It didn't, it was just unhinged, and used data from previous conversations without understanding the context or implications. Users also baited Tay into saying certain responses, and then claimed that it said them on its own.