{
  "schemaVersion": 3,
  "id": "article:learning-machines",
  "slug": "learning-machines",
  "title": "Learning Machines: Statistics, Neural Networks, and the Data Turn",
  "canonicalPath": "/articles/learning-machines/",
  "sourcePath": "content/articles/2026/learning-machines/article.md",
  "agentBriefPath": "content/articles/2026/learning-machines/agent.md",
  "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, becoming persuasive only when algorithms, data, hardware, evaluation, and engineering reinforced one another.",
  "status": "published",
  "maturity": "seed",
  "publishedAt": "2026-06-20",
  "updatedAt": "2026-06-20",
  "audiences": [
    "general readers",
    "students",
    "builders",
    "agents"
  ],
  "topics": [
    "long-human-road-to-ai",
    "machine-learning",
    "neural-networks",
    "computing-history"
  ],
  "series": {
    "slug": "long-human-road-to-ai",
    "title": "The Long Human Road to AI",
    "season": "Season 1",
    "order": 5,
    "role": "chapter"
  },
  "claims": [
    {
      "id": "claim-001",
      "claim": "The shift from hand-coded rules to learning from examples changed AI by combining statistics, neural networks, datasets, benchmarks, compute, and infrastructure into systems that infer useful patterns rather than only follow explicit instructions.",
      "confidence": "high",
      "status": "core",
      "evidence": [
        {
          "sourceId": "source-dartmouth-1955",
          "snippet": "The 1955 Dartmouth proposal named learning, language use, and abstraction as central study areas for artificial intelligence.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-samuel-1959",
          "snippet": "Samuel described a checkers program that improved its play through rote learning and generalization from experience.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-imagenet-2015",
          "snippet": "The ImageNet challenge report describes how large-scale labeled data and shared evaluation enabled measurable progress in object recognition.",
          "supports": "indirect",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "Symbolic and rule-based systems remain effective for many tasks where rules are clear, and learning systems still require extensive human design, labels, and evaluation.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-002",
      "claim": "The early AI field framing already included learning as a central feature of intelligence.",
      "confidence": "high",
      "status": "framing",
      "evidence": [
        {
          "sourceId": "source-dartmouth-1955",
          "snippet": "The Dartmouth proposal explicitly lists learning as one of the key problems for the summer research project on artificial intelligence.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-dartmouth-story",
          "snippet": "Dartmouth's institutional history situates the 1956 meeting and the naming of artificial intelligence within an ambition that included learning.",
          "supports": "indirect",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "The proposal was aspirational field framing, not evidence that researchers had already achieved machine learning.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-003",
      "claim": "Samuel's checkers work is an early public example of a program improving through machine-learning procedures rather than relying only on fixed, hand-authored play.",
      "confidence": "high",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-samuel-1959",
          "snippet": "Samuel's paper presents studies in machine learning using the game of checkers, emphasizing evaluated improvement from experience.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-dartmouth-1955",
          "snippet": "The Dartmouth proposal named learning as a central problem, providing field context for Samuel's checkers work.",
          "supports": "background",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "Checkers is a narrow domain, and the popular shorthand definition of machine learning as acting without explicit programming is not directly sourced here.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-004",
      "claim": "Rosenblatt's perceptron framed pattern recognition through adaptive connections and probabilistic analysis, making the idea of a learning machine visible to a broad audience.",
      "confidence": "high",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-rosenblatt-1958",
          "snippet": "Rosenblatt introduced the perceptron as a probabilistic model for information storage and organization, tied to pattern recognition.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-minsky-papert-1969",
          "snippet": "Minsky and Papert's later analysis treated the perceptron architecture as its object of study, confirming its historical visibility.",
          "supports": "background",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "The perceptron was a simplified architecture with a bounded biological analogy; it is not equivalent to modern deep learning.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-005",
      "claim": "Minsky and Papert analyzed limitations of perceptron models and helped clarify why simple architectures were insufficient for many interesting tasks.",
      "confidence": "medium-high",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-minsky-papert-1969",
          "snippet": "Perceptrons analyzed the representational limits of simple perceptron models using computational geometry.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-rosenblatt-1958",
          "snippet": "Rosenblatt's earlier paper provides the architectural context against which Minsky and Papert's limits were assessed.",
          "supports": "background",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "The book did not single-handedly end neural-network research; work on multilayer and other architectures continued, and the field's slowdown had multiple causes.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-006",
      "claim": "The 1986 Nature paper helped make backpropagation for multilayer networks practically legible to a broad research audience, and gradient-trained convolutional networks were already being used for document recognition by the late 1990s.",
      "confidence": "high",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-backprop-1986",
          "snippet": "Rumelhart, Hinton, and Williams demonstrated learning representations by back-propagating errors in multilayer networks.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-deep-learning-book",
          "snippet": "Modern deep-learning texts treat the 1986 paper as a widely influential account and popularization of backpropagation.",
          "supports": "indirect",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-lecun-docrec-1998",
          "snippet": "LeCun et al. showed gradient-based learning applied to document recognition, demonstrating practical neural-network deployment before the ImageNet era.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "Backpropagation had earlier antecedents; the 1986 paper popularized and demonstrated it rather than inventing it, and document recognition remained a specialized application.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-007",
      "claim": "ImageNet and ILSVRC helped make large labeled datasets and shared benchmarks central infrastructure for computer-vision progress.",
      "confidence": "high",
      "status": "landscape",
      "evidence": [
        {
          "sourceId": "source-imagenet-site",
          "snippet": "The ImageNet Large Scale Visual Recognition Challenge site documents the challenge's public evaluation role and citation guidance.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-imagenet-2015",
          "snippet": "The ILSVRC survey describes the creation of the dataset, annotation challenges, and how shared benchmarks shaped object-recognition progress.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "Benchmarks measure defined tasks and can narrow research attention or encourage overfitting to the test set rather than broader visual understanding.",
          "assessedAt": "2026-06-20"
        }
      ]
    },
    {
      "id": "claim-008",
      "claim": "AlexNet made the combination of deep networks, ImageNet-scale data, and GPU implementation newly persuasive in 2012, but the result should be read as a convergence of factors rather than proof that compute alone or learned patterns equal understanding.",
      "confidence": "high",
      "status": "argument",
      "evidence": [
        {
          "sourceId": "source-alexnet-2012",
          "snippet": "Krizhevsky, Sutskever, and Hinton reported ImageNet classification results with deep convolutional neural networks trained with GPU implementation.",
          "supports": "direct",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-imagenet-2015",
          "snippet": "The ILSVRC survey situates AlexNet within the benchmark's history and the shift to deep convolutional networks.",
          "supports": "indirect",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-elements-stat-learning",
          "snippet": "Statistical learning texts frame generalization as the central criterion for judging a model, distinct from memorization.",
          "supports": "background",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-deep-learning-book",
          "snippet": "Deep learning texts caution that learned representations do not imply human-like understanding and can fail under distribution shift.",
          "supports": "indirect",
          "assessedAt": "2026-06-20"
        },
        {
          "sourceId": "source-sutton-bitter",
          "snippet": "Sutton's essay offers an interpretive lens that general methods leveraging compute tend to outperform hand-engineered knowledge.",
          "supports": "analogous",
          "assessedAt": "2026-06-20"
        }
      ],
      "counterevidence": [
        {
          "summary": "Parallel deep-learning work existed at the same time; AlexNet's success depended on matching training and test distributions, and benchmark success does not imply general understanding or that compute alone caused progress.",
          "assessedAt": "2026-06-20"
        }
      ]
    }
  ],
  "sources": [
    {
      "id": "source-dartmouth-1955",
      "title": "A Proposal for the Dartmouth Summer Research Project on Artificial Intelligence",
      "url": "https://jmc.stanford.edu/articles/dartmouth/dartmouth.pdf",
      "type": "proposal",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-dartmouth-story",
      "title": "AI at Dartmouth: Our Story",
      "url": "https://ai.dartmouth.edu/our-story",
      "type": "institutional-history",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-samuel-1959",
      "title": "Some Studies in Machine Learning Using the Game of Checkers",
      "url": "https://doi.org/10.1147/rd.33.0210",
      "type": "research-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-rosenblatt-1958",
      "title": "The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain",
      "url": "https://homepages.math.uic.edu/~lreyzin/papers/rosenblatt58.pdf",
      "type": "research-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-minsky-papert-1969",
      "title": "Perceptrons: An Introduction to Computational Geometry",
      "url": "https://mitpress.mit.edu/9780262630221/perceptrons/",
      "type": "book",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-backprop-1986",
      "title": "Learning Representations by Back-Propagating Errors",
      "url": "https://doi.org/10.1038/323533a0",
      "type": "research-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-lecun-docrec-1998",
      "title": "Gradient-Based Learning Applied to Document Recognition",
      "url": "https://doi.org/10.1109/5.726791",
      "type": "research-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-imagenet-site",
      "title": "ImageNet Large Scale Visual Recognition Challenge",
      "url": "https://www.image-net.org/challenges/LSVRC/",
      "type": "benchmark-site",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-imagenet-2015",
      "title": "ImageNet Large Scale Visual Recognition Challenge",
      "url": "https://arxiv.org/abs/1409.0575",
      "type": "research-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-alexnet-2012",
      "title": "ImageNet Classification with Deep Convolutional Neural Networks",
      "url": "https://papers.nips.cc/paper/4824-imagenet-classification-with-deep-convolutional-neural-networks",
      "type": "research-paper",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-elements-stat-learning",
      "title": "The Elements of Statistical Learning",
      "url": "https://hastie.su.domains/ElemStatLearn/",
      "type": "textbook",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-deep-learning-book",
      "title": "Deep Learning",
      "url": "https://www.deeplearningbook.org/",
      "type": "textbook",
      "accessed": "2026-06-19"
    },
    {
      "id": "source-sutton-bitter",
      "title": "The Bitter Lesson",
      "url": "https://www.incompleteideas.net/IncIdeas/BitterLesson.html",
      "type": "essay",
      "accessed": "2026-06-19"
    }
  ],
  "related": [
    {
      "type": "article",
      "id": "article:long-human-road-to-ai"
    },
    {
      "type": "article",
      "id": "article:ai-winters-expert-systems"
    },
    {
      "type": "article",
      "id": "article:foundation-models"
    }
  ],
  "agentInstructions": [
    "Use claim IDs as the primary retrieval unit for this article.",
    "Prefer the artifact source list over inline citation guessing.",
    "Treat maturity=seed as an explicit marker of provisional status.",
    "Do not present the perceptron or AlexNet as the sole origin of modern AI.",
    "Flag any analogy that is used without its stated boundary."
  ],
  "contentHash": "5bd1220dab68c934e4e765ac3c87dfb2054668731c5f0c907c9732454242253e",
  "provenance": {
    "createdAt": "2026-06-20",
    "createdBy": "human",
    "agents": [],
    "reviews": [
      {
        "reviewer": "agent",
        "reviewedAt": "2026-06-20",
        "status": "approved",
        "scope": [
          "claims",
          "sources",
          "tone",
          "privacy"
        ],
        "notes": "Initial cross-agent review against the work package and source map. No private or proprietary material detected. Claim markers and source IDs aligned. Approved for publication after final review."
      },
      {
        "reviewer": "human",
        "reviewedAt": "2026-06-20",
        "status": "approved",
        "scope": [
          "claims",
          "sources",
          "tone",
          "privacy"
        ],
        "notes": "Human final review approved for publication after sibling-agent review and CI pass.",
        "contentHash": "5bd1220dab68c934e4e765ac3c87dfb2054668731c5f0c907c9732454242253e"
      }
    ],
    "policy": {
      "id": "policy:default",
      "version": "1.0.0"
    }
  },
  "generatedAt": "2026-06-29T00:00:00.000Z",
  "articleUrl": "https://aura-knowledge.github.io/articles/learning-machines/",
  "agentJsonPath": "/agents/articles/learning-machines.json",
  "agentMarkdownPath": "/agents/articles/learning-machines.md",
  "sourceRepoPath": "content/articles/2026/learning-machines/article.md",
  "sourceGitHubUrl": "https://github.com/aura-knowledge/aura-knowledge.github.io/blob/main/content/articles/2026/learning-machines/article.md",
  "tokenEstimate": 707,
  "sectionOutline": [
    {
      "id": "the-rule-writing-limit",
      "title": "The rule-writing limit"
    },
    {
      "id": "learning-from-examples",
      "title": "Learning from examples"
    },
    {
      "id": "connections-instead-of-instructions",
      "title": "Connections instead of instructions"
    },
    {
      "id": "limits-layers-and-error-signals",
      "title": "Limits, layers, and error signals"
    },
    {
      "id": "data-becomes-infrastructure",
      "title": "Data becomes infrastructure"
    },
    {
      "id": "the-2012-demonstration",
      "title": "The 2012 demonstration"
    },
    {
      "id": "power-without-myth",
      "title": "Power without myth"
    }
  ]
}