Three AI systems with documented, checkable results show how differently famous technologies work. IBM's Deep Blue beat world chess champion Garry Kasparov 3.5 to 2.5 in May 1997 using special-purpose search hardware, IBM's Watson won a televised Jeopardy! match in February 2011 by combining many answer-finding components, and the Transformer, described in a June 2017 paper, is a neural network design for language that was tested on translation.
Key takeaways
- Fame and method are separate questions: a system can be widely known and still be narrow, as Deep Blue was with chess.
- Deep Blue evaluated about 200 million chess positions per second, according to IBM's historical account, and its designers list search and a complex evaluation function among the reasons it won.
- Watson answered Jeopardy! clues without an internet connection, using a pipeline described in a 2010 AI Magazine paper.
- The 2017 Transformer paper reported 28.4 BLEU on English-to-German translation, more than two points above the best earlier results.
- Each system is a milestone for one task, not evidence of general intelligence.
Deep Blue: search at machine speed
Deep Blue played Kasparov twice. In Philadelphia in February 1996 Kasparov won the six-game match 4 to 2, although Deep Blue took the first game, the first computer win against a reigning world champion under regular time controls. In the May 1997 rematch in New York, Deep Blue won 3.5 to 2.5: Kasparov took game one, Deep Blue game two, games three to five were draws, and Deep Blue won game six.
IBM's account says the machine evaluated 200 million positions per second. The designers Murray Campbell, A. Joseph Hoane Jr., and Feng-hsiung Hsu, in their 2002 paper in the journal Artificial Intelligence, list a single-chip search engine, a multi-level parallel system, heavy use of search extensions, a complex evaluation function, and a grandmaster game database. Chess is a fully observable, deterministic game, which suits exhaustive search.
Watson: many components, one answer
Watson's Jeopardy! match ran over three days in February 2011 against Brad Rutter and Ken Jennings, the show's two most successful champions. Watson finished with USD 77,147, against USD 24,000 for Jennings and USD 21,600 for Rutter, according to IBM's historical account. The original machine filled a room: 10 racks, 90 servers, and 2,880 processor cores, and it was not connected to the internet during play.
The design is documented in Building Watson: An Overview of the DeepQA Project, published in AI Magazine in the Fall 2010 issue by David Ferrucci and colleagues. Its abstract says a core team of about 20 researchers worked on it for roughly three years, and that the aim was a live contestant matching top human players in accuracy, confidence, and speed. The architecture is a pipeline in which many algorithms propose and score candidate answers, not a single learned model.
The Transformer: a design, not a product
The third entry is different in kind. "Attention Is All You Need," by eight authors led by Ashish Vaswani, appeared on arXiv on June 12, 2017. The paper's abstract reports 28.4 BLEU on the WMT 2014 English-to-German task, improving on the best existing results, including ensembles, by over 2 BLEU, and a single-model record of 41.8 BLEU on English-to-French after 3.5 days of training on eight GPUs. BLEU is an automatic score that compares machine translations with human reference translations.
A Transformer is an architecture that other teams can train on other data. That is why its fame is indirect: the paper has no consumer product attached, and the chat tools people know are built by other organizations on later versions of the idea.
Three milestones side by side
| System | Date | Task | Method in one line | What it did not show |
|---|---|---|---|---|
| Deep Blue | May 1997 | Chess match, 3.5 to 2.5 | Special-purpose chips running fast search with a hand-built evaluation | Skill at any task other than chess |
| Watson | February 2011 | Jeopardy! quiz | Pipeline of many answer generators and scorers | Understanding beyond the clue-and-answer format |
| Transformer | June 2017 | Machine translation benchmark | Neural network built around attention | Reliability outside the tested benchmarks |
Questions to ask when a technology is called famous
- What exact task was measured, and against whom?
- Does the source describe the method, or only the result?
- Was the result a single event, such as one match, or a repeated benchmark?
- Does the headline number come from the builders, and is there independent confirmation?
- Would the same system work on a different task without being rebuilt?
FAQ
Which of these is the most famous AI technology?
No source ranks them, and fame depends on audience. Deep Blue and Watson became widely known through televised contests. The Transformer is better known among researchers than the public, though much recent language software builds on it.
Was Deep Blue artificial intelligence?
It is usually counted as an AI milestone because it played a task once seen as needing human intelligence. Its method, fast search guided by an evaluation function, differs from the learning-based systems common today.
Did Watson search the web during the match?
No. IBM's account states that Watson was not connected to the internet while playing.
What does BLEU measure?
BLEU scores how closely a machine translation matches one or more human translations. It is a convenient benchmark, but a higher score does not guarantee better translations in every case.
Related reading
- How deep learning works and where it is used
- Well-known AI systems around the world
- Benefits of emerging technologies such as AI
- What measured benefits of AI look like
More in AI Basics.
Last verified: October 2026
Sources
- Deep Blue, IBM History, International Business Machines, accessed 2026
- Deep Blue, Campbell M, Hoane A J Jr, Hsu F, Artificial Intelligence 134(1-2), 2002
- Watson and Jeopardy!, IBM History, International Business Machines, accessed 2026
- Building Watson: An Overview of the DeepQA Project, Ferrucci D et al., AI Magazine 31(3), 2010
- Attention Is All You Need, Vaswani A et al., arXiv 1706.03762, 2017