Our CEO, Greg Godbout, published an article about his development of an AI predictive model named "Side Bet Sally," which is designed to beat NFL point spreads. The AI model, currently in version 1.0, was created before the 2024 NFL season, and has achieved a 76.5% accuracy rate after five weeks, successfully predicting 26 out of 34 games. Sally's performance has outpaced human NFL handicappers, and the AI has generated significant profits for a small group of friends and family.
Key points from the article include:
Lift in AI Models: The concept of "lift" is discussed as a measure of how much better the AI model performs compared to random guessing. Sally's 76.5% accuracy provides a significant lift over the expected 50% accuracy of random betting.
Human in the Loop: The article highlights the importance of human oversight in AI systems, using the author as a human decision-maker to avoid bets when teams appear to have mentally "quit" on themselves.
AI Evolution and Governance: Sally is evolving from a predictive model to a more comprehensive AI product, incorporating multiple data sources, decision models, and a focus on governance and risk management to handle AI's inherent flaws.
The article also notes that future plans include automating the AI governance process to better detect when teams are not performing to expectations. Emphasis is placed on accepting AI's imperfections while using governance to mitigate risks.
For more detailed insights, you can access the full article here.