The Ghost in the Machine: How Language Shapes Our Perception of Artificial Intelligence

In the digital age, the lexicon we use to describe the world is evolving at the pace of our technology. When we discuss artificial intelligence, we often find ourselves reaching for words that describe the uniquely human experience: think, know, understand, remember, want. These "mental verbs" are the bedrock of our interpersonal communication, yet when transposed onto lines of code and neural networks, they trigger a psychological phenomenon known as anthropomorphism—the attribution of human traits, intentions, or emotions to non-human entities.

A groundbreaking study recently published in Technical Communication Quarterly sheds light on this linguistic habit, revealing that while we often fear that AI is being "humanized" in the media, the reality is far more complex, nuanced, and perhaps less alarming than previously suspected.

The Anatomy of the Study: Investigating the AI Narrative

The research, titled "Anthropomorphizing Artificial Intelligence: A Corpus Study of Mental Verbs Used with AI and ChatGPT," was led by a team of scholars including Jo Mackiewicz, a professor of English at Iowa State University, and Jeanine Aune, director of the advanced communication program at the same institution. They were joined by Matthew J. Baker of Brigham Young University and Jordan Smith of the University of Northern Colorado.

The team’s objective was to quantify the extent to which journalists—the primary gatekeepers of public information—imbue AI with human agency. To achieve this, they analyzed the "News on the Web" (NOW) corpus, a massive dataset encompassing over 20 billion words from English-language news articles published across 20 countries. By tracking the frequency of mental verbs in conjunction with terms like "AI" and "ChatGPT," the researchers aimed to map the current state of public discourse surrounding machine intelligence.

The Risk of the "Humanized" Machine

The concern driving this research is not merely academic; it is rooted in the practical consequences of public perception. When a news outlet reports that "ChatGPT knows the answer" or "AI has decided to prioritize efficiency," the linguistic implication is that the system possesses consciousness, moral agency, or intentionality.

"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 Professor Jo Mackiewicz. "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 danger, as the researchers highlight, is twofold:

1. Unrealistic Expectations: Attributing cognitive processes like "understanding" to an algorithm can lead the public to overestimate the reliability of AI. If a system "knows," one might trust it implicitly, ignoring the reality that the system is merely performing complex statistical pattern matching.
2. Obfuscation of Responsibility: Perhaps more critically, anthropomorphizing AI shifts the focus away from the human architects behind the curtain. When we say an AI "made a mistake," we implicitly absolve the engineers, developers, and organizations who designed the training data and the oversight protocols.

Findings: The Counter-Intuitive Reality

When the research team began their analysis, they anticipated finding a pervasive trend of anthropomorphism across the news media. However, the data told a different story.

Mental Verbs Are Less Common Than Expected

Contrary to the belief that AI is being constantly "humanized" in the press, the study found that mental verbs are used sparingly in professional news writing. "Anthropomorphism has been shown to be common in everyday speech, but we found there’s far less usage in news writing," Mackiewicz noted.

In the vast landscape of the NOW corpus, the frequency of these terms was surprisingly low. For example, while the term "needs" appeared with AI 661 times, other more distinctly "human" mental verbs were used with remarkably low frequency. For ChatGPT, the most common mental verb pairing was "knows," yet it appeared only 32 times throughout the entire dataset.

The Role of Editorial Standards

The team attributes this relative restraint to the professional norms of journalism. Many news organizations adhere to strict style guides, such as those provided by the Associated Press (AP), which explicitly caution against attributing human characteristics to machines. By maintaining a professional distance, journalists appear to be acting as a buffer against the rampant anthropomorphism that characterizes social media discourse.

Contextual Nuance: When "Need" Doesn’t Mean "Want"

A significant portion of the study focused on the distinction between semantic anthropomorphism and functional usage. Not all mental verbs carry the same weight.

Take, for instance, the word "needs." When a journalist writes, "AI needs large amounts of data," they are not suggesting that the software has a human-like craving or biological necessity. They are describing a technical prerequisite. In this context, the word functions identically to how we might say, "The engine needs oil," or "The recipe needs more salt."

Furthermore, the researchers identified that when "needs" was used, it was frequently paired with the passive voice—"AI needs to be trained" or "AI needs to be implemented." By using the passive voice, the sentence structure subtly forces the reader to consider the human actors who must perform the training or the implementation, thereby re-centering the responsibility on the developers.

The Spectrum of Anthropomorphism

The researchers emphasize that anthropomorphism is not a binary switch but a spectrum. Some instances of language are clearly figurative, while others lean closer to suggesting genuine cognitive autonomy.

"These instances showed that anthropomorphizing isn’t all-or-nothing and instead exists on a spectrum," says Jeanine Aune.

A phrase like "AI needs to understand the real world" occupies a dangerous middle ground. Unlike "AI needs data," the verb "understand" invokes a cognitive state. Even if the author intends it as a metaphor, it suggests an expectation of ethical reasoning or contextual awareness that current Large Language Models do not possess. It is in this gray area where public perception is most vulnerable to being misaligned with technical reality.

Implications for the Future of Tech Communication

As AI continues to be integrated into everything from healthcare to judicial sentencing, the language we use to describe it will become increasingly consequential. The findings of this study offer a roadmap for how we can improve our discourse:

For Journalists and Writers

The research suggests that writers should remain hyper-aware of the mental verbs they deploy. A simple edit—replacing "ChatGPT knows" with "ChatGPT provides data based on"—can profoundly change how the reader interprets the machine’s authority. Transparency in reporting, specifically regarding the limitations of AI, is an essential tool for maintaining public trust.

For Developers and Corporations

Companies building AI systems have a vested interest in how their products are described. While marketing departments may be tempted to use humanizing language to make products feel more "friendly" or "accessible," this study suggests that such language may ultimately create a backlash when the technology fails to meet the "human" standards it was assigned.

Future Research Directions

The research team acknowledges that their study is merely the beginning. Future inquiries must delve deeper into the impact of this language. Does a reader’s trust in AI increase when they encounter anthropomorphic language? Do they hold developers less accountable when a machine is described as "thinking" rather than "calculating"?

Conclusion: Bridging the Gap

The study published in Technical Communication Quarterly serves as a sobering reminder that our language is the primary lens through which we view innovation. While we have largely avoided the trap of consistently characterizing AI as a human-like peer in our professional news, the potential for slippage remains.

"For writers, this nuance matters: the language we choose shapes how readers understand AI systems, their capabilities and the humans responsible for them," Mackiewicz concludes.

As we stand on the precipice of a new technological era, the duty of the communicator is clear. By choosing our words with precision, we can ensure that we view AI not as a ghost in the machine with a mind of its own, but as what it truly is: a powerful, human-designed tool that requires our constant, vigilant oversight. The goal is not to stop talking about AI, but to talk about it in a way that respects the boundaries between silicon and soul.