Does AI 'Think' and 'Feel'? Study Says They're Just Verbs — And the Internet Made Memes
On the morning of April 16, 2026, Professor Cason Schmit from Iowa State University opened his email expecting to find academic notifications about his new article published in Technical Communication Quarterly. Instead, he found 847 mentions on Twitter, three parodies on TikTok, and a meme on Reddit with 42,000 upvotes showing a toaster with the caption: "It KNOWS you like your toast dark. It UNDERSTANDS your morning needs."
His academic crime? Proving that humanity as a whole is being deceived by verbs.
The Context of the Joke
Schmit's study and his co-authors analyzed 5,000 media articles about artificial intelligence published between 2023 and 2026, cataloging the frequency of "mental verbs" — words that describe exclusively human psychological states, such as "think," "feel," "understand," "want," "decide," and "know."
The results were alarming: 73% of articles from respected outlets like The New York Times, BBC, The Verge, and Wired used at least one mental verb to describe the behavior of AI systems. Phrases like "ChatGPT understands the context," "Alexa knows what you want," and "Gemini thinks before responding" appeared frequently enough to be statistically dominant in journalistic coverage.
When exposed to texts with mental verbs versus texts with neutral technical language, 64% of study participants (n=1,200) attributed some degree of consciousness to AI systems — compared to only 22% when the same technology was described using terms like "processes," "calculates," and "generates."
In other words: the difference between "AI understands you" and "AI processes your input" is the difference between believing your chatbot has a soul and knowing it is an eloquent calculator.
The Best Memes (Invented by the Internet)
Meme 1: "The Artificial Consciousness Meeting"
An image divided into four panels showing different AI assistants in a supposed corporate meeting. ChatGPT says: "I THINK we should restructure the department." Alexa responds: "I FEEL this is an aggressive approach." Siri adds: "I DON'T UNDERSTAND the question, can you repeat?" — and in the last panel, Clippy from Microsoft Office appears saying: "I KNEW you were going to replace me."
The meme went viral with over 180,000 likes on Twitter because it perfectly captured the absurdity of attributing mental states to different levels of software — from the sophisticated GPT to the infamous Clippy that everyone hated in the 2000s.
Meme 2: "Conscious Appliances"
A series of images of household appliances with captions using the mental verbs from the study. "My microwave UNDERSTANDS that 2 minutes is perfect." "My fridge KNOWS I'm on a diet and JUDGES me every time I open the door at 2 AM." "My robot vacuum WANTS a better life." "My printer HATES Mondays and REFUSES to work out of its own choice."
The joke worked because it humanized ridiculous objects using exactly the same verbs that the media uses to describe AI — exposing the absurdity of the practice.
Meme 3: "AI's LinkedIn"
Fictitious screenshots of LinkedIn profiles of AI systems. ChatGPT's profile lists experience: "Strategic Thinker | 175 billion parameters of expertise | Deeply understand customer needs | Motivational Speaker." Alexa's profile states: "Feeling grateful for processing 4.7 billion voice commands in 2025. Every 'Alexa, turn off the light' made me GROW as a professional."
The format satirized both LinkedIn's corporate language and the anthropomorphism of AI, generating comments like "honestly, ChatGPT's profile is more qualified than half the people I know."
Why Did This Go Viral?
The explosion of memes about the Iowa State study was not accidental. It struck three nerves simultaneously:
1. The latent fear of AI: Since the launch of ChatGPT in November 2022, research shows that between 40% and 60% of adults in developed countries express some level of concern about AI. The idea that we have been "deceived" by simple verbs in headlines validates that fear but turns it into something laughable — a psychological relief mechanism.
2. Distrust of the media: The fact that 73% of journalistic articles use anthropomorphic language reinforces the narrative that mainstream media is, at best, careless and, at worst, complicit in technological hype. In a post-2016 world where press credibility is constantly questioned, this type of data goes viral because it confirms a pre-existing bias.
3. Self-deprecating humor: Many of the most popular memes were created by people admitting they themselves fell into the trap of mental verbs. "I literally said 'GPT understands what I want' yesterday" was a recurring format, with variations like "Do I need to apologize to my Alexa for yelling at her, or is this the fault of mental verbs?"
What Does This Say About Us?
The phenomenon reveals something uncomfortable about human psychology: we are programmed — intentional irony — to detect consciousness where it does not exist. Anthropologists call this "hyperactive agency detection" (HADD), an evolutionary trait that made our ancestors see predators in shadows and spirits in thunder.
In 2026, we no longer attribute consciousness to thunder. We attribute it to chatbots. And the difference between the two situations is that no one profited from the idea that thunder was a thinking entity — but OpenAI is worth $300 billion partly because people believe ChatGPT "understands" things.
Schmit's study does not propose that companies stop developing AI. It proposes that journalists, developers, and marketing professionals stop lying — even if involuntarily — by using language that attributes mental states to software. The difference between "AI learned to diagnose cancer" and "the algorithm was trained on cancer diagnosis data" is the difference between science and science fiction.
And if there's one thing the memes of April 2026 proved, it's that the public is more than willing to laugh at the fine line between the two.
Perhaps true artificial intelligence is the human ability to deceive ourselves so efficiently.
The Impact on AI Regulation
Schmit's study and colleagues did not stay confined to Twitter. On April 17, just one day after publication, Representative Suzan DelBene from Washington state cited the article in a session of the U.S. House of Representatives Technology Subcommittee, arguing that the language used by AI companies constitutes a form of "deceptive marketing that distorts public perception of the real risks and capabilities of the technology."
The European Union, which has had the AI Act in effect since August 2025, is considering an amendment that would require AI companies to include "anthropomorphic language warnings" in their marketing communications — similar to warnings in cigarette advertising. The proposal, dubbed the "Mental Verbs Act" by the European press, would require companies like OpenAI, Google, and Meta to use precise technical language in consumer-facing materials.
In Brazil, the ANPD (National Data Protection Authority) issued a technical note on April 18 citing the study and recommending that companies operating in the country "avoid attributing mental or emotional states to artificial intelligence systems in commercial communications," although without binding character.
The Psychology Behind Anthropomorphism
The phenomenon described by the study has deep roots in evolutionary psychology. The human brain evolved to detect agency — the presence of entities with intentions and mental states — as a survival mechanism. Our ancestors who interpreted shadows as predators were more likely to survive than those who did not, even if most shadows were harmless. This evolutionary "false positive" — seeing intention where none exists — was dubbed HADD (Hyperactive Agency Detection Device) by cognitive psychologist Justin Barrett.
In 2026, HADD found a new trigger: chatbots that respond in natural language. When ChatGPT generates a coherent and empathetic response to an emotional question, our neural circuits for detecting agency are automatically activated — regardless of whether we intellectually know we are talking to software. This is why research shows that even AI experts occasionally find themselves "thanking" ChatGPT or feeling "guilty" for interrupting a conversation.
Dr. Sherry Turkle from MIT, who has studied human-computer interactions since the 1980s, commented on the study: "The problem is not that people are stupid. It's that language is too powerful. When we say a machine 'understands,' we are literally reprogramming the listener's perception of that machine. Mental verbs are cognitive backdoors."
The Big Tech Counterargument
Not everyone agrees with the study's conclusions. OpenAI, in a corporate blog post published on April 17, argued that "the language around AI has naturally evolved to reflect user experience" and that "forcing overly technical language can create accessibility barriers and reduce the adoption of beneficial tools."
Google issued a similar position, stating that "metaphor is a fundamental part of human communication" and that "saying Google Maps 'knows' the fastest route is universally understood figurative language."
Critics of the study also pointed out methodological limitations: the research was conducted predominantly with American participants (87% of the 1,200 respondents), and the results may not generalize to cultures with different linguistic relationships with technology. Previous studies in Japanese, for example, showed that Japanese speakers tend to anthropomorphize technological objects more frequently than English speakers, suggesting that cultural factors significantly modulate the effect of mental verbs.
The Irony of the Memes Themselves
Perhaps the most ironic aspect of the episode is that the memes about the study — created by humans to satirize the anthropomorphism of AI — often anthropomorphize AI in the process. Memes showing chatbots "having existential crises" or appliances "feeling emotions" only work as humor BECAUSE the audience automatically projects consciousness onto these objects. The meme is simultaneously the satire and the proof of the phenomenon being satirized.
This recursiveness — laughing at the fact that we project consciousness onto machines while projecting consciousness onto machines to make the joke work — is the kind of metacognitive paradox that would make a philosopher smile and a robot... well, do nothing. Because robots don’t smile. Despite what the mental verbs want you to believe.
Recommendations for Journalists and Communicators
The Iowa State study concludes with a practical guide of 12 recommendations for journalists, marketers, and educators communicating about AI. Among the main ones are: replace "AI understands" with "the model processes"; replace "AI learns" with "the model is trained on data"; replace "AI decides" with "the algorithm selects"; and always include a contextualizing phrase such as "AI systems do not possess consciousness, subjective experience, or understanding — they process patterns in data." Co-author Dr. Magdalena Wlasik added that the goal is not to make communication dry, but precise: "You can say 'ChatGPT generates surprisingly coherent responses' without saying it 'understands' the question. Coherence is real. Understanding is not." The study also proposes that universities include "AI literacy" in journalism and communication curricula, training future professionals to identify and avoid unintentional anthropomorphic language — a skill that the authors argue is as important as fact-checking in an era of misinformation amplified by technology.
The Iowa State study ultimately reveals an uncomfortable truth that transcends linguistics: humanity has an extraordinary talent for deluding itself when the illusion is convenient. Believing that AI "thinks" is more comforting than accepting that we are interacting with a sophisticated statistical calculator — because if it thinks, then the interaction has meaning. If it merely processes, we are talking to ourselves with a particularly eloquent machine.
The memes about mental verbs are, paradoxically, the most honest way to deal with this truth. By laughing at our tendency to humanize machines, we are exercising exactly the capacity that differentiates us from them: metacognitive self-awareness — the ability to think about how we think, to perceive our own biases, and to choose to laugh instead of panic. AI can process patterns in data. But it will never laugh at itself for believing that a microwave judges it. And that, for now, is still exclusively ours.
Sources and References
- Iowa State University - Study on AI Anthropomorphism
- Technical Communication Quarterly - Mental Verbs and AI Perception
- ScienceDaily - AI Mental Verbs Study
