On AI Comprehension…

Look at this picture. It’s funny – I get it, and so do you.

Now, mentally catalog all the different things a computer/AI would have to understand to “get” this picture – to understand this picture the same way your mind does.

It’s …sobering. Humans have a background and context of knowledge that’s awfully hard to replicate.

(I didn’t come up with this exercise. I read about it in a blog post, but I can’t find it anymore. If you know of it, post a link in the comments.)

I’m reading a book called “Structured Semantics,” which is about taking a sentence apart and trying to figure out what it means. It’s astonishing the amount of tacit knowledge we bring to understanding language. There are SO many aspects of the human condition that we just …know.

(Example: “A rock fell on John’s sandwich and crushed it. Tom had pushed it.” What did Tom push? The rock? The sandwich? John? Clearly, it’s the rock – but ask yourself how many unspoken things you had to understand to know that.)

If you had to write a sentence (or explain a picture) without assuming ANY understanding or pre-existing knowledge, it would be far more difficult than you think.