Abstract
Picture a crowded lecture hall where students once scribbled notes in the margins of textbooks, a bustling café where friends debated ideas late into the night, or even a family dinner table where stories and traditions passed from one generation to the next. Each of these moments was shaped by the dominant tools of its era. The printing press expanded access to knowledge, the telephone and radio reshaped conversation, social media transformed public discourse, and now large language models (LLMs) are beginning to influence not only how we communicate but also how we think. Today, many conversations begin not with pen or voice, but with predictive text, autocomplete, or artificial intelligence (AI)-generated drafts—evidence that machines are no longer just recording our words, but helping to shape them. ChatGPT reached over 100 million users within months, setting a record for the fastest-growing consumer application in history. 1
The printing press standardized spelling and grammar and democratized knowledge. The telegraph and telephone collapsed geographic distances and created new norms of brevity. Radio and television introduced mass communication, shaping entire generations with shared messages. The typewriter and later the word processor standardized business and academic style. Social media compressed language into hashtags, memes, and micro-messages. Now, LLMs are accelerating change again, this time in ways that blur authorship, challenge individuality, and insert algorithmic prediction into the very fabric of expression.
To better understand how LLMs may reshape our linguistic landscape, we can examine three core principles that have historically guided changes in human communication.
Principles of Change
The first is the principle of least effort, which suggests that humans simplify language whenever possible—shortening words, dropping syllables, and eliminating redundancy. LLMs mirror this tendency. Their predictive patterns often favor compressed or familiar phrasing, reinforcing the natural pull toward efficiency. This can be beneficial in some cases, reducing friction in communication and helping those who struggle with language barriers. Yet if machines consistently privilege what is simplest, we risk losing the richness of expression that complexity and nuance provide. Metaphors, jokes, and regional idioms can convey meaning far more powerfully than stripped-down sentences. When efficiency becomes dominant, individuality and cultural color can fade. 2 Recent analyses confirm these broader linguistic shifts: conversational engagement patterns in British English changed measurably between 1994 and 2014. 3
The second is the principle of density, which holds that our language adapts to the communities we inhabit. Just as humans begin to sound like their colleagues, friends, or families, LLMs echo the linguistic habits of their training data. This mirroring explains why AI-generated text often carries the idioms, rhythms, and preferences of the datasets it was fed. When millions of people receive similar AI-generated phrasing, subtle convergence may occur. A teacher grading essays, a manager reviewing reports, and a journalist reading drafts may notice the same balanced, somewhat bland style. Over time, feedback loops could standardize phrasing globally, eroding linguistic diversity and flattening discourse. This was demonstrated in a Max Planck Institute study, which analyzed more than 740,000 hours of spoken content from podcasts and YouTube and found measurable upticks in AI-favored words after ChatGPT’s release. 4
The third is the principle of imitation. Humans have always adopted the expressions of those they admire—poets, politicians, artists, or celebrities. LLMs magnify this effect, since their very function is to imitate, and to do so at scale. They not only replicate admired turns of phrase but can generate entire styles that people adopt as their own. In this way, technology does not merely follow human language, but begins to lead it. The danger is subtle: when imitation crosses into influence, the creativity of human voices may be overshadowed by the predictability of machine-generated ones.
Benefits and Risks
LLMs also bring undeniable benefits to communication. They can generate outlines, assist with translation, and support collective intelligence by synthesizing diverse inputs at unprecedented scale. 5 For students, they offer scaffolding for essays and projects; for patients, they can provide language to describe symptoms more clearly; for professionals, they can anticipate counterarguments and strengthen analysis. These advantages democratize access to polished expression. Someone with limited fluency in English, for example, can now write with greater clarity and confidence. In these cases, efficiency does not erase creativity but amplifies it—provided we remain conscious of the limitations embedded in the systems.
Yet the risks of homogenization are significant. AI-generated text gravitates toward neutral, balanced formulations, often recycling favored terms like “robust,” “seamless,” or “holistic.” This tone can be useful for professional writing but may also flatten individuality. If student essays, corporate reports, and even political speeches begin to share the same stylistic fingerprints, language risks becoming monotonous. Creativity thrives on deviation, but divergence is exactly what predictive models tend to smooth out. The long-term cultural consequence could be a narrowing of discourse, with fewer surprising, idiosyncratic, or subversive voices breaking through. In email, smart replies now suggest “Thanks!” or “Sounds good” before we even think of them—subtle shifts that Hohenstein and colleagues showed make senders appear less cooperative and personable. 6
Health Care and Future Scenarios
Health care illustrates both the promise and the pitfalls of this linguistic shift. Patients increasingly use AI tools to draft medical histories or describe symptoms, sometimes borrowing the formalized tone of electronic health records. Clinicians, too, are beginning to experiment with AI for charting and correspondence. While this can save time and improve clarity, it can also make communication less personal. A patient’s unique voice, with its hesitations and emotional nuances, may be flattened into standardized phrasing. For some, this offers empowerment and efficiency; for others, it risks alienation. Ultimately, medicine is not only about accurate records but about empathy, and how AI mediates that balance will be critical.
Patients also increasingly share AI-assisted drafts in peer support groups, where technical phrasing can both help validate experiences and sometimes distance emotional connection. Clinicians note that templated AI summaries are efficient for records but can obscure subtle cues of distress or coping. Training both patients and providers to use these tools without losing empathy will be essential.
Language does more than transmit information—it shapes cognition. Psychologists and linguists have long argued that the words we use influence how we categorize and perceive the world. If LLMs nudge us toward certain shortcuts, they may also guide thought in subtle ways. Some early evidence already suggests that reliance on AI drafting tools changes how people approach problem-solving, encouraging acceptance of suggested phrasing rather than exploration of new ones. In education, this could mean students learning to summarize rather than to question; in civic life, it could mean public debate becoming more scripted, less spontaneous. These cognitive shifts may not be dramatic at first but could accumulate over time.
Looking ahead, we may see the emergence of AI dialects—styles optimized for medicine, law, or customer service. These could improve efficiency within fields while further narrowing expression. Humor, irony, and regional color, already difficult for LLMs, could become rarer in public discourse. Alternatively, as AI-speak becomes more visible, human voices that resist it may stand out more sharply, valued precisely for their originality. This tension—between homogenization and differentiation—will define the cultural impact of AI language in the coming decade.
In health care, for example, an AI dialect could enforce consistency in describing symptoms and treatments, improving clarity but reducing personalization. In law, AI could reinforce rigid formats, narrowing rhetorical creativity. In academia, AI assistance may encourage balance and caution but also suppress bold claims. These scenarios illustrate the double-edged nature of linguistic automation—streamlining communication while potentially suppressing innovation.
It is important to note that fears of new technologies altering language are not new. Socrates worried that the invention of writing would weaken memory. Critics of the typewriter lamented the loss of penmanship. Each innovation brought change, but creativity endured. The difference today is speed and scale. What once took generations now occurs within a few years. That acceleration gives us less time to adapt socially, culturally, and cognitively. It also means that humans must play a more deliberate role in guiding how these technologies are integrated. LLMs should be seen not as replacements for human communication, but as collaborators—tools that require active oversight to ensure they enrich rather than impoverish our language.
Language is always evolving, but in just a few years, LLMs have accelerated and concentrated that process. Imitation may echo AI, but influence should expand our horizons. The challenge is to embrace the efficiency and reach of LLMs without surrendering the creativity, nuance, and individuality that define human communication. As Swan reminds us, human writing remains essential precisely because it conveys creativity, nuance, and individuality that AI cannot replicate. 7 The future of language will not be written by machines alone—but by the choices we make in how to use them.
Brenda K. Wiederhold,Editor-In-Chief