Little Dog: “Well, my machine gets trained through a process called reinforcement learning. It tries out various lines of conversation and sees how everyone reacts to it. For the comments that get a more positive reaction, it will reinforce that type of behavior. So, I need some experimental subjects for my robot to gain experience on.”

Big Dog: “So, let me get this straight. You’re going to potentially insult Fluffy’s family by having the robot make inappropriate comments and hope your stupid machine figures out which topics to avoid?”

Little Dog: “Well, that’s why I was hoping for some sort of agenda. That way I could initialize my machine to start with some nifty topics.”

Big Dog: “Little Dog, dinner conversations don’t work that way. Life doesn’t work that way. That’s why it’s impossible to predict the future. Canine behavior is too complex to put in a box. That’s why so-called experts are incredibly bad at predicting the future. Whether it’s the economy, or the weather, or the future price of biscuits, the experts never get it right. I have an idea. Why don’t you program your robotic dog to just sit there instead, and carefully listen? That might help it figure out how to behave.”

Little Dog: “Oh, you mean supervised learning. I suppose I could try that to start.”