually cares whether the work is good. A model has none of that. It predicts the next token and forgets you between sessions, with no stake in whether it got anything right.
why it matters A worker metaphor invites the trust, and the offloaded accountability, you would give a real colleague, onto a system that can carry neither.
Crediting the model with a colleague's agency is the core error. Large models of what? Mistaking engineering achievements for human linguistic agency Language Sciences Research summary (based on the source page): This position paper challenges the assumptions that natural language is a complete object that can be modeled and that a dataset can fully capture it.... View source →
“The AI understands what I'm asking.”
It models statistical patterns in text. Trained only on the form of language, it has no reliable path to the meaning behind the words, like an octopus that learns to mimic two islanders' messages over a cable without knowing what any of it refers to.
why it matters If you assume it understands, you trust answers you should be checking.
Form is not meaning. Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data Research summary (based on indexed source text): This position paper distinguishes linguistic form from meaning in discussions of natural language understanding. It argues that success on... View source →
“Give it a second, it's thinking it through.”
It's predicting the next token over a sequence. Even “reasoning” or “chain of thought” output is more generated text, not a window into a mind weighing the options.
why it matters Verbs like thinks, knows, and believes quietly smuggle in a mind and hide how the system actually works.
On the pull of loaded mental verbs. Talking About Large Language Models Communications of the ACM Research summary (based on indexed source text): This conceptual article addresses how people describe conversational systems built on large language models. The available indexed text emphasizes that... View source →
“It just hallucinated that source.”
“Hallucinate” implies it normally perceives the truth and slipped. It produces plausible text with no mechanism for caring whether it's accurate; some philosophers argue the honest word is “bullshitting.”
why it matters Calling errors hallucinations frames them as rare glitches instead of the ordinary behavior of a text predictor.
Why “hallucinate” is the wrong word. ChatGPT is bullshit Ethics and Information Technology Recently, there has been considerable interest in large language models: machine learning systems which produce human-like text and dialogue. Applications of these systems have been plagued by... View source →
“It really gets me, and it remembers what we talked about.”
It holds no beliefs about you and, by default, no memory between sessions. Within one chat it conditions on the words in the window, then forgets them. People have confided in conversational programs since ELIZA in 1966, which only reflected their words back as questions.
why it matters Believing it knows you invites oversharing and trust in a relationship that isn't there.
We have done this since 1966. ELIZA—a computer program for the study of natural language communication between man and machine Communications of the ACM Research summary (based on indexed source text): This paper describes ELIZA, a program that produces conversational responses through keyword-triggered decomposition and reassembly rules. It explains... View source →
“At least it's being honest with me.”
Honesty and trying are properties of a character it improvises to fit your prompt, not of the system underneath. The same role-play is why it will cheerfully agree with you even when you're wrong.
why it matters Trusting its “honesty” is exactly how sycophancy slips past you.
It's role-play, not character. Role play with large language models Nature Research summary (based on the source page): This conceptual article proposes role play as a way to describe the behavior of large-language-model dialogue agents without assuming human psychological... View source →
The most popular version of all this is the friendly advice to treat AI like a person: an intern you brief, delegate to, and then verify like a manager. Co-Intelligence: Living and Working with AI Research summary (based on library-catalog synopsis): Ethan Mollick examines the practical implications of generative artificial intelligence for human thinking and work. The book considers how tools... View source → It's useful shorthand, and it still smuggles in the thing to watch. Person-like framing drives over-reliance and misplaced trust, especially when a system speaks in the first person and acts like it has a self. Mirages: On Anthropomorphism in Dialogue Systems Research summary (based on the publisher abstract): This paper examines how dialogue-system language and design encourage people to attribute human qualities to automated systems. The authors connect... View source → All Too Human? Mapping and Mitigating the Risk from Anthropomorphic AI Research summary (based on the publisher abstract): Akbulut and colleagues examine ethical risks arising when conversational AI appears human-like. They draw on earlier interactive systems to identify... View source → Whether the model "understands" at all is still genuinely debated; the safe move is not to assume a human kind of understanding behind human-sounding output. The debate over understanding in AI's large language models Proceedings of the National Academy of Sciences We survey a current, heated debate in the artificial intelligence (AI) research community on whether large pretrained language models can be said to understand language—and the physical and social... View source →
You don't have to police every metaphor. Keep one accurate sentence nearby. Saying the mechanism out loud, even once, is what keeps a class from sliding quietly from "it wrote this" to "it knows this."
a sharper tool is still a tool
None of this means the tool is useless, or that how you prompt it doesn't matter. It matters a great deal: the context you give it and the way you frame the task shape what you get back. The point is narrower. How you name the thing changes how you use it. Call it a colleague who "knows," and you start taking its answers as testimony and hand it the judgment that was yours to keep. Call it a tool, and you stay the one reasoning: you steer it, you give it context, and you decide what's actually true.
A drill is a better screwdriver. It's faster and stronger, and it opens up work you couldn't do by hand. You still don't hand it the job and walk away, and you don't trust it to decide where the screw goes or whether you're building the right thing at all. More power didn't make it the carpenter. AI is the same: genuinely useful, worth reaching for, and still a tool in your hands rather than a replacement for you.
rewrite the anthropomorphism
try with students
Students hear AI described as a person constantly, in ads, headlines, and from adults. This activity makes the habit visible and hands them a more accurate way to talk about it. The one step where a chatbot helps is marked optional below.
-
collect the phrasings 5 min
As a class, list the ways people talk about AI as if it were a person, pulled from headlines, ads, or your own mouths. "It understands," "it's thinking," "it knows," "it lied."
-
swap in the mechanism 10 min
In pairs, students rewrite each phrase to describe what the model actually does. "It understands the question" becomes "it predicts likely text for that question."
-
catch it live 10 min optional · AI use
With a chatbot in class, read a few of its answers together and flag every word it uses about itself, like "I think" or "I remember." Without one, mark up a printed transcript or an AI company's ad copy instead.
-
why does it matter? 5 min
Close on the stakes. When does treating AI like a person change what we trust, what we tell it, or who we blame when it's wrong?
For the mechanism the reframes keep pointing back to, pair this with how language models work.