By Iffa Jayyana 
In the digital age, the lexicon we use to describe the world is shifting to accommodate a new kind of inhabitant. We talk of machines that "think," algorithms that "know," and neural networks that "understand." While these terms are convenient shorthand for complex computational processes, a growing body of academic research suggests that this linguistic habit—known as anthropomorphism—may be subtly distorting our understanding of what artificial intelligence actually is, and more importantly, what it is not.
A recent study published in Technical Communication Quarterly, titled "Anthropomorphizing Artificial Intelligence: A Corpus Study of Mental Verbs Used with AI and ChatGPT," has peeled back the curtain on how media outlets frame these technologies. Led by a team of researchers from Iowa State University, Brigham Young University, and the University of Northern Colorado, the study argues that by imbuing software with human-like mental states, we risk creating a dangerous disconnect between public perception and technical reality.
The Mirage of Sentience: Why Mental Verbs Matter
At the core of the study is an examination of "mental verbs"—words like think, know, understand, believe, and want. When applied to humans, these words describe conscious, subjective experiences. When applied to a Large Language Model (LLM) or a machine learning algorithm, they are technically metaphors.
However, Jo Mackiewicz, a professor of English at Iowa State University and lead author of the study, warns that the metaphor often becomes a cognitive trap. "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," Mackiewicz notes. "It helps us relate to them. But at the same time, there’s a risk of blurring the line between what humans and AI can do."
The danger, according to the research team, is twofold. First, it fosters a false sense of autonomy. When a headline proclaims that "AI decided to optimize its output," it implies a degree of independent agency or intention that simply does not exist in code. AI systems operate by calculating statistical probabilities and pattern recognition; they do not possess beliefs, desires, or a moral compass. Second, this language obscures the human element. By treating the machine as an entity with "intentions," we inadvertently hide the developers, engineers, and corporate stakeholders who hold the ultimate responsibility for the system’s design, training data, and subsequent outputs.
A Deep Dive into the Data: The Methodology
To determine if the media is truly guilty of over-anthropomorphizing AI, the research team—which included Jeanine Aune, director of the advanced communication program at Iowa State; Matthew J. Baker of Brigham Young University; and Jordan Smith of the University of Northern Colorado—conducted a massive linguistic audit.
They turned to the News on the Web (NOW) corpus, a colossal dataset comprising over 20 billion words derived from English-language news articles published across 20 countries. This provided a panoramic view of how journalists, pundits, and technical writers characterize AI in the wild. The team specifically tracked the frequency of mental verbs in proximity to keywords like "AI" and "ChatGPT."
The Unexpected Verdict: Restraint in the Newsroom
The findings of the study were, in many ways, surprising. The researchers hypothesized that they would find rampant anthropomorphism, given how common these figures of speech are in casual conversation. Instead, they discovered a surprising level of editorial restraint.
"Anthropomorphism has been shown to be common in everyday speech, but we found there’s far less usage in news writing," Mackiewicz observed. The study suggests that institutional norms, such as the Associated Press style guidelines, which explicitly caution journalists against attributing human emotions or traits to non-human entities, are effectively curbing the impulse to "humanize" the technology.
While "needs" was the most common verb associated with AI—appearing 661 times in the dataset—its usage was frequently functional rather than sentient. Phrases like "AI needs large amounts of data" are structurally similar to saying "a car needs gasoline" or "a recipe needs salt." In these contexts, "needs" denotes a requirement for operation, not a psychological craving.
The Nuance of Language: Context as a Shield
The study’s most compelling finding is that the presence of a mental verb does not automatically constitute anthropomorphism. Language is a spectrum, and context is the deciding factor.
The Spectrum of Humanization
The researchers identified a sliding scale of intent:
- Functional Necessity: "AI needs more processing power." (Neutral, non-anthropomorphic).
- Passive Agency: "AI needs to be trained by experts." (Shifts responsibility back to the human, non-anthropomorphic).
- Pseudo-Cognition: "AI needs to understand the real world to be effective." (Borderline; implies a capability of "understanding" that AI currently lacks).
The third category is where the researchers urge caution. When writers suggest that a system must "understand" or "know" something to be successful, they are creating a narrative expectation of human-level reasoning. This sets the stage for public disillusionment when the technology inevitably fails to display human-like ethics or common sense.
Implications for the Future of Technology Communication
As we move deeper into the era of generative AI, the way we describe these systems will have profound consequences for policy, regulation, and public trust.
1. Shifting Responsibility
When we treat AI as an autonomous agent that "knows" or "decides," we inadvertently absolve the creators of their duty to ensure safety and transparency. If a system is described as having "intent," the blame for harmful outputs might be misplaced on the machine itself rather than the biased training data or the flawed architecture curated by humans.
2. Managing Expectations
Unrealistic expectations are the primary fuel for the "hype cycle." If the public believes AI can "think," they are more likely to trust it with sensitive decisions—such as medical diagnoses or legal sentencing—that require genuine judgment. The research team emphasizes that precise language is a safeguard against this dangerous over-reliance.
3. A Call to Action for Practitioners
The study serves as a primer for technical writers, journalists, and corporate communicators. By shifting the focus from the machine’s "mind" to its "function," communicators can provide a more accurate representation of the technology. Instead of writing "ChatGPT knows the answer," writers might opt for "ChatGPT generated a response based on its training data." It is a subtle shift, but one that preserves the distinction between computation and cognition.
Conclusion: Mind Your Language
The research led by Mackiewicz and Aune confirms that while news outlets have shown a commendable degree of caution in describing AI, the risk of anthropomorphism remains a live issue. As AI becomes more sophisticated and better at mimicking human conversation, the temptation to use "mental verbs" will only increase.
For those tasked with documenting the evolution of artificial intelligence, the message is clear: language is not merely a tool for description; it is a tool for framing reality. By choosing our words with care, we can ensure that society views AI for what it is—a transformative, powerful, and complex tool—rather than a sentient entity that, in reality, does not exist. The ghost in the machine is, ultimately, just a reflection of the humans who built it.
Ammar Sabilarrohman 
Ali Ikhwan 