Legislators were only slightly less likely to respond to constituent email messages generated by AI (15.4%) than those authored by humans (17.3%), according to new research from Cornell. A field experiment investigated how GPT-3, the predecessor to the most recent model, would fare.