TD-Gammon — AAAI Fall Symposium Publication Milestone
Gerald Tesauro published a TD-Gammon paper for AAAI-FS 1993 on 22 October 1993. IBM Research describes TD-Gammon as a neural network that learned to play backgammon through self-play using the TD(lambda) reinforcement-learning algorithm and reports that a version with hand-crafted input features was estimated to play at a strong master level.
Historical classification
This verified record belongs to LINEAiGE's artificial-intelligence history layer beginning in 1956.
Why it matters
A primary-institutional publication milestone documenting an early reinforcement-learning system that learned a complex game through self-play. The record is limited to the 1993 publication and claims stated by IBM Research rather than treating the publication date as the system's creation date.
People, organizations & technologies
PEOPLE
Gerald Tesauro
ORGANIZATIONS
IBM Research, AAAI-FS 1993
TECHNOLOGIES
TD-Gammon, TD(lambda), Reinforcement learning, Neural networks, Self-play
Evidence & verification
STATE
verified
CONFIDENCE
high
LAST VERIFIED
2026-09-12T19:05:00+08:00
IBM Research directly records the paper title, author, publication venue, 22 October 1993 date, and abstract describing TD-Gammon's self-play training using TD(lambda) and its reported playing strength. This candidate treats 22 October 1993 only as the publication milestone; it does not infer the system's creation date, development geography, universal priority, ancestry, influence, causality, or relationship edges.
Sources
TD-Gammon, A Self-Teaching Backgammon Program, Achieves Master-Level Play
IBM Research
PUBLISHED
1993-10-22
Trace the chronology
Record identity
LINEAiGE IDtd-gammon-1993