Putting ChatGPT on the Stand as an Expert Witness
A judge asked ChatGPT if the term "landscaping" might cover the installation of a "ground-level trampoline" 1 . He was ruling on a lawsuit involving a landscaper and his insurance company. The insurance company wouldn't pay for damages incurred by the landscaper while installing said trampoline, because the policy covered only "landscaping".
I believe AI will increasingly be used in support of legal argument. As AI gains credibility and ubiquity on the stand, we will one day realize that a rubicon has been crossed; that the age of AI has begun.
Imagine an AI tapped to answer a technical question in lieu of a human expert witness. Instead of calling a doctor to the stand to opine on the cause of death in a homicide, or a financial analyst to vet profit projections for damages in a corporate fraud case, the attorney would bring to court an AI. This AI, just like an expert witness, would be subject to voir dire—an initial questioning by attorneys from both sides to test the witness's qualifications and biases.
To get past an adversarial voir dire is no easy matter. The opposing attorney will use any and all tricks they can find to fool the AI into saying something ridiculous. It would be like an adversarial Turing test 2 . Establishing rapport with the jurors will also be difficult, amidst widespread (and justifiable) AI skepticism. But if an AI manages that feat too, and the case indeed turns on jurors being persuaded by the AI's subsequent testimony, this might herald AGI3 . It would be hard to pretend that nothing significant just happened.
