AI and Common Sense

Let's show this image to an AI and ask it to infer the "common sense" behind the scene.

Q: "Why is the man in the foreground pointing at the man across from him?"

AI: "To let the waitress know that this man ordered the dish."

Q: "Why did you do that?"

AI: "Generally, waitresses don't know who's ordering what."

This is a test called Visual Commonsense Reasoning, and humans have an accuracy rate of 85%, while AI has an accuracy rate of 72%. In the not-too-distant future, this AI-equipped robot may be sitting at your table as the fourth customer.