Unintelligent Artifice

It should come as a surprise to no-one that if one trains an Artificial Intelligence on what people post to the Internet, a chatbot powered by that AI will sound like the average person on the Internet. And as anyone who has had the misfortune of spending any amount of time interacting with people on the Internet will tell you, it's not going to sound informed, compassionate or even mentally healthy. And given people's incentives to demonstrate the shortcomings of AI, prompts to bring out the worst in it are inevitable. Active, deliberate and intensive troll-testing is an absolute must under these circumstances. There is no reason why Microsoft should be burned in this a second time, given that it's a known quantity.