We saw that most objects can’t be significantly compressed.
That’s disturbing! |
|
Why? |
|
|
|
|
It means that AIT is concerned with only a small proportion of all possible objects. |
|
So what? It just means that most objects are boring. |
|
|
|
|
And so, you mean that an AI program should pay attention only to compressible objects? |
|
Or only to objects that it is able to compress. |
|
|
|
|
If I follow you correctly, Kolmogorov complexity can be used to measure intelligence. |
|
?? |
|
|
|
|
|
Yes. Kolmogorov complexity is based on an ideal compressor. If making sense of data amounts to achieving compression, then KC offers an upper limit of what intelligent processing can do. |
|
Ah, OK... I see.
Personally, I felt more concerned by the practical side of it. |
|
|
|
|
You mean, actual compression using Zip? |
|
Not only. It’s nice to know that we can assess complexity through frequency measures. |
|
|
|
|
Yes. It connects Kolmogorov complexity to Shannon’s information. |
|
And using a search engine to compute semantic distance is really cool! |
|
|
|
Ah, the story about the universal distance and the "Google distance"? Yes, nice!
But my favorite is still the story about Zipf’s law. |
|
Oh yes. Incredible, isn’t it? The law is so simple: F = k/r.
And it applies to everything we write! |
|
|
|
|
It seems to make sense, though.
I mean, when you look at it in terms of minimal description. |
|
Yes. It suggests that Kolmogorov complexity is not only a tool.
It explains things. |
|
|
|
|
