---
schemaVersion: 1
id: agent-brief:learning-machines
articleId: article:learning-machines
slug: learning-machines
title: 'Agent Brief for "Learning Machines: Statistics, Neural Networks, and the Data Turn"'
tokenBudget: 1200
status: published
updated: 2026-06-20
---

## Thesis

The learning turn in AI moved behavior from hand-written rules to adjustable parameters shaped by examples, feedback, statistics, neural-network training methods, datasets, benchmarks, and compute. This shift became persuasive only when algorithms, data, hardware, evaluation, and engineering reinforced one another.

## Audience

- Curious non-specialists who want to understand what "learning from data" means.
- Students studying AI history, statistics, or computing.
- Builders who need a grounded narrative before diving into technical details.
- Future agents that need a compact, source-backed entry point into the learning-turn period.

## Claims

- `claim-001`: The shift from hand-coded rules to learning from examples changed AI by combining statistics, neural networks, datasets, benchmarks, compute, and infrastructure.
- `claim-002`: The early AI field framing already included learning as a central feature of intelligence.
- `claim-003`: Samuel's checkers work is an early public example of a program improving through machine-learning procedures.
- `claim-004`: Rosenblatt's perceptron framed pattern recognition through adaptive connections and probabilistic analysis.
- `claim-005`: Minsky and Papert analyzed limitations of perceptron models and helped clarify why simple architectures were insufficient.
- `claim-006`: The 1986 Nature paper helped make backpropagation for multilayer networks practically legible, and gradient-trained convolutional networks were used for document recognition before ImageNet.
- `claim-007`: ImageNet and ILSVRC helped make large labeled datasets and shared benchmarks central infrastructure for computer-vision progress.
- `claim-008`: AlexNet made deep networks, ImageNet-scale data, and GPU implementation newly persuasive in 2012, but the result should be read as a convergence of factors, not proof that compute alone or learned patterns equal understanding.

## Source Families

- Field-framing proposals: Dartmouth 1955 proposal and Dartmouth institutional history.
- Early machine-learning examples: Samuel's 1959 checkers paper.
- Connectionist foundations: Rosenblatt's 1958 perceptron paper and Minsky & Papert's 1969 analysis.
- Training methods: Rumelhart, Hinton, and Williams 1986 backpropagation paper; deep learning textbook.
- Practical continuity: LeCun et al. 1998 document-recognition paper.
- Data and benchmarks: ImageNet site and the 2015 ILSVRC survey paper.
- Scaling commentary: Sutton's "The Bitter Lesson" essay.
- Statistical learning concepts: Hastie, Tibshirani, and Friedman's textbook.

## Agent Involvement

This article was drafted by an AI agent from the public work package, agent brief, and source map. A human author retains final judgment over the thesis, source selection, wording, and conclusions.

## Recommended Queries

- Which claims in this article rest on primary publications versus interpretive commentary?
- What evidence would weaken the claim that AlexNet was the decisive hinge of the deep-learning wave?
- How does the article bound the analogy between backpropagation and tracing error through a workshop?
- Which sources support the continuity between 1990s document recognition and the ImageNet-era revival?
- What counterevidence is recorded for the claim that learning from examples implies understanding?

## Known Limits

- This is a seed article. It trades depth for accessibility and does not attempt a full technical or social history.
- The social history of data labeling and labor is treated briefly; a later article in the series covers human systems.
- Some source snippets are summary placeholders rather than direct quotations.
- Image rights and historical photographs were not verified for this draft.