The Impact of AI on Language Skills: A Call to Action

Vertige (Vertigo) / Léon Spilliaert / 1908

There is a specific kind of panic that sets in when I realise my vocabulary is disappearing. Not the vocabulary I recognise when reading – that stays intact, sometimes even expands. The vocabulary I can actually use. The words that come to me when I need them, in conversation or spontaneous writing, without having to stop and fish around in my mental archive like I am searching for a file I know exists but cannot locate.

The Reality of Vocabulary Loss

This is happening to many people right now. Not just those with neurological conditions or elderly individuals experiencing normal cognitive decline. Young professionals, academics, and writers are all affected. They describe the same phenomenon in slightly different terms: mental fogginess, difficulty finding words, and a sense that their thinking has become slower or less precise. When pressed, they admit they have been using AI extensively for written communication. The correlation is not coincidental.

What is occurring is a neurological process called disuse atrophy. The brain is an expensive organ metabolically. It cannot afford to maintain neural pathways that are not being activated regularly. When I stop engaging in active language production – when I stop constructing sentences from scratch, retrieving words from long-term memory, and building syntactic structures in real time – those pathways weaken. The neurons are still there. The connections between them are not. This is not permanent damage, but it is not trivial either. Rebuilding those connections requires sustained, effortful practice of exactly the kind that AI tools are designed to eliminate.

The Illusion of AI-Generated Language

The process is invisible at first because AI-generated language is fluent. It reads well and sounds professional. I can edit it, approve it, send it, and nobody notices the substitution. But editing is not the same cognitive process as generating. When I edit, I am evaluating something that already exists. I am checking for errors, adjusting tone, and ensuring coherence. These are important skills, but they do not activate the generative networks responsible for creating language from scratch. Those networks require production, not evaluation. And production requires effort – the kind of cognitive load that strengthens synaptic connections through repeated activation.

This explains why people who use AI extensively for writing often report difficulty with spontaneous speech. Spoken language and written language are neurologically linked. They share overlapping networks for lexical retrieval, syntactic processing, and discourse planning. When I weaken one, I weaken the other. The effect is particularly noticeable in professional settings where complex, precise language is required. I find myself pausing mid-sentence, searching for a word I used to access effortlessly, or simplifying my phrasing because the more sophisticated structure I wanted feels just out of reach.

The Broader Implications of AI Use

The problem extends beyond vocabulary. AI tools also change how I approach thinking itself. Language is not simply a vehicle for expressing pre-formed thoughts. It is the medium in which much of my thinking occurs. When I write something out, the act of articulation forces me to clarify vague intuitions, resolve internal contradictions, and discover implications I had not consciously considered. This is why experienced writers often say they do not know what they think until they write it down. The writing is not a transcript. It is the thinking.

When AI does the writing, this clarification process never happens. I start with a rough idea, I get a polished paragraph, and I move on. The idea never undergoes the iterative refinement that comes from struggling to articulate it. It remains at the level of vague intuition, dressed up in fluent prose that gives the illusion of understanding. I can present the idea convincingly because the AI has provided convincing language. But if someone challenges the underlying reasoning, I cannot defend it because I never actually worked through the reasoning myself. I borrowed the AI's articulation without developing my own comprehension.

Cognitive scientists call this the illusion of explanatory depth. People consistently overestimate how well they understand complex systems until they are asked to explain those systems in detail. The act of explanation reveals gaps in understanding that were invisible during passive consumption. AI exacerbates this problem because it provides explanations that sound complete without requiring me to construct them. I read the explanation, I feel I understand, but the understanding is shallow and fragile. It collapses under interrogation.

The Consequences of Outsourcing Thought

This has implications that go well beyond individual productivity. When large numbers of people outsource their thinking to AI, the collective capacity for original thought diminishes. Ideas start to homogenise because everyone is drawing from the same training data. Language becomes predictable because the statistical models favour common patterns over unusual ones. Discourse flattens because the tools are optimised for clarity and professionalism, which means they avoid ambiguity, irony, metaphorical density, and all the other features that make language interesting and thought-provoking.

Metaphor is particularly vulnerable. Metaphorical thinking is how I extend language into new domains. It is how I talk about abstract concepts using concrete imagery, how I map structure from one domain onto another, and how I generate insight by noticing unexpected similarities. AI systems can recognise conventional metaphors because those appear frequently in training data. But they avoid generating novel metaphors because novelty is unpredictable and unpredictability is penalised during training. The result is language that is literally accurate but metaphorically impoverished.

This matters because metaphor is not decoration. It is a fundamental cognitive tool. When I describe time as a river, I am not just making language prettier. I am activating a conceptual mapping that allows me to reason about time using spatial logic. The metaphor scaffolds thought. Without it, certain kinds of reasoning become much harder. And when AI trains me to avoid metaphorical language because it is risky or ambiguous, it is training me to think in narrower, more literal ways.

The Decline of Stylistic Variation

The same applies to stylistic variation. AI output tends toward a neutral, professional register because that is the safest choice across contexts. But language has many registers, each suited to different purposes. Academic language, legal language, poetic language, colloquial language, technical jargon, ironic understatement – these are not interchangeable. They encode different relationships between speaker and audience, different epistemological stances, and different social identities. When I stop practising these registers because AI defaults to a single neutral mode, I lose the ability to code-switch effectively. My language becomes functionally competent but stylistically monotonous.

This is already visible in professional communication. Emails sound increasingly similar across industries and contexts. The distinctive voice that used to signal personality or expertise has been smoothed away. Everyone sounds vaguely corporate, vaguely friendly, and vaguely professional. The homogenisation is efficient, but it is also aesthetically impoverishing. And it has cognitive costs. Stylistic variation is not superficial. It reflects the ability to adopt different perspectives, to modulate tone for different audiences, and to signal nuance through linguistic choice. When that ability atrophies, communication becomes less flexible and less expressive.

The Challenge for English Learners

The decline is particularly acute for those learning English as an additional language. Advanced learners often reach a plateau where they are functionally fluent but cannot progress further. They can handle everyday communication and professional tasks, but they struggle with complex abstract discussion, nuanced argumentation, and creative language use. AI tools seem like a solution because they provide immediate access to sophisticated language. But they actually entrench the plateau. The learner becomes dependent on the tool for producing anything beyond basic communication. Their active vocabulary stops expanding. Their grammatical range stops diversifying. They remain permanently intermediate, capable of editing AI output but incapable of generating equivalently complex language independently.

This is what educators call learned helplessness in a linguistic context. The tool becomes a crutch. The user stops believing in their own capacity to produce sophisticated language without assistance. And because the tool is always available, there is no pressure to develop that capacity. The result is functional dependency – not inability to communicate, but inability to communicate at a high level without technological support.

Breaking the Dependency

Breaking this dependency requires deliberate practice of precisely the skills that AI has made optional. I need tasks that force active language production without external scaffolding. I need prompts that cannot be answered with cached phrases or formulaic responses. I need cognitive challenges that are genuinely difficult, that create the kind of productive struggle which drives neural adaptation.

This is why randomized, constraint-based tasks are particularly effective. When I am forced to connect two unrelated concepts, I cannot rely on pre-existing explanations. I have to construct a bridge through active reasoning. When I am given three random words and asked to build a narrative, I cannot fall back on standard story templates. I have to improvise structure in real time. When I encounter an ambiguous stimulus and have to interpret it without guidance, I cannot defer to authoritative sources. I have to generate meaning independently.

These tasks are cognitively demanding. They require high working memory load, flexible retrieval from long-term memory, and real-time integration of disparate elements. This is exactly the kind of cognitive work that strengthens the neural networks responsible for creative language use. And it is exactly the kind of work that AI tools allow people to avoid.

The Importance of Generative Practice

The training does not need to be lengthy. Research on cognitive skill acquisition suggests that even brief periods of high-intensity practice can produce measurable improvements if the practice is sufficiently challenging and occurs regularly. Twenty minutes of genuine cognitive effort – the kind where I am forced to generate solutions rather than evaluate pre-made ones – is more effective than two hours of passive review or AI-assisted editing.

The key is that the practice must be generative rather than receptive. Reading sophisticated language improves comprehension but does not improve production. Editing AI-generated text improves evaluation skills but does not improve generation. I have to actually produce language under challenging conditions to maintain productive capacity. There is no shortcut. There is no way to outsource this and still retain the underlying skill.

Establishing Boundaries with AI Tools

This does not mean AI tools are inherently harmful. They are useful for specific purposes: catching grammatical errors, suggesting alternative phrasings, and providing information quickly. The problem is not the tools themselves but the way they are being used – as replacements for thinking rather than supplements to it. When AI becomes the default for any task involving language, when I reach for it automatically rather than trying to solve the problem myself first, that is when cognitive atrophy begins.

The solution is not to abandon the tools but to establish boundaries around their use. I can use AI for refinement after I have done the generative work myself. I can use it to check my grammar, not to write my sentences. I can use it to explore alternative perspectives after I have articulated my own position. I should treat it as a second opinion rather than a first draft. This keeps the cognitive load on me while still benefiting from the tool's capabilities.

Balancing Efficiency and Capacity

The broader point is that efficiency and capacity are often in tension. Optimising for efficiency – getting things done faster with less effort – can undermine the development and maintenance of capacity. Skills require practice. Practice requires effort. When I eliminate effort in the name of efficiency, I eliminate the condition necessary for skill maintenance. This is true for physical skills, and it is equally true for cognitive ones.

The concern is not that AI will make humans obsolete. The concern is that humans will make themselves less capable by delegating too much of their own cognition. The atrophy is voluntary. It is reversible. But it requires recognition that the problem exists and deliberate action to counteract it. Platforms like Grandomastery exist specifically to provide this kind of counter-pressure – structured environments where cognitive effort is required, where AI cannot do the work for me, and where the only way forward is through genuine mental exertion.

The Stakes of Cognitive Capacity

The stakes are not abstract. This is not about preserving outdated skills for sentimental reasons. This is about maintaining the cognitive capacities that allow humans to think originally, to reason flexibly, and to communicate with precision and creativity. These capacities do not disappear overnight. They erode gradually, almost imperceptibly, as people make hundreds of small decisions to let AI handle tasks they used to do themselves. Each decision seems rational in isolation. Cumulatively, they produce a population that is functionally literate but creatively impoverished, capable of consuming and evaluating language but increasingly unable to generate it.

That is the trajectory we are on. And it is not inevitable. But changing it requires more than good intentions. It requires structured practice, deliberate effort, and a willingness to struggle with problems that do not have easy solutions. The brain adapts to the demands placed on it. If I want to remain capable of complex, creative thought, I need to keep placing those demands on myself.