In 2009, I launched FantasySCOTUS. At the time, the idea was novel: a league where people could predict the outcome of Supreme Court cases. I soon realized I had the makings of a nascent prediction market. We began to aggregate the votes and offer expert predictions on all of the pending cases before the Court. The best players would predict up to 85% of the cases accurately. By 2014, I partnered with colleagues to develop a machine learning algorithm that would predict Supreme Court cases. We called it {Marshall+}. At its peak circa 2016, the algorithm was nearly as accurate as our best users--approximately 70% accurate.
Admittedly, over time, interest in FantasySCOTUS has faded. I chalk it up to a few factors. First, the Supreme Court has become more predictable. With the new appointees, fewer votes are up for grabs in most cases. I think the novelty of predicting outcomes has somewhat worn off. Instead, the focus is not on affirm/reverse, but how the Court rules. Second, coding each case for our algorithm proved to be too time-consuming. The juice was no longer worth the squeeze. Third, AI has completely obliterated all of our work. It is now possible to ask an AI agent to predict a case, and that work will be far more accurate than what our algorithm might have generated with hours of preparation. I maintain FantasySCOTUS at this point mostly as a novelty. There are a dedicated group of players who still make predictions, and I am thankful for their commitment. But I doubt this league will ever return to its heyday.
In the early days, I was often asked if the Supreme Court Justices would check FantasySCOTUS. I had no way of knowing, but I would joke with reporters that Justice Kennedy might check the league to see which way to vote.
