Predicting SCOTUS

The Economist reports:

This june may be the most harried for the Supreme Court’s justices in some time. On top of 30-odd rulings due by Independence Day, the court faces a steady stream of emergency pleas. Over 16 years, George W. Bush and Barack Obama filed a total of eight emergency applications in the Supreme Court (scotus). In the past 20 weeks, as many of his executive orders have been blocked by lower courts, Donald Trump has filed 18.

Into this maelstrom, The Economist is introducing a tool to help analyse how the high court is acquitting itself under pressure. A year ago Adam Unikowsky, a regular litigator before the justices, enlisted Claude, Anthropic’s large language model (llm), to decide 37 Supreme Court cases. Claude’s decision matched the court’s 27 times. Inspired by this example, we tested several models of our own and settled on o3, Openai’s best reasoning engine for Chatgpt. We fed our scotusbot the main briefs and oral-argument transcripts for ten of the court’s biggest pending cases—plus three cases that have already been decided—and asked it to predict how each justice would vote and why.

It will be fascinating to see how well the bot does in predicting the Supreme Court decisions.

It is possible it might even end up influencing Court decisions. If you’re a Supreme Court Justice, you might not like being predictable, and could even end up varying what you decide or write, just so the AI isn’t correct!