Pennings, A.J. (2026, Jun 15) How Spreadsheet Logic became the Foundation of AI, Part 2. apennings.com https://apennings.com/data-analytics-and-meaning/how-spreadsheet-logic-became-the-foundation-of-ai-part-2/

Introduction

This is my second post on how AI drew on spreadsheet logic. It starting with the fundamentals of the spreadsheets. Words, numbers, lists, columns, rows, tables, functions and formulas are remediated into the spreadsheet.[1] Then it shows how this combination became the foundation for AI. And ultimately how AI has progressed beyond the digital spreadsheet.

Long before modern AI models were trained on billions of parameters, the fundamental architecture of human data organization was already being perfected in the humble spreadsheet. One of my main interest is in how new media incorporate old media, in a process called remediation.[2] The spreadsheet clearly remediated some basic media such as writing, lists, tables, columns and rows, as well as functions and formulas. The purpose of remediation is to form a better, “healed,” or more authentic access to reality.[3]

Artificial Intelligence didn’t replace spreadsheet logic; it grew directly out of it. The core concepts that make a spreadsheet work are the exact mathematical and philosophical stepping stones that computer scientists used to build machine learning and large language models (LLMs).

Here is a breakdown of how the fundamentals of spreadsheets became the foundation for AI, and how AI has since more beyond spreadsheet logic.

Part 1: From the Spreadsheet Grid to the Neural Network

If you look closely at how modern AI operates, you can see the ghost of VisiCalc and Excel in almost every process. Every core spreadsheet element has a direct transition into machine learning architecture.

Words, Numbers, and Lists -> Data Types, Arrays, and Tensors

In a spreadsheet, you start with the basics, specifically numbers (floats/integers) and words (strings). When you group these into a single row or column, you get a list.

In AI, a list of numbers is called a vector (or a 1D array). When you stack those lists into multi-dimensional blocks, you get a tensor. AI models don’t read words like humans do; they convert words into long lists of numbers (called word embeddings) so they can process text using pure math.

Columns and Rows -> Features and Observations

A spreadsheet relies on a two-dimensional grid where rows represent individual entries (e.g., a specific customer or a single transaction) and columns represent attributes of those entries (e.g., age, income, purchase history).

In Machine Learning (ML), rows are called observations or data points, and columns are called features. When an AI predicts whether a credit card transaction is fraudulent, it is essentially scanning a giant, invisible spreadsheet row and calculating the mathematical relationships between the column features.

Tables -> Relational Datasets & Tabular Foundation Models

A table is a collection of structured, related data with clear boundaries.

Early AI relied heavily on cleaning these tables for training. Today, AI has advanced to the point of creating Tabular Foundation Models (TFMs). Just like ChatGPT was trained on billions of pages of text to understand human language, TFMs are trained on millions of diverse spreadsheets to inherently understand structural relationships, allowing them to predict missing cell values or forecast trends instantly without needing task-specific retraining.

Functions and Formulas -> Weights, Biases, and Activation Functions

In a spreadsheet, a formula (like =A1+B1) or a function (like SUM or VLOOKUP) is a deterministic rule. You give it input X, and it applies a fixed mathematical rule to give you output Y.

A neural network is essentially millions of interconnected spreadsheet cells containing adaptive formulas. Instead of a human manually writing =IF(A1>10, “Yes”, “No”), the AI assigns fractional “weights” and “biases” to data inputs. It uses activation functions (the AI equivalent of a complex nested IF statement) to decide whether to pass information to the next layer. The crucial difference? AI writes its own formulas through trial and error.

Part 2: How AI Has Progressed Beyond the Digital Spreadsheet

While spreadsheets provided the structural logic for AI, modern artificial intelligence has broken free from the constraints of rows, columns, and deterministic calculations.

To understand how technology transitioned from rigid data grids into adaptable intelligence, we can look at three major shifts in how software handles data structure, logic, and reasoning.

First, consider Data Structure. In the spreadsheet era, computers were strictly confined to highly structured environments. Data had to be neatly compartmentalized into grids, tables, and CSV files; if information couldn’t fit into a specific row or column, the system couldn’t process it. The modern AI era has shattered these boundaries by mastering unstructured data. Today’s systems can natively interpret and find meaning in the chaotic formats of the real world, including audio recordings, video feeds, free-form text, and even complex 3D spaces.

Second, the fundamental Logic governing these systems has flipped. Spreadsheets operate on deterministic logic, meaning they rely entirely on explicit, hard-coded rules written by humans. If a user types a specific formula, the spreadsheet executes that exact calculation every single time, with no room for nuance. Modern AI, by contrast, operates on probabilistic logic. Instead of following rigid instructions, the computer analyzes vast datasets to learn its own fluid patterns. This allows the AI to make calculated predictions and navigate ambiguity, determining what is “most likely” correct rather than demanding absolute mathematical certainty.

Finally, the style of Reasoning has undergone a massive paradigm shift. Spreadsheet logic is inherently linear and sequential—it processes data step-by-step, moving from cell to cell along a fixed path. Modern AI utilizes parallel, contextual, and semantic reasoning. It can process massive amounts of interconnected information simultaneously, evaluating not just individual data points, but the overarching context and deeper meaning behind them.

From Rigid Grids to Multi-Dimensional Context (Transformers)

Spreadsheets are bound by two dimensions: X and Y. Even if you link multiple sheets, the paths are static. Modern AI utilizes Transformer architectures (the tech behind LLMs) to look at data in thousands of dimensions simultaneously. It maps semantic meaning, tone, and abstract concepts into a “latent space” where words and ideas are clustered by contextual relevance, completely removing the need for an explicit table layout.

Handling the “Unstructured” World

Spreadsheets break down completely when fed a photograph, a voice memo, or a live video feed. You can’t fit a pixel’s emotional context into a cell. AI has progressed by developing specialized architectures—like Convolutional Neural Networks (CNNs) for vision—that treat unstructured data as dynamic arrays, allowing computers to “see” and “hear” without requiring human-engineered column headers.

Deterministic vs. Probabilistic Logic

If an Excel formula fails or encounters a missing value, it throws an error (#VALUE!, #N/A). It demands perfection. AI operates on probability. When an LLM generates a sentence or an AI predicts stock inventory, it isn’t executing a fixed mathematical absolute; it is calculating the most likely next state based on historical patterns. This makes AI highly resilient to messy, incomplete, and chaotic real-world data that would break a traditional spreadsheet template.

Generative Autonomy

A spreadsheet is reactive; it sits silently until a human updates an input or writes a script. Modern AI is generative and agentic. It can evaluate its own outputs, write its own code (including Python or SQL to manipulate data behind the scenes), and take multi-step actions to achieve a goal specified in plain natural language.[2]

Conclusion

Artificial Intelligence didn’t replace spreadsheet logic; it grew directly out of it. The core concepts that make a spreadsheet work are the exact mathematical and philosophical stepping stones that computer scientists used to build machine learning and large language models (LLMs). Long before modern AI models were trained on billions of parameters, the fundamental architecture of human data organization was already being perfected in the humble spreadsheet.

Ultimately, spreadsheets taught computers how to organize the world into clean math. AI took that math, abstracted it, and used it to map the messiness of human reality.

Summary

This post explores how the fundamental building blocks of spreadsheets served as the direct conceptual and mathematical stepping stones for modern Artificial Intelligence.

Part 1: From Grid to Neural Network details how foundational spreadsheet elements were remediated into AI architecture. Simple words, numbers, and lists were changed into high-dimensional vectors and tensors. Rows and columns became machine learning observations and features, while traditional data tables paved the way for Tabular Foundation Models (TFMs). Finally, static formulas and functions transformed into dynamic weights, biases, and activation functions, allowing AI to essentially “write its own formulas.”

Part 2: Beyond the Digital Spreadsheet outlines how AI shattered the limitations of the 2D grid. The post contrasts the two eras across three paradigm shifts: moving from structured grids to unstructured data (like audio and video), swapping deterministic human rules for probabilistic machine-learned logic, and upgrading from linear processing to parallel, contextual reasoning via Transformer architectures. Ultimately, while spreadsheets forced reality into clean math, AI uses that math to navigate the messy, autonomous, and multi-dimensional real world.

Notes

[1] Pennings, A.J. (2025, Aug 12) Excel vs. Java: Two Languages, Two Worlds of Thought: Symbolic Grid vs. the Algorithmic Script. apennings.com https://apennings.com/meaning-makers/digital-spreadsheets-power-over-people-and-resources/
[2] My presentation with Saebom Jin at APEC 2025 is instructive about remediation. Pennings, A.J., Jin, S. (2025, Sep 13) APEC Presentation: Healing the COP (Common Operating Picture) with AI and APIs for Disaster Management. apennings.com https://apennings.com/crisis-communications/apec-presentation-healing-the-cop-common-operating-picture-with-ai-and-apis-for-disaster-management/
[3] Bolter, Jay David, and Richard Grusin. Remediation: Understanding New Media. Cambridge, MA: MIT Press, 2000.
[4] Innovations around the merger of spreadsheets and AI are quite intriguing. For example, see Quadratic AI

Prompt(s) Conversation with Gemini. I am looking for more details about how AI drew on spreadsheet logic. Start with the fundamentals of the spreadsheets: words, numbers, lists, columns, rows, tables, functions and formulas and show how they became the foundation for AI. And how AI has progressed beyond the digital spreadsheet.

© ALL RIGHTS RESERVED

Not to be considered financial advice. AI is often used, and results are thoroughly interrogated. Links are used for some citations.

---

Anthony J. Pennings, PhD is a Professor at the Department of Technology and Society, State University of New York, Korea and a Research Professor for Stony Brook University. He teaches AI and broadband policy. From 2002-2012 he taught digital economics and information systems management at New York University. He also taught in the Digital Media MBA at St. Edwards University in Austin, Texas, where he lives when not in Korea.

Related

Category: artificial intelligence, blockchain, cryptocurrencies, data analytics and meaning, spreadsheet logic
Tags: Neural Network > remediation theory > Spreadsheet logic

Comments

Comments are closed.