Human interactions have a social/cultural context. You can't just recognise $thing, you have to recognise how $thing depends on $context for the correct interpretation.
Current AI either ignores context completely or doesn't parse it correctly.
It's a rediscovery of the ancient "Time flies like an arrow, fruit flies like a banana" problem.
If you're out on a date and you say "Let's go back to mine" it implies one thing. If you say it to some friends after an evening out it usually means something completely different.
Sometimes it means the same thing - but you need to know a lot about the people involved to be able to infer that accurately.
And sometimes humans can't parse these nuances accurately either.
AI-by-grep or stat bucket can't handle them at all, because the inferences are contextual and specific to situations and/or individuals. They can't be extracted from just the words themselves.
Minsky & co researched some of this in the 70s, and eventually it motivated the semantic web people. But it was too hard a problem for the technology of the time. Now it seems somewhat forgotten.
Human interactions have a social/cultural context. You can't just recognise $thing, you have to recognise how $thing depends on $context for the correct interpretation.
Current AI either ignores context completely or doesn't parse it correctly.
It's a rediscovery of the ancient "Time flies like an arrow, fruit flies like a banana" problem.
If you're out on a date and you say "Let's go back to mine" it implies one thing. If you say it to some friends after an evening out it usually means something completely different.
Sometimes it means the same thing - but you need to know a lot about the people involved to be able to infer that accurately.
And sometimes humans can't parse these nuances accurately either.
AI-by-grep or stat bucket can't handle them at all, because the inferences are contextual and specific to situations and/or individuals. They can't be extracted from just the words themselves.
Minsky & co researched some of this in the 70s, and eventually it motivated the semantic web people. But it was too hard a problem for the technology of the time. Now it seems somewhat forgotten.