I Like To Watch

I Like To Watch

"I like to watch / CopVision is a program that watches television. Specifically, it watches COPS on Fox ... CopVision learns its language from closed captioning subtitles transmitted in the television signal. Everything that is said on COPS is tucked away in its memory to help it understand what it's seeing. It analyzes every frame, searching the field for outlines that remind it of something it has seen before. When it recognizes a contour it tags it with a guess as to what might be going on, gathered from its experience of words and pictures that go together."

With respect to research in commonsense reasoning, I found this interesting, because one difficulty in such reasoning is collecting lots of commonsense knowledge about the world - e.g. information such as "When you are hungry, you eat." We use a tremendous amount of simple information like this in our everyday lives, and it's not obvious how to collect all of it in one place. There have been attempts to collect this information through hand-entering by experts, through passive analysis of Web text (and other text as well), through collaborative effort by Internet users, and indirectly through online games. Another possible source of low-effort commonsense collection could be through parsing the text from closed-captioned TV programs.