Could AI Predictions Change the Future of Sports?

Northeastern University researcher explores AI’s potential in group challenges: Lorenzo Torresani aimed to test AI models’ ability to understand complex, group-oriented activities like sports, surgery, or military operations.
Finding suitable datasets: A challenge emerged as there’s a lack of proper datasets to validate advanced AI concepts like agency, planning, and causal reasoning.
Utilization of sports footage for testing: Torresani and Gedas Bertasius created a dataset with extensive sports videos and reports to evaluate AI models’ predictions on cause-and-effect.
AI performance on different capabilities: The AI excelled in basic perception (74% accuracy) but struggled with higher-order tasks like causal reasoning (40-50% accuracy) and agency (5% accuracy), which are crucial for understanding complex scenarios.
Limitations identified: AI currently excels at describing events but fails in explaining why they occur, anticipating outcomes, and suggesting actions, highlighting its limitations in real-world applications.