In the AI era, language is infrastructure, not code
In earlier technological eras, words explained the system. Now words have become the system. That should change who sits at the table when machines are built.
For as long as we have built machines, words merely explained them. That has changed. Today's technological leap runs on language itself. Artificial intelligence, and the large language models behind generative AI, are built entirely on words, with no inherent grasp of context or emotion. What a model "knows" comes from the data it is trained on, and that data carries our stereotypes and biases, all of them encoded in language. Feed it "workforce optimization" and it reads efficiency; feed it "employee displacement" and it reads harm. The facts are identical. Only the words have moved. Words no longer describe the system. They are the system.
Earlier, the best systems were won on scale and speed, on the belief that more computing meant better outcomes. LLMs break that rule. Their performance turns on linguistic precision, not volume. And language is never neutral. A small semantic shift changes how a model frames a problem and what it recommends. "Consumer engagement" and "behavioural surveillance" can describe the same activity on identical facts, yet one invites investment and the other a lawsuit. How information is framed now decides how well a machine reasons. Words have become its operating terms.
This is not about an individual coaxing a better answer from a chatbot. It runs deeper. Training a model is institutional work. Information is classified and tagged, and those choices become the machine's foundation for reading the world. The labels an organization assigns, safe or risky, trustworthy or harmful, productive or wasteful, harden into datasets, interfaces and workflows. They teach the machine what to notice, what to ignore, how to read human behaviour and what to produce by default. The future influence of technology leaders will rest on how well they shape these semantic systems, not on technical know-how alone.
It is time organizations rethought where communication sits. For fifty years it has been treated as downstream work, a voice given to ideas already settled behind closed doors. Its job was the faade: to gloss the presentation, to add the face-lift once the strategy was done. AI collapses that order. Communication is no longer symbolic; it is infrastructural. Semantics no longer shape only human perception; they govern how the machine interprets. A careless taxonomy can distort a model's output, and through it, people's sense of what is true. An ethically thin classification can normalize harm, then scale it. As AI enters hiring, healthcare, finance, governance and education, these word choices stop being cosmetic. They start carrying material, sometimes life-altering, consequences.
India raises the stakes. Here meaning is layered across languages, castes, classes, regions and registers, relational and situational rather than literal, where the same sentence shifts depending on who says it to whom. Yet most AI is trained on flattened, English-dominant data that cannot hold that texture. The danger is not only exclusion but a quiet standardization of interpretation that casts one version of reality as the only one. A system that cannot read social nuance scales the misreading into millions of decisions. This is not to say compute no longer counts. Models still need power, data and first-rate engineering, and India needs far more of all three. But the decisive capability is migrating. When everyone can buy comparable computing and license comparable models, the differentiator is no longer who processes language fastest, but who understands it most deeply: its ethics, its politics, its cultural freight.
Ultimately the shift is philosophical. Human judgement has never run on logic alone; it draws on context, memory, metaphor, ethics and culture, the faculties that resist clean quantification. Through language, machines are now entering that human terrain. The future of AI governance will depend less on who owns the most advanced model than on who understands language well enough to govern it.
So the practical work is clear. Boards should treat the labelling and classification of data as a question of governance, not a backroom technical chore. Technology leaders should bring linguists, ethicists and communicators into the room where models are designed, not summon them afterwards to explain the result. And India should invest in data built in its own languages and registers, rather than borrow a worldview encoded in someone else's. These people are not soft adjuncts to the engineering. They are load-bearing. For too long, institutions treated words as a layer on top of technology, useful for explaining it, selling it, defending it. That hierarchy is inverting. In the AI era, words are the architecture through which machines understand the world. And whoever shapes the words is already building the machine.
(Dr Ruchi Tewari is the Associate Professor and Associate Dean – Marketing, Communication and Public Affairs, Chief Marketing Officer at MICA. An academician with over 25 years of experience, she holds a PhD in Management with a focus on CSR communication in the Indian IT sector. She also holds an MPhil and an MA in English Literature.)
For as long as we have built machines, words merely explained them. That has changed. Today's technological leap runs on language itself. Artificial intelligence, and the large language models behind generative AI, are built entirely on words, with no inherent grasp of context or emotion. What a model "knows" comes from the data it is trained on, and that data carries our stereotypes and biases, all of them encoded in language. Feed it "workforce optimization" and it reads efficiency; feed it "employee displacement" and it reads harm. The facts are identical. Only the words have moved. Words no longer describe the system. They are the system.
Earlier, the best systems were won on scale and speed, on the belief that more computing meant better outcomes. LLMs break that rule. Their performance turns on linguistic precision, not volume. And language is never neutral. A small semantic shift changes how a model frames a problem and what it recommends. "Consumer engagement" and "behavioural surveillance" can describe the same activity on identical facts, yet one invites investment and the other a lawsuit. How information is framed now decides how well a machine reasons. Words have become its operating terms.
This is not about an individual coaxing a better answer from a chatbot. It runs deeper. Training a model is institutional work. Information is classified and tagged, and those choices become the machine's foundation for reading the world. The labels an organization assigns, safe or risky, trustworthy or harmful, productive or wasteful, harden into datasets, interfaces and workflows. They teach the machine what to notice, what to ignore, how to read human behaviour and what to produce by default. The future influence of technology leaders will rest on how well they shape these semantic systems, not on technical know-how alone.
It is time organizations rethought where communication sits. For fifty years it has been treated as downstream work, a voice given to ideas already settled behind closed doors. Its job was the faade: to gloss the presentation, to add the face-lift once the strategy was done. AI collapses that order. Communication is no longer symbolic; it is infrastructural. Semantics no longer shape only human perception; they govern how the machine interprets. A careless taxonomy can distort a model's output, and through it, people's sense of what is true. An ethically thin classification can normalize harm, then scale it. As AI enters hiring, healthcare, finance, governance and education, these word choices stop being cosmetic. They start carrying material, sometimes life-altering, consequences.
India raises the stakes. Here meaning is layered across languages, castes, classes, regions and registers, relational and situational rather than literal, where the same sentence shifts depending on who says it to whom. Yet most AI is trained on flattened, English-dominant data that cannot hold that texture. The danger is not only exclusion but a quiet standardization of interpretation that casts one version of reality as the only one. A system that cannot read social nuance scales the misreading into millions of decisions. This is not to say compute no longer counts. Models still need power, data and first-rate engineering, and India needs far more of all three. But the decisive capability is migrating. When everyone can buy comparable computing and license comparable models, the differentiator is no longer who processes language fastest, but who understands it most deeply: its ethics, its politics, its cultural freight.
Ultimately the shift is philosophical. Human judgement has never run on logic alone; it draws on context, memory, metaphor, ethics and culture, the faculties that resist clean quantification. Through language, machines are now entering that human terrain. The future of AI governance will depend less on who owns the most advanced model than on who understands language well enough to govern it.
So the practical work is clear. Boards should treat the labelling and classification of data as a question of governance, not a backroom technical chore. Technology leaders should bring linguists, ethicists and communicators into the room where models are designed, not summon them afterwards to explain the result. And India should invest in data built in its own languages and registers, rather than borrow a worldview encoded in someone else's. These people are not soft adjuncts to the engineering. They are load-bearing. For too long, institutions treated words as a layer on top of technology, useful for explaining it, selling it, defending it. That hierarchy is inverting. In the AI era, words are the architecture through which machines understand the world. And whoever shapes the words is already building the machine.
(Dr Ruchi Tewari is the Associate Professor and Associate Dean – Marketing, Communication and Public Affairs, Chief Marketing Officer at MICA. An academician with over 25 years of experience, she holds a PhD in Management with a focus on CSR communication in the Indian IT sector. She also holds an MPhil and an MA in English Literature.)
