Part III · Era 4 — Statistical Learning (1994–2011)

draft

AI swaps hand-written rules for patterns learned from data. Statistics and machine learning become the engine of the field, and the ingredients of the next revolution — data, GPUs and deep networks — accumulate.

Narrative

Without fanfare, AI is reborn as data science. Statistical methods (SVM, forests) surpass rule-based systems on real tasks. Chess falls to Deep Blue (1997); the web generates data at an unprecedented scale; benchmarks and competitions emerge. Hinton rehabilitates deep networks (2006), and Fei-Fei Li assembles ImageNet (2009) — the dataset that, combined with GPUs, would ignite the 2012 revolution.

Milestones

  • 1995 — *Support Vector Machines* · Cortes & Vapnik · maximum-margin

    classifier; dominates ML for a decade.

  • 1997 — LSTM · Hochreiter & Schmidhuber · solves long-term memory in

    sequences.

  • 1997 — Deep Blue beats Kasparov · IBM · first defeat of a world chess

    champion by a machine.

  • 2001 — Random Forests · Breiman · robust ensembles for tabular data.
  • 2006 — *Deep Belief Networks* · Hinton · reignites interest in deep

    networks; popularizes "deep learning".

  • 2009 — ImageNet · Fei-Fei Li · 14 million labeled images; the benchmark

    that changes computer vision.

  • 2009–2010 — GPUs for neural networks · accelerated training makes *deep

    learning* viable.

  • 2011 — Watson wins Jeopardy! · IBM · natural-language QA on national

    television.

People

Vladimir Vapnik (1936–) — *coauthor of SVMs and of statistical learning theory.* Formalized generalization (VC dimension). 🖼️ Photo F-III.10.

Jürgen Schmidhuber (1963–) — coauthor of LSTM (1997). Pioneer of recurrent networks and meta-learning.

Fei-Fei Li (1976–) — creator of ImageNet. Showed that data would be as decisive as algorithms. 🖼️ Photo F-III.11.

Yoshua Bengio (1964–) — one of the three "godfathers" of deep learning. Neural language models, attention; Turing Award 2018.

🎨 Figure F-III.12 — The tide of data. Brief: illustration of a wave formed by millions of images/records (the ImageNet), with a Deep Blue chessboard submerging on one side and GPU boards emerging on the other. Suggests the silent accumulation of inputs.

Connection to the Compendium: this era consolidates the probabilistic/ statistical paradigm (../02-types-of-ai/03-probabilistic.kmd) and sets up the Data stage of Part I as the protagonist (data-centric AI).