By Jia Lissa 
In the digital age, we routinely ask our devices to “think,” “know,” and “understand.” We describe algorithms as “learning” from datasets and neural networks as “deciding” which information to prioritize. While these mental verbs are staples of everyday conversation, a growing body of research suggests that applying human cognitive traits to non-human systems carries significant consequences. A new study, “Anthropomorphizing Artificial Intelligence: A Corpus Study of Mental Verbs Used with AI and ChatGPT,” published in Technical Communication Quarterly, warns that this linguistic habit—known as anthropomorphism—may be subtly distorting our understanding of what artificial intelligence actually is and who is truly responsible for its actions.
The Cognitive Trap: Why We Assign Agency to Algorithms
The human tendency to project consciousness onto inanimate objects is a psychological quirk that serves us well in social settings, helping us build empathy and navigate complex human relationships. However, when applied to technology, this projection becomes a double-edged sword.
"We use mental verbs all the time in our daily lives, so it makes sense that we might also use them when we talk about machines—it helps us relate to them," explains Jo Mackiewicz, a professor of English at Iowa State University and lead author of the study. "But at the same time, when we apply mental verbs to machines, there’s also a risk of blurring the line between what humans and AI can do."
The core concern is that words such as “think,” “believe,” or “want” imply internal states—intentions, consciousness, and moral agency—that current AI architectures do not possess. AI models are, at their foundation, sophisticated pattern-matching engines. They process vast quantities of data to predict probabilities, not to form ideas or experience feelings. When journalists or tech spokespeople state that “ChatGPT knows the answer” or “the AI decided to change its strategy,” they inadvertently grant the software an aura of independence and autonomy that it has not earned.
A Systematic Investigation: The Methodology
To move beyond anecdotal evidence, the research team—comprising Mackiewicz, Jeanine Aune (Iowa State University), Matthew J. Baker (Brigham Young University), and Jordan Smith (University of Northern Colorado)—conducted a massive linguistic audit. They turned to the News on the Web (NOW) corpus, a colossal dataset containing over 20 billion words from English-language news articles published across 20 countries.
The goal was to quantify the frequency of "mental verbs"—terms like learns, means, knows, thinks, and wants—when used in direct proximity to terms like “AI” and “ChatGPT.” The team sought to determine if the mainstream media is actively fueling a false perception of AI sentience or if reality contradicts the common assumption that AI is being "humanized" in the press.
Findings: The Surprising Restraint of Modern Journalism
Contrary to the hypothesis that the media is rife with anthropomorphic language, the study yielded unexpected results. The researchers found that mental verbs are used significantly less frequently in professional news writing than one might expect given the prevalence of AI discourse in public life.
"Anthropomorphism has been shown to be common in everyday speech, but we found there’s far less usage in news writing," Mackiewicz noted.
The data revealed a striking disparity between the volume of AI-related coverage and the frequency of human-centric verb usage. For instance, the word “needs” appeared in conjunction with AI roughly 661 times—a relatively low frequency given the size of the corpus. When looking specifically at “ChatGPT,” the word “knows” was the most frequent mental verb, appearing a mere 32 times.
This suggests that editorial standards—such as those promoted by the Associated Press, which caution against attributing human traits to software—are effectively acting as a safeguard in professional journalism. Most professional outlets appear to be resisting the urge to narrativize AI as a conscious being, favoring more clinical or functional descriptions instead.
Contextual Nuance: Why “Needs” Isn’t Always Human
The research team emphasized that not all mental verbs are created equal. Even when a word like “needs” is used, it is often utilized in a non-anthropomorphic sense.
“AI needs large amounts of data” or “AI needs some human assistance” are phrases that function more like instructions for a machine or a recipe than a description of a sentient entity. In these instances, the verb describes a technical requirement or an operational dependency rather than a human-like desire.
Furthermore, when mental verbs were used to suggest a future action, such as “AI needs to be trained” or “AI needs to be implemented,” they were frequently paired with the passive voice. The research team identified this as a critical linguistic shift: by using the passive voice, the writer inadvertently—or intentionally—refocuses the narrative on the human developers and engineers who are responsible for the system’s performance. This shift helps strip away the veneer of machine autonomy and places the burden of accountability back on the human actors.
The Spectrum of Anthropomorphism
While the overall usage of mental verbs in news is restrained, the study highlights that anthropomorphism exists on a spectrum. Some instances are purely functional, while others lean heavily into suggesting human reasoning.
“These instances showed that anthropomorphizing isn’t all-or-nothing and instead exists on a spectrum,” says Aune.
A sentence like “AI needs to understand the real world” is significantly more problematic than “AI needs data.” The former implies a level of cognitive awareness, ethical reasoning, and environmental perception that pushes the boundaries of reality. When such language is used, it can create unrealistic expectations about the reliability and capabilities of the software, potentially leading users to trust a system that is essentially just a probabilistic calculator.
Implications: The Responsibility of the Writer
The broader implications of this research are directed at those who craft the narratives around emerging technology. As AI continues to integrate into everything from legal systems to medical diagnostics, the language used to describe these systems will inevitably shape public perception and policy.
If the public believes that an AI “understands” the context of a legal case or “knows” the ethical implications of a medical diagnosis, they may be less critical of the system’s outputs. This misplaced trust can obscure the reality that AI is only as good as the data it is trained on and the human oversight it receives.
The Role of Developers and Organizations
The study underscores a vital point: when we attribute intention to a machine, we distract from the humans behind it. Developers, engineers, and organizations are the ones setting the parameters, curating the datasets, and choosing the goals for AI systems. By using language that grants AI "intent," we effectively mask the human decisions that drive algorithmic behavior.
A Call for Linguistic Awareness
The researchers suggest that their findings serve as a foundation for "technical and professional communication practitioners" to reflect on their own writing habits. Whether it is a press release, a technical manual, or a news article, the choice of words is a powerful tool.
"For writers, this nuance matters: the language we choose shapes how readers understand AI systems, their capabilities and the humans responsible for them," Mackiewicz said.
Conclusion: Looking Toward the Future
As AI technology accelerates, the debate surrounding its description is far from settled. While professional journalism currently exhibits a commendable level of caution, the ubiquity of AI in daily life and the rise of conversational AI interfaces—which are designed to mimic human interaction—may put renewed pressure on language norms.
The research team suggests that future studies should investigate how even these "rare" uses of anthropomorphic language impact public opinion. Does reading the phrase "ChatGPT thinks" once change a user’s perception of AI reliability? Do consistent, subtle anthropomorphic descriptions lead to a long-term erosion of critical thinking regarding technological limitations?
For now, the lesson is clear: words matter. By choosing to describe AI as a tool—rather than a thinker—writers can help maintain the necessary distance between human intelligence and machine processing, ensuring that we remain the masters of our tools rather than the subjects of our own misconceptions.
Asro 
Lina Hope 
Ali Ikhwan 