
A black king and a pale knight stand on a wooden chessboard in front of a computer cabinet. AI-generated editorial illustration by AI World Today.
On May 11, 1997, IBM’s Deep Blue beat Garry Kasparov to win their six-game match, 3.5 points to 2.5. The computer used custom chips to search chess moves. It also needed a way to judge which moves were worth a closer look.
IBM describes the result as the first match win by a computer over a reigning world chess champion under standard tournament time controls. Those details matter. A single win is one game. This test measured how well the machine could play across a full match.
Kasparov won their first match
Deep Blue and Kasparov first met in 1996. The computer took the first game, and Kasparov went on to win the match. IBM came back with a new version of Deep Blue in 1997. In that match the computer won twice. Kasparov won once. The other three games were draws. The Computer History Museum’s 1997 timeline records that split.
The two matches help put the famous score in context. A year passed while the team worked on the machine. The result measures how it played in that May 1997 match. Kasparov was still the world champion. His title was separate from the score of this event.
Chess gave the team a clear test. Both players used the same rules and had to make a move within the time allowed. They could win, draw, or lose. A game also gives other people a record of every move. That mix helps explain the fascination with chess in AI’s early years.
How a chess computer looks ahead
Think about the moves you can play on your next turn. Each one gives your opponent a set of replies. Each reply gives you more choices. After a few turns, the list can get very long. Even Deep Blue needed a way to focus its time on moves that might matter.
The team described its approach in the technical paper on Deep Blue. They used chips built for chess, a way to split the search among many processors, and a set of rules for scoring positions. They drew on a database of grandmaster games. They also gave the search room to look deeper when a line of play called for it.
Why give a score to a board? A search must stop somewhere, often with neither side in checkmate. The computer still needs to decide how good the position looks. That job belongs to the scoring rules, which chess programmers call an evaluation function.
Say one move wins a pawn but leaves your king open to attack. A second keeps your king safe and sets up a threat of its own. If the search can see far enough to spot the attack, the scores can favor the second move. This made-up example shows why the pace of the search and the way it judges a board work together.
What its custom chips did
IBM says Deep Blue could assess roughly 200 million chess positions per second. That number describes chess work. The rules of the game leave too many possible paths to read to the end of every one. The team had to choose which operations to speed up and where to spend that speed.
The museum’s record of a Deep Blue chip credits Feng-Hsiung Hsu with the design of the custom chess processors. A chip built for a narrow set of tasks can do those tasks quickly. Deep Blue paired those chips with software that chose what to search and made use of the results.
Counting boards per second gives one view of the machine. Two systems may spend their time looking at very different boards. One may have rules that let it stop early on weak lines. To judge how well those choices work, it helps to look at the games themselves.
You can see the same issue in a timed game between people. One player might think through a dozen short lines, while the other spends most of a turn on a single threat. The threat can matter more than all the quiet lines put together.
The path from ChipTest to Deep Blue
Deep Blue was the work of a team over many years. A 1996 report from IBM Research traces the project back through ChipTest and Deep Thought. The team had to bring together people who knew how to build chips, write a fast search, and turn chess knowledge into code.
For a loose comparison, think of someone making a pocket calculator. People pick the keys, build the device, and decide how to handle each operation. Once those parts work, the user can press a key and get a result from the device itself. A chess program adds the hard problem of choosing among many moves.
Our Alan Turing profile traces an older question about what a machine can do by following rules. Deep Blue put that question in front of a live audience, where the rules were plain and the score was easy to follow.
Why a chess victory was a useful test
Deep Blue showed what a computer built for chess could do under match rules. The board gave it a fixed set of moves and a clear goal. A question such as how to help a lost traveler leaves far more open. The traveler may care about the time, cost, comfort, or whether a route is safe at night.
That is why it helps to ask what a test measures. Can you see the full input? Who sets the rules? Can an outside reader check the score? In chess there are good answers to all three. Those answers help us see the value of Deep Blue’s particular achievement.
From one board to the next
Nearly two decades later, AlphaGo gave the public another match to watch. Go has its own rules, and AlphaGo used trained networks to help judge a position and choose where to search. The shared idea was to look ahead. The way each system weighed its choices had changed.
It is easy to remember one famous date from the chess story. The first match, the custom chips, and the score for each game make that date worth remembering. They let us name the advance and see the work that led up to it.
For a later snapshot, read what AI could do when GTA 5 launched. More milestones are in the AI history collection.
Read also
Popular news

Microsoft brings Salesforce records and actions into Copilot

Podcast: Nvidia RTX Spark, What Local AI Keeps on Your Computer

Deutsche Telekom’s investment arm puts €100 million into Mistral

Canva lets Free and Pro users turn off generative AI

Google’s Gemini can now click through apps on a Mac



