💬 NLP

Interview with Kalle Lyytinen on "Implications of Theories of Language for Information Systems"

In this interview, Kalle Lyytinen reflects on the origins and evolution of his seminal work on the linguistic foundations of information systems, discussing their continued relevance in the era of large language models and generative AI while outlining future research directions.

Original authors: Kalle J. Lyytinen, Pierre Maier, Paul Ackah Toffey

Published 2026-07-28

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Kalle J. Lyytinen, Pierre Maier, Paul Ackah Toffey

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the world of Information Systems (IS) not as a collection of cold, hard wires and glowing screens, but as a giant, bustling library where the books are made of light and the librarians are algorithms. At its heart, IS is the study of how we use technology to talk to each other and get things done. Think of it like a massive translation project: humans have messy, complicated feelings and ideas, and computers speak in strict, binary code (ones and zeros). The job of IS researchers is to figure out how to build bridges between these two worlds so that the computer understands what we mean, and we understand what the computer is doing. For decades, scientists have argued over how language works to build these bridges. Some think language is like a rigid set of math rules (grammar), while others think it's more like a social game where meaning changes depending on who is talking and where. This matters to everyone because every time you send a text, order a pizza online, or use a GPS, you are relying on these invisible bridges. If the bridge is built on the wrong idea of how language works, the message gets lost, the pizza goes to the wrong house, or the GPS drives you into a lake.

In this lively conversation, Kalle Lyytinen, a veteran researcher who first wrote about these "language bridges" forty years ago, sits down with two curious interviewers to see how the landscape has changed. The paper is essentially a time-traveling chat where Lyytinen looks back at his famous 1985 article and asks, "Did we get it right? And what do we do now that computers can talk back?"

Lyytinen explains that his original idea was simple but deep: Information Systems are not just machines; they are linguistic systems. He compares them to "bit strings" (long chains of zeros and ones) that are completely meaningless until humans give them a story. A computer is just a robot that shuffles these bit strings around; it doesn't know what they mean. The meaning comes from us, the humans, who use these systems to communicate and work. He argues that for a long time, researchers treated language as a boring, background fact—like air that we just breathe without thinking about it. But Lyytinen insists that the "air" is actually the most important part. If you don't understand how people use words to create meaning, you can't build a system that actually helps them.

When the interviewers ask if his old advice still holds up, Lyytinen admits that the world has shifted. Back in the 80s, computers were mostly used to automate boring office tasks, like keeping track of inventory or accounting. These were like digital versions of old paper files. Today, however, we have Large Language Models (LLMs) and Generative AI. These are the new kids on the block. Lyytinen describes them as fascinating but tricky. Unlike the old computers that followed strict rules, these new AI models are like super-smart parrots. They don't actually "know" what a word means; they just know which words usually appear next to each other based on reading billions of sentences. They are incredibly good at guessing the next word in a sentence, but they have no concept of the real world. If you ask them a question, they don't "think" about the answer; they just calculate the most likely response based on patterns.

This leads to a major twist in the story. Lyytinen points out that while these AI models are amazing at mimicking human speech, they lack a "normative view." In plain English, they don't know what is right or wrong in the real world, only what is common in their training data. They are like a chef who has tasted every recipe in the world but has never actually eaten a meal or understood what hunger feels like. Because of this, Lyytinen suggests that we can't just trust them blindly. We humans are still the "final arbitrators"—we have to decide if what the AI says actually makes sense.

The conversation also dives into how this changes the "data layer" of our technology. Lyytinen uses a creative analogy here: in the past, data was like a neatly organized filing cabinet with labeled folders. Now, with AI, the data layer has become a chaotic but powerful soup of "tokens" (tiny pieces of words) and "models" (the recipes for mixing them). This is a brand new kind of artifact that researchers are still trying to understand. It's not just about storing data anymore; it's about managing these new, fluid capabilities. Lyytinen warns that we are currently in a "learning by trying" phase. We don't have a perfect map for how to manage these new AI tools in big companies yet. It's like trying to build a skyscraper while the ground is still shifting.

Finally, Lyytinen offers some advice for the next generation of researchers (like the PhD students asking the questions). He suggests that the best way to make a mark isn't just to follow the latest hype, but to ask a "higher-level" question. Instead of just asking "How do we code this AI?", ask "How does this AI change the way we think about work?" or "How does it change the meaning of data?" He believes that the most exciting research will come from figuring out how these new, performative tools (tools that do things rather than just show things) fit into our social lives.

In short, the paper suggests that while our tools for talking to computers have gotten incredibly fancy, the fundamental rule remains: meaning is a human thing. Computers can shuffle words and guess patterns, but they cannot create meaning on their own. As we move forward, the challenge isn't just building smarter AI, but figuring out how to weave these new, pattern-matching machines into the messy, social, and linguistic fabric of human life without losing our grip on what actually matters. The paper doesn't claim to have solved this puzzle; rather, it suggests that we are just at the beginning of a long, interesting journey to understand it.

Technical Summary: Interview with Kalle Lyytinen on "Implications of Theories of Language for Information Systems"

Problem Statement
The paper addresses the foundational identity of Information Systems (IS) as a discipline and the role of language within it. Forty years after Kalle Lyytinen's seminal 1985 MIS Quarterly article, the interview explores whether the "linguistic essence" of IS remains a central, actionable theoretical lens or if it has been relegated to a background assumption. The core problem is the lack of a unified, systematic approach to integrating different theories of language (e.g., Fregean, Chomskyan, Skinnerian) into IS research and practice. Furthermore, the paper investigates how the emergence of Large Language Models (LLMs) and Generative AI challenges traditional linguistic views, particularly regarding the nature of meaning, semantics, and the structure of data layers in modern digital artifacts.

Methodology
This paper is not an empirical study or a theoretical derivation but a structured qualitative interview conducted on November 7, 2025, between Kalle Lyytinen and researchers Pierre Maier and Paul Ackah Toffey. The methodology relies on:

Retrospective Analysis: Lyytinen reflects on the origins, motivations, and theoretical underpinnings of his 1985 article.
Comparative Discussion: The interview contrasts the state of IS research in the 1980s (focused on office automation, conceptual modeling, and entity-relationship diagrams) with the current landscape (Generative AI, LLMs, and complex socio-technical systems).
Critical Evaluation: The participants evaluate the applicability of Lyytinen's original "five language views" to contemporary phenomena, specifically examining where these views hold, where they fail, and what new theoretical gaps exist.

Key Contributions
The paper provides a retrospective and forward-looking synthesis of the linguistic perspective in IS, offering several key contributions:

Reaffirmation of the Linguistic Essence: Lyytinen asserts that IS are fundamentally "semiotic systems" or "bit strings" that derive meaning only through human social construction. He argues that while the field often takes this for granted, the linguistic dimension is inextricable from the definition of IS, distinguishing it from pure Computer Science (which focuses on the manipulation of bit strings) and other management fields (which often "black-box" the artifact).
Critique of the "Contingency Approach": Lyytinen admits that his 1985 suggestion to systematically integrate multiple language views was not fully adopted by the community. He attributes this to the difficulty of shifting between views and the evolution of development methodologies (e.g., Agile) which embed linguistic considerations intuitively rather than through explicit theoretical frameworks.
Theoretical Re-evaluation of LLMs: A significant contribution is the distinction between traditional conceptual modeling and the "cognitive substrates" of LLMs. Lyytinen argues that LLMs lack a normative view of the world or an inherent sense of semantics; they operate on statistical correlations of tokens. Consequently, they do not possess "concepts" in the traditional sense, challenging the validity of applying standard semantic theories directly to them.
Redefinition of the Data Layer: The paper introduces a novel architectural perspective where the data layer of enterprise systems now includes "tokens" and "models" as distinct artifacts alongside traditional data types. This creates a new, complex hierarchy (raw data $\to$ tokens $\to$ models $\to$ instances) that requires new management strategies and theoretical frameworks.
Guidance for Future Research: Lyytinen outlines specific directions for PhD students and researchers, emphasizing the need to move beyond merely applying new technologies to asking "novel questions" about how these technologies change the organization of work, the nature of expertise, and the performative value of systems.

Results and Findings

Status of Language Views: The five language views proposed in 1985 (Fregean, Chomskyan, Piagetian, Skinnerian, and Ordinary Speaking) are still relevant but their relative importance has shifted. The field has not successfully integrated them into a single systematic method; instead, practitioners often rely on intuitive, pragmatic approaches (e.g., in Agile development) that implicitly handle linguistic complexity.
Nature of LLMs: LLMs are identified as a "novel phenomenon" that operates differently from human language. They are descriptive and stochastic rather than normative. Their "meaning" is not intrinsic but is determined by the human user's interpretation of the output.
Conceptual Modeling Evolution: The role of conceptual modeling has shifted from "greenfield" design (creating models from scratch) to "matching" and "orchestrating" existing enterprise grammars (e.g., ERP systems) and integrating them with new AI capabilities.
Reproducibility Challenges: The use of LLMs in research introduces significant reproducibility issues. Because prompts can yield different results across runs, researchers must account for variance and demonstrate transparency in their inferences, similar to qualitative research but with added technical complexity.
Identity of IS: The discipline remains defined by the study of "work systems" where digital technologies are embedded. The "artifact" cannot be removed from the system definition; however, the focus has expanded from the artifact itself to the broader socio-technical orchestration and the "affordances" it provides.

Significance of the Paper
The paper claims significance as a critical reflection on the theoretical foundations of IS. It serves to:

Contextualize Current Trends: It frames the rapid rise of Generative AI not just as a technological shift but as a fundamental change in the "data layer" and the nature of linguistic representation within organizations.
Highlight Research Gaps: It identifies a lack of systematic research on how to manage the new artifacts (tokens and models) at the enterprise level and how to theorize the interaction between traditional data and stochastic AI models.
Motivate Rigorous Inquiry: By revisiting the origins of the linguistic perspective, the paper encourages the IS community to avoid superficial adoption of new technologies. Instead, it calls for deep, theoretically grounded research that asks how these technologies alter the "forms of life" and the organization of work, urging researchers to move beyond technical implementation to understanding the broader socio-technical implications.

The paper concludes that while the specific methods of 1985 may seem dated, the core insight—that IS are linguistic systems requiring a deep understanding of meaning, semantics, and social construction—remains vital, albeit in a more complex and performative form.

Technical Summary: Interview with Kalle Lyytinen on "Implications of Theories of Language for Information Systems"

More like this