Your understanding of the picture is therefore grounded in your experiences as a person in the world. Such an understanding is not possible for CaptionBot, because CaptionBot has no such grounding (nor, of course, does it purport to). CaptionBot is completely disembodied from the world, and as Rodney Brooks reminded us, intelligence is embodied. I emphasize that this is not an argument that AI systems cannot demonstrate understanding but rather that understanding means more than being able to map a certain input (a picture containing Matt Smith) to a certain output (the text “Matt Smith”). Such a capability may be part of understanding, but it isn’t by any means the whole story.
Writer - Critic - Poet - Editor