An MIT study found AI assistants improved misinformation detection in the moment, but appeared to weaken users' ability to spot falsehoods on their own.