The six-game contest became one of the defining events in the history of artificial intelligence.
IBM describes the result as something no machine had previously achieved: Deep Blue defeated the reigning world champion under standard tournament time controls. The computer could evaluate as many as 200 million chess positions per second, compared with the tiny number of positions a human player could consciously examine.
But the real significance of Deep Blue was more complicated than “machine becomes intelligent.”
In The A.I. Age, AI entrepreneur and author Adam Riccoboni provides a particularly interesting interpretation of the match. Riccoboni argues that the psychologically decisive moment came when Deep Blue ceased to look like a computer to Kasparov. The machine produced moves that the greatest chess player in the world interpreted as evidence of human-like strategic insight. In Riccoboni’s words, Deep Blue had passed “a sort of Turing test”, not by convincing someone it was human through conversation, but by producing chess strategy that appeared indistinguishable from the work of a great human mind.
That distinction makes Deep Blue an important bridge in AI history:
Turing’s early chess experiments → computer chess → Deep Blue → AlphaGo → modern machine learning → foundation models and AI reasoning systems.
And there is an extraordinary twist.
The mysterious behaviour that helped convince Kasparov that the machine possessed unexpected strategic intelligence was later attributed, at least in part, to a computer bug.
The human champion may have been psychologically defeated not because he understood the computer too well, but because he attributed too much intelligence to behaviour he could not explain.
That makes the Deep Blue story not only a milestone in computing, but a story about how humans perceive machine intelligence, a question that has become even more important in the age of ChatGPT.
Deep Blue vs Garry Kasparov: the key facts
| Question | Answer |
|---|---|
| When did a computer first beat a world chess champion in a match? | May 1997 |
| Which computer won? | IBM Deep Blue |
| Who did Deep Blue defeat? | Garry Kasparov |
| What was the final score? | Deep Blue 3.5 to Kasparov 2.5 |
| Where was the match played? | New York City |
| How many games were played? | Six |
| Could Deep Blue evaluate millions of chess positions? | Yes, up to about 200 million positions per second |
| Had Deep Blue previously played Kasparov? | Yes. Kasparov defeated it 4–2 in 1996 |
| Was Deep Blue really AI? | This remains surprisingly debatable. IBM itself said its “short answer” was no |
| Why was the victory important? | It demonstrated that a machine could outperform the world’s best human at an activity long associated with sophisticated intelligence |
| What was Adam Riccoboni’s interpretation? D | eep Blue effectively passed a form of Turing Test when its strategy became indistinguishable from human strategic brilliance to Kasparov |
IBM’s own history confirms the central facts: Kasparov defeated Deep Blue in their first six-game match in 1996, but an improved Deep Blue won the May 1997 rematch 3.5–2.5.
Why was chess so important to artificial intelligence?
For much of AI history, chess was considered one of the ultimate tests of machine intelligence.
Chess appears simple.
There are 64 squares, two players and a limited number of types of pieces.
But playing chess at the highest level requires a remarkable combination of abilities:
calculation;
memory;
pattern recognition;
planning;
anticipation;
observation;
concentration;
creativity;
strategic judgment; and
the ability to predict an opponent.
As Riccoboni explains in The A.I. Age, this is why chess became one of the longest-studied subjects in artificial intelligence. He notes that researchers even described chess as the “drosophila” of AI, the equivalent of the fruit fly used so extensively in biological experimentation.
The AI pioneer Allen Newell captured the ambition behind machine chess when he argued that building a successful chess machine appeared to offer a route into the core of human intellectual activity.
If a machine could master chess, researchers reasoned, perhaps machines could eventually master many other activities associated with intelligence.
The strange prehistory of machine chess: the Mechanical Turk
The fascination with intelligent chess-playing machines began long before electronic computers existed.
In the eighteenth century, audiences across Europe marvelled at a device known as the Mechanical Turk.
The elaborate automaton appeared to play chess autonomously.
It defeated experienced players and reportedly played figures including Napoleon Bonaparte and Benjamin Franklin.
There was only one problem.
The Mechanical Turk was a fraud.
A human chess player was hidden inside the machine.
Yet the hoax had an important intellectual consequence. Riccoboni recounts how Charles Babbage encountered the Turk and became interested in the broader proposition that machinery could perform tasks previously associated with human cognition.
The dream represented by the Turk would eventually become real.
Alan Turing and one of the first computer chess programs
The history of chess AI also leads back to Alan Turing, one of the foundational figures of computer science and artificial intelligence.
Between roughly 1948 and 1952, Turing worked on a chess-playing algorithm usually known as Turochamp.
There was an unusual problem.
Turing did not have access to a computer sufficiently capable of executing the program.
So he became the computer.
Turing manually followed the algorithm’s instructions, sometimes taking around half an hour to calculate a single move.
The program was not capable of defeating strong players, but it demonstrated the conceptual possibility of representing chess strategy computationally. Riccoboni uses Turing’s experiment to illustrate just how long chess and artificial intelligence have been intellectually intertwined.
Humans remained much better than computers for decades
Early AI researchers were extremely optimistic about how quickly computers would overtake humans.
Reality proved more difficult.
In 1968, computer scientists Donald Michie and John McCarthy wagered Scottish International Master David Levy that a chess program would defeat him within ten years.
Levy accepted.
Ten years later, Levy collected the money.
Human grandmasters remained considerably stronger than the machines.
Chess programs continued improving, however, as computing power increased and programmers became more sophisticated at evaluating positions.
Eventually they encountered a human opponent who represented arguably the highest possible benchmark.
Who was Garry Kasparov?
Garry Kasparov was not merely a good chess player.
He was the reigning world chess champion and one of the most dominant players in the history of the game.
Riccoboni notes that from 1985 until his retirement in 2005, Kasparov was ranked world number one for 225 out of 228 months.
He had also repeatedly demonstrated his superiority over computers.
In 1985, the 22-year-old Kasparov simultaneously played 32 chess computers.
His score was:
Kasparov 32 to Computers 0.
He beat every one of them.
By the mid-1990s, however, computing power was advancing extremely rapidly.
IBM believed it finally had a machine capable of challenging him.
What was IBM Deep Blue?
Deep Blue was an IBM chess-playing supercomputer developed specifically to play chess at extraordinary speed.
Its central advantage was computation.
According to IBM, the 1997 system could evaluate approximately 200 million chess positions every second.
Riccoboni makes the contrast with Kasparov particularly vivid:
Deep Blue: approximately 200 million positions per second.
Kasparov: approximately three positions per second.
But this does not mean Deep Blue was 67 million times more intelligent than Kasparov.
The two possessed fundamentally different capabilities.
Riccoboni summarises the difference as:
Deep Blue had enormous calculation ability but limited chess understanding; Kasparov had enormous chess knowledge and strategic insight but limited calculation capacity.
That distinction is essential to understanding the match.
How did Deep Blue play chess?
Deep Blue did not examine every theoretically possible chess game.
That would have been computationally impossible.
Instead, it explored trees of possible moves.
For a potential move, it could calculate likely responses from Kasparov, responses to those responses, and further branches.
It then evaluated positions according to programmed criteria and selected moves assessed as advantageous.
Riccoboni describes Deep Blue as searching possible moves several “plies” into the game and evaluating factors including material gains and positional strength.
The approach was extraordinarily powerful.
But Kasparov believed it contained a weakness.
A computer could calculate vast numbers of short- and medium-term consequences.
A great human could construct a holistic strategy.
Kasparov therefore hoped to lure the machine toward superficially attractive moves that would create long-term strategic weaknesses.
It was a contest between:
brute computational force and human strategic intuition.
Kasparov had already beaten Deep Blue
The famous 1997 contest was actually a rematch.
Kasparov and Deep Blue first played a six-game match in February 1996.
Deep Blue made history by winning the opening game, the first time a computer defeated a reigning world champion in a game under regular time controls.
But Kasparov adapted.
He eventually won the match 4–2. IBM confirms that after Deep Blue’s opening victory, Kasparov recovered and defeated the computer overall.
The machine could beat the champion in a game.
It had not yet proved it could beat him in a match.
IBM spent the following year improving Deep Blue.
The 1997 rematch
In May 1997, Kasparov and an upgraded Deep Blue met again in New York.
The contest attracted enormous international attention.
Riccoboni records that Newsweek captured the atmosphere with a dramatic cover line:
“The Brain’s Last Stand.”
The implicit question was much larger than chess.
Could humanity’s greatest player still defeat the machine?
The six-game match proceeded:
Game 1: Kasparov wins
Game 2: Deep Blue wins
Game 3: Draw
Game 4: Draw
Game 5: Draw
Game 6: Deep Blue wins
Final score:
Deep Blue 3.5, Garry Kasparov 2.5. IBM had made history.
IBM did not rely on computing power alone
One of the most interesting parts of Riccoboni’s account is that Deep Blue was not simply a computer locked in a room calculating moves.
Humans played a significant role.
IBM recruited experienced chess grandmasters to help the development team.
The machine therefore embodied human chess knowledge as well as computer calculation.
IBM’s current account confirms that its team strengthened Deep Blue between the matches by improving its evaluation functions and endgame databases and bringing additional grandmasters into the project.
Riccoboni goes further into the psychology.
He describes how grandmaster advisers helped IBM understand not just chess strategy but gamesmanship.
A human player can influence an opponent through timing and behaviour.
Make a move immediately and the opponent may wonder whether you anticipated their strategy.
Pause over an easy move and they may infer uncertainty.
According to Riccoboni’s account, similar timing techniques were deliberately incorporated into Deep Blue’s play to make its behaviour less predictable and exert psychological pressure on Kasparov.
The machine did not feel psychology.
But its designers understood that its opponent did.
The enormous advantage machines have over humans: no nerves
Riccoboni highlights another difference between Kasparov and Deep Blue that has implications far beyond chess.
Machines do not become nervous.
Deep Blue did not know millions of people were watching.
It did not fear humiliation.
It did not contemplate its career.
It did not lose confidence after making a mistake.
It did not suffer adrenaline, anxiety or self-doubt.
Riccoboni compares this to elite sport. Even the greatest athletes can fail at decisive moments because humans experience pressure.
Deep Blue did not.
The machine remained computationally identical whether it was playing an informal test game or the most famous chess match in history.
This remains an important distinction between human and artificial decision-making today.
Machines have their own weaknesses.
But fear is not one of them.
The moment Deep Blue appeared human
This leads to what is perhaps the most fascinating part of Riccoboni’s interpretation of Deep Blue.
The real breakthrough was not simply that the machine calculated faster.
Kasparov knew computers calculated faster.
That did not frighten him.
The psychological transformation occurred when Deep Blue made moves that did not conform to Kasparov’s model of how a computer should behave.
Kasparov expected the computer to favour immediate, calculable gains.
Instead, Deep Blue sometimes appeared to sacrifice obvious opportunities in favour of subtle positional strategy.
Riccoboni argues that the experience created a kind of alternative Turing Test.
Alan Turing’s famous test asks whether a machine can communicate in such a way that a human cannot distinguish it from another human.
Deep Blue did something conceptually similar through chess.
It produced a strategy that the greatest chess player alive could no longer confidently classify as machine-like.
As Riccoboni puts it, Deep Blue passed “a sort of Turing test.”
Kasparov himself later described moments in which he perceived a level of machine play he had not expected. His subsequent book Deep Thinking examined not only the chess but the psychological consequences of confronting a machine whose behaviour he could not reliably interpret.
The extraordinary Deep Blue bug
The story then becomes even more interesting.
Kasparov interpreted some of Deep Blue’s strange moves as signs that the computer possessed deeper strategic understanding than he had expected.
But one famous unexplained move was subsequently attributed by Deep Blue developer Murray Campbell to a software bug.
According to the later account, Deep Blue failed to identify its preferred move within the allotted process and fell back on an emergency behaviour.
Kasparov did not know this.
He instead tried to rationalise the inexplicable move.
The machine must know something he did not.
The apparently irrational behaviour therefore became evidence, in Kasparov’s mind, of superior intelligence.
Riccoboni highlights the irony:
Kasparov had not considered that what looked like extraordinary machine intelligence might simply be an error.
Later accounts of Deep Blue developer Murray Campbell’s recollection similarly describe a bug producing a strange move that affected Kasparov’s perception of the machine.
This story deserves to be remembered alongside the victory itself.
It tells us something profound about artificial intelligence.
Humans often infer intelligence from behaviour they cannot explain.
That question is even more relevant with modern generative AI.
When an LLM generates an unexpectedly brilliant response, are we seeing reasoning?
Pattern recognition?
Statistical prediction?
Emergent intelligence?
Or behaviour whose mechanism the human observer simply does not understand?
Deep Blue confronted Kasparov with an early version of that problem.
Was Deep Blue actually artificial intelligence?
Surprisingly, IBM’s own answer at the time was:
“The short answer is No.”
IBM explained that Deep Blue relied primarily on computational power combined with search and evaluation rather than attempting to reproduce the mechanisms of human thought.
The company’s historical FAQ described it as functioning more like an extremely powerful expert system than an independently thinking intelligence.
Riccoboni makes the same important distinction in The A.I. Age.
He concludes that, despite the enormous symbolism surrounding the event, Deep Blue was fundamentally more dependent on computational power than the kind of learning intelligence associated with later AI systems.
This creates an apparent paradox.
IBM today includes Deep Blue prominently within its history of artificial intelligence and describes its victory as an inflection point that heralded a future in which supercomputers and AI could simulate elements of human thinking.
Yet IBM’s own contemporary description insisted that Deep Blue itself was not really “AI.”
Both statements can make sense.
Deep Blue was unquestionably an important milestone in the history of AI, even if its internal approach was very different from what people today mean when they discuss machine learning or generative artificial intelligence.
Deep Blue was specialised intelligence, not general intelligence
Deep Blue was extraordinarily capable at one thing:
chess.
It could defeat arguably the greatest human chess player ever.
But it could not:
recognise a face;
hold a conversation;
drive a car;
write an essay;
learn a new occupation;
explain why chess mattered;
understand that Garry Kasparov was a person;
or even understand in a human sense that it was playing chess.
This is an early and striking example of specialised or narrow machine capability.
A system can outperform every human being on Earth at one intellectual task while remaining incapable of completing elementary tasks that an ordinary child performs effortlessly.
That distinction became central to later debates about artificial general intelligence.
From Deep Blue to AlphaGo
Deep Blue was not the end of machine mastery of games.
It was the beginning of a new era.
Nearly two decades later, Google’s DeepMind produced AlphaGo.
Go presented a much more difficult computational problem for traditional brute-force approaches because of the enormous number of possible board positions.
AlphaGo therefore relied much more heavily on machine learning, neural networks and learned evaluations rather than simply expanding a search tree through raw calculation.
The transition is significant:
- Deep Blue (1997): human knowledge, programmed evaluation and immense computational search.
- AlphaGo (2016): machine learning, neural networks and search.
- Foundation models: large-scale learned representations capable of operating across many domains.
- Modern reasoning systems and AI agents: models able to analyse problems, use tools, plan sequences of actions and work across many different kinds of task.
Deep Blue demonstrated that a computer could outperform the world’s best human within a defined intellectual domain.
Later AI demonstrated that machines could learn sophisticated strategies.
Today’s central question is whether similar capabilities can generalise across thousands of intellectual domains.
Why was Deep Blue important to AI?
Deep Blue mattered for several distinct reasons.
1. A machine defeated the best human
The victory was psychologically enormous. Chess had stood for decades as a symbol of human intellect, and the reigning world champion was regarded as one of the strongest players in the history of the game. When Deep Blue won the match, the public saw, for the first time, a computer beating the best human at an activity long associated with sophisticated thinking. The event travelled far beyond the chess world and shaped popular expectations of what machines might eventually do.
2. It showed that narrow capability could exceed human performance
Deep Blue demonstrated that a system built for one task could outperform every human on Earth at that task, while remaining incapable of anything else. That combination, superhuman in one domain and helpless outside it, became the template for understanding narrow artificial intelligence, and it framed the later question of whether such capability could ever generalise.
3. It combined human knowledge with machine calculation
Deep Blue was not calculation alone. Grandmasters helped IBM shape its evaluation functions and opening and endgame knowledge, and human designers built its timing and behaviour. The machine embodied human expertise expressed in a form a computer could exploit at scale. That pattern, human knowledge encoded and then amplified by computation, recurs throughout the history of artificial intelligence.
4. It exposed how humans interpret machine behaviour
Riccoboni’s reading of the match is that its most important lesson concerned Kasparov rather than the computer. When the champion encountered moves he could not explain, he inferred a depth of understanding that, in at least one famous case, was later attributed to a bug. Humans project intelligence onto behaviour they cannot account for. That observation connects Deep Blue to ELIZA before it and to modern large language models after it.
5. It marked the end of one approach and the beginning of another
Deep Blue was the high point of search-based, hand-engineered game playing. Within two decades, AlphaGo showed that learning systems could master games too complex for brute-force search, and foundation models later extended learned capability across many domains. Deep Blue closed one chapter of artificial intelligence and, in doing so, made the next chapter visible.
What Deep Blue tells us about artificial intelligence today
Read in the age of large language models, the Deep Blue match reads less like a relic and more like a rehearsal.
The technical questions have changed. Modern systems learn rather than search; they generate rather than evaluate; they operate across thousands of tasks rather than one. But the human questions raised in New York in May 1997 remain in place.
How do we judge whether a machine understands anything? How much of what we call intelligence is simply behaviour we cannot explain? What should a person do when a machine outperforms them at something they have spent a lifetime mastering?
Kasparov faced those questions in a six-game match. Business leaders, professionals and policymakers now face them at scale.
Frequently Asked Questions
When did a computer first beat a world chess champion?
IBM’s Deep Blue defeated reigning world champion Garry Kasparov 3.5 to 2.5 in a six-game match in New York in May 1997. It was the first time a computer had beaten a reigning world champion in a full match under standard tournament conditions.
Had a computer beaten Kasparov before 1997?
Deep Blue won a single game against Kasparov in their first match in February 1996, the first time a computer had beaten a reigning world champion in a game under regular time controls. Kasparov went on to win that match 4 to 2. The 1997 match was the first that a computer won overall.
What was the score of Deep Blue vs Kasparov?
Deep Blue won the 1997 rematch 3.5 to 2.5, with two wins, one loss and three draws across six games.
Was Deep Blue artificial intelligence?
IBM’s own contemporary answer was no. Deep Blue relied on computational search and hand-built evaluation rather than learning. In The A.I. Age, Adam Riccoboni draws the same distinction, describing Deep Blue as fundamentally dependent on calculation rather than on the kind of learned intelligence associated with later systems. Deep Blue nevertheless remains one of the most important milestones in the history of artificial intelligence.
How many positions per second could Deep Blue evaluate?
According to IBM, the 1997 version of Deep Blue could evaluate approximately 200 million chess positions per second.
Why did Kasparov think Deep Blue was so intelligent?
Deep Blue made moves that did not match Kasparov’s expectations of how a computer would play, including apparently positional choices that sacrificed immediate gains. Kasparov interpreted these as evidence of deep strategic understanding. One famous unexplained move was later attributed by Deep Blue developer Murray Campbell to a software bug.
What did Adam Riccoboni say about Deep Blue?
In The A.I. Age, Riccoboni argues that Deep Blue passed “a sort of Turing test” when its play became indistinguishable, to Kasparov, from human strategic brilliance. He also highlights the role of grandmaster advisers, the deliberate use of timing to apply psychological pressure, the advantage machines hold in feeling no nerves, and the irony that behaviour Kasparov read as intelligence may have been an error.
How did Deep Blue lead to AlphaGo and modern AI?
Deep Blue showed that a machine could beat the best human in a defined intellectual domain using search and programmed evaluation. AlphaGo, developed by Google DeepMind and unveiled nearly two decades later, showed that machines could learn strategies for a game too complex for brute-force search. Foundation models then extended learned capability across many domains. Deep Blue is the first link in that chain.
Conclusion: the match that taught us how we see machines
Deep Blue’s 1997 victory over Garry Kasparov is remembered as the moment a computer first beat a reigning world chess champion in a match. It deserves to be remembered for something more.
The machine won on calculation. But the decisive moment, in Adam Riccoboni’s reading of the match, came when Kasparov could no longer tell whether he was playing a computer or something that thought like a person. Deep Blue passed a kind of Turing test at the chessboard. And at least one of the moves that persuaded him was, according to its own developer, a bug.
That combination, superhuman performance, human misreading of machine behaviour, and a gap between what a system does and what observers believe it understands, has defined every major encounter between people and artificial intelligence since. It is the story of ELIZA in the 1960s, of Deep Blue in 1997 and of large language models today.
The board has changed. The question has not.
Primary sources and further reading
IBM: Deep Blue. IBM’s own history of the Deep Blue project and the 1996 and 1997 matches against Garry Kasparov. IBM: Deep Blue
Adam Riccoboni: The A.I. Age. Riccoboni’s account of the Deep Blue match, the “sort of Turing test” interpretation and the history of chess in artificial intelligence. Published by Critical Future, 2020. Google Books: The A.I. Age
Garry Kasparov: Deep Thinking: Where Machine Intelligence Ends and Human Creativity Begins. Kasparov’s own account of the matches and their psychological consequences. PublicAffairs, 2017.
Murray Campbell, A. Joseph Hoane Jr. and Feng-hsiung Hsu: “Deep Blue”. The Deep Blue team’s technical paper in Artificial Intelligence, volume 134 (2002).
Critical Future: Artificial Intelligence Development and Strategy. The company founded and led by Adam Riccoboni. Critical Future