Automated AI commentary built from foundation models
IBM first pioneered the use of AI to curate video highlight reels in 2017, work that earned the IBM Consulting team a 2023 Emmy® Award. The solution uses gesture recognition (such as fist pumps and players’ reactions), crowd noise and game analytics (such as break points) to identify highlight-worthy videos in both golf and tennis.
This year, fans can add AI-generated spoken commentary to Wimbledon highlight reels, hearing play-by-play narration for the start and end of each reel, along with key points. Fans can also turn on closed captions to further enhance accessibility, a key consideration for AELTC.
The solution is built from a foundation model developed using watsonx, IBM’s enterprise-grade AI platform designed to manage the entire lifecycle of AI models, from curating trusted data sources to governing responsible, trusted AI. Work began with watsonx.data, a data store that connects disparate data sources and allows developers to filter the data for things like profanity, hate speech or personally identifiable information. For AI Commentary, the team drew source material from nearly 130 million documents.
The data was then used to train a large language model chosen from watsonx.ai, a next-generation studio for building and training generative AI models for business use cases. The IBM team then fine-tuned the model, adding the specific domain expertise of Wimbledon, including the use of unique Wimbledon nomenclature, such as “gentlemen’s draw” rather than “men’s draw.” The final model boasts 3 billion parameters, and the team will continue to monitor its performance using governance tools, ensuring the model performs as expected.
“The watsonx platform has allowed us to quickly leverage the power of generative AI without sacrificing trust or transparency,” says Aaron Baughman, a Distinguished Engineer and Master Inventor at IBM. “This is what we mean when we say ‘AI for Business.’ To use AI in a commercial setting, you need to have confidence that a model is scalable, reliable and trusted.”
Co-creation with IBM iX
IBM iX, the experience design arm of IBM Consulting, works with the Club year-round to design, develop, maintain and secure the tournament’s website and mobile apps. The goal of this work is to continually enhance the digital experience with new, innovative features, while maintaining the tradition, beauty and design simplicity of Wimbledon itself.
Twice a year, the IBM iX and the Wimbledon team meet for workshops guided by the IBM Garage™ methodology, an enterprise design thinking collaboration that sets a roadmap for building and iterating the next generation of features that drive fan engagement. They use personas and journey maps to guide the design process, and agile development techniques to quickly iterate and build new features. In addition to the AI Commentary feature, the team is introducing several other enhancements, including:
AI draw analysis
As soon as any tournament draw is released, players and fans alike intuitively assess each player’s luck and path through the field: do they have a “good draw” or a “bad draw”? This year, IBM AI Draw Analysis helps them make more data-informed predictions by providing a statistical factor (a draw ranking) for each player in the singles draw.
The analysis leverages two previous innovations IBM built for the Club: the IBM Power Index, an AI-powered analysis of recent player performance and momentum (plus sentiment gleaned from natural language processing of media discussion by IBM Watson Discovery), and Likelihood to Win, a prediction of who will win a singles match.
The draw analysis, derived from structured and unstructured data, determines the level of advantage or disadvantage for each player and is updated throughout the day as the tournament progresses and players are eliminated. Every player has their draw ranked from 1 (most favorable) to 128 (most difficult). Fans can also click on individual matches to see a projected difficulty for that round.
Path to the final
Based on the AI Draw Analysis, users of the Wimbledon app can also see which opponent each player is most likely to play. It lists all potential matchups in the draw, ranked by the IBM Likely to Play. For each match in progress, fans can also follow the live scores and stats provided by IBM SlamTracker.
As a proven consultative process that taps into enterprise-scale automation and AI toolsets, the partnership between IBM and Wimbledon continues to deliver an innovative fan experience to millions around the world.
Source: ibm.com
Posts
0 comments:
Post a Comment