Go: machine AlphaGo against human Lee Sedol
Machines passed top human Go players on 15 March 2016, when DeepMind’s AlphaGo beat Lee Sedol 4–1 over five games on a full board with no handicap, refereed and broadcast live. Overtaq records a confirmed crossover. A 2023 result, in which an amateur beat leading engines using an adversarial policy, shows machine superiority in Go is not uniform across all positions.
The record in full
| Discipline | Go BOARD GAME · 19×19 |
|---|---|
| What is measured | Match play against top-ranked professional players, no handicap |
| Human benchmark | Lee Sedol — Lee Sedol, 9 dan · 18 world titles SOUTH KOREA · 15 MARCH 2016 · GOOGLE DEEPMIND CHALLENGE MATCH, SEOUL |
| Machine benchmark | AlphaGo — AlphaGo 15 MARCH 2016 · GOOGLE DEEPMIND CHALLENGE MATCH, SEOUL |
| Attributed to | Google DeepMind UNITED KINGDOM · UNITED STATES |
| Difference | No single comparable number. The machine passed the best humans in a contest with an unambiguous result. |
| Crossover | 15 March 2016 — AlphaGo 4–1 Lee Sedol, Seoul AlphaGo beat a top professional 4–1 over five games with no handicap. |
| Verification status | VERIFIED Verified |
| Comparability status | DIRECT Directly comparable |
| Overtaq classification | CONFIRMED A machine passed the human benchmark in a contest or measurement Overtaq judges both well evidenced and legitimately comparable. |
| Data last checked | 23 AUGUST 2026 |
Conditions and comparability
Machine conditions
Five games, full board, no handicap, standard time controls, refereed, broadcast live.
Why this is directly comparable
The cleanest crossover in the dataset. Same rules, same board, no handicap, independent referees, a public result. The caveat is not the match but the years since: in 2023 an amateur player beat top open-source Go engines repeatedly by exploiting an adversarial weakness, which shows machine superiority in Go is not uniform across all positions.
Machine progression
6 recorded results. Overtaq keeps every published result rather than only the current best, so the frontier can be read as a history. This table is the full data behind the chart on the interactive record.
| Date | Result | System | Status | Note |
|---|---|---|---|---|
| 9 October 2015 | AlphaGo 5–0 Fan Hui | AlphaGo Fan DeepMind |
VERIFIED DIRECT | First machine win against a professional player on a full board without handicap. |
| 15 March 2016 | AlphaGo 4–1 Lee Sedol CROSSING | AlphaGo Lee DeepMind |
VERIFIED DIRECT | The crossover. Lee Sedol won game four — the last game a top human has won against a leading Go program in a formal match. |
| 4 January 2017 | 60–0 online, as "Master" | AlphaGo Master DeepMind |
VERIFIED BROAD | Played anonymously against top professionals on internet Go servers and won every game. Fast time controls, online play. |
| 27 May 2017 | AlphaGo 3–0 Ke Jie | AlphaGo Master DeepMind |
VERIFIED DIRECT | Defeated the world number one in Wuzhen. DeepMind retired AlphaGo from competition afterwards. |
| 18 October 2017 | AlphaGo Zero 100–0 AlphaGo Lee | AlphaGo Zero DeepMind |
VERIFIED BROAD | Trained from self-play alone with no human game data, and beat the version that beat Lee Sedol in every game. |
| 17 February 2023 | Amateur beats KataGo via adversarial policy | KataGo Open source |
REPORTED MODIFIED | A human amateur, guided by an adversarial policy discovered by researchers, won repeatedly against top-level engines. A counter-example to uniform machine superiority, not a reversal of the crossover. |
Evidence
4 sources. Every source is a direct link to the primary record where one exists; where Overtaq has not been able to confirm an exact permanent URL, the entry says so rather than linking to a search page.
- Nature 529, 484–489 — Mastering the game of Go with deep neural networks and tree search
28 JANUARY 2016 · DOI:10.1038/nature16961
AlphaGo’s architecture and the Fan Hui match; the Lee Sedol match conditions.
- Google DeepMind — AlphaGo
27 MAY 2017
The 60–0 online run as "Master" and the 3–0 result against Ke Jie.
- Nature 550, 354–359 — Mastering the game of Go without human knowledge
18 OCTOBER 2017 · DOI:10.1038/nature24270
AlphaGo Zero’s self-play training and 100–0 result against AlphaGo Lee.
- ICML 2023 / arXiv:2211.00241 — Adversarial Policies Beat Superhuman Go AIs
17 FEBRUARY 2023
The adversarial exploit and the human amateur wins against KataGo.
What is still missing
- No public Elo scale reliably spans human professionals and modern Go engines, so Overtaq does not publish a Go rating gap.
Social cards
Every record is rendered into four formats in both themes from the same data, so the verification and comparability status travels with the numbers wherever the card is posted.
Related records
ALL RECORDS OVERTAKE TIMELINE HOW THIS IS VERIFIED SUBMIT A CORRECTION