The AI’s success rate was up to 67 percent in the preliminary round, and in the playoff phase, it predicted the gold and bronze games correctly.
The forecast was implemented in an exceptional way: using generative AI. The most significant success was to demonstrate that generative AI can also be used as a tool for prediction.
As the championships progressed, the stumbling block for AI became an inadequate data base.
National team games are rarely played, only once a year. So the data is from a year ago or even longer.
“It was known that predicting the World Championships is more challenging than, for example, the Finnish Liiga, as there is less data, and it is somewhat outdated. In addition, in the World Championships, the playoffs are dependent on one game and not a series of matches, so the importance of a single match is great. Now we saw a lot of even matches that ended in penalties,” says Juhana Juppo, Chief Technology Officer from Digia.
Generative AI as a prediction tool - why and when?
While Digia’s previous Liiga forecast was based on a random forest model, which helps in decision-making by calculating the most likely outcome from several different decision trees, the World Championship forecast was implemented using generative AI.
Digia’s forecast for the World Championships was implemented using an RAG-based language model solution.
“Yes, this predicts, and in the preliminary round, it even seemed that the prediction accuracy exceeds coin tossing. At best, it was even 67 percent. The model learned to behave humanly from a small amount of data,” says Juhana Juppo.
So when should you consider using generative AI in making forecasts?
“When the data is not clearly numerical,” Juppo summarizes.
Good results could also bring a combination of generative AI and another model, such as a random forest model.
“In the context of ice hockey, AI based on a language model could search for information about injuries from the Lions’ announcements or information about the emotional atmosphere from social discussions. In the business world, the model could find market indications and other non-numerical data from various text materials, which helps to refine forecasts,” Juppo describes.
He reminds that AI is an everyday tool like any other technology. AI solutions can be implemented technically in different ways, and one and the same tool rarely suits all needs.
“The key is to understand what different implementation methods are suitable for, and what are their strengths and weaknesses. Continuous experimentation is important for learning and finding tools suitable for your own needs. When used correctly, AI is a very good aid for decision-making and many other needs,” says Juppo.
More information:
Juhana Juppo
Chief Technology Officer, Digia Plc
tel. +358 40 172 2859
juhana.juppo(a)digia.com