IBM Watson is now taking its mechanical smarts to a new passion – predicting tennis matches. Can the famous AI computer beat the odds and become a sports betting legend?
The Watson AI system may not, in reality, be making bets or generating revenue from winnings.
But for the first time ever, Watson generated its own daily player rankings and predictions, starting at the 2021 US Open.
Watson has a long been a mainstay at the New York-based WTA Grand Slam tournament. But in the past, it has focused on easier tasks befitting of a machine, such as generating fact sheets and summaries about players and other related tennis tidbits.
IBM’s move to work Watson into public player rankings presents a big step forward in the increasing popularization of natural language processing (NLP)
It also speaks to the artificial intelligence industry’s increasing confidence in the tools.
How IBM Watson compares to the competition
Each day during the US Open, Watson is producing a new iteration of its leaderboard.
IBM introduced the process at Wimbledon and is fueled by both statistical analysis and NLP, which is part of Watson Discovery.
The nuts and bolts
There are two main things that go into determining who ranks where:
- structured data (e.g., player performance specifics) and
- unstructured data (e.g., media commentary)
For structured data, more than 100,000 statistics go into how each player is ranked throughout the tournament, according to Tyler Sidell, a technical program manager at IBM.
Examples include:
- how quickly a player beats their opponent (commonly called win velocity),
- margin of victory (how close the win was based on sets and games),
- injuries, and more.
IBM declined to state how each factor or piece of data is weighted by Watson.
For unstructured data, the Watson system uses NLP to determine how the sentiment is on a certain player based on trusted news sources — like ESPN, Tennis Channel, and USOpen.org.
For example, when articles mentioned that Australian Ashleigh Barty was “tired” or “exhausted” after spending five months away from home country, this likely adversely impacted the system’s ranking of Barty.
How accurate is Watson?
Like human sports bettors, Watson is still imperfect.
For example, for the 2021 tournament, Watson had 24-year-old Japanese tennis pro Naomi Osaka — the defending 2020 US Open champ — ranked #9.
It had 19-year-old Canadian Leylah Fernandez ranked at #32.
But Fernandez surprised both the crowd and Watson after besting Osaka in the third round.
