@ChrSzegedy@pwang@ESYudkowsky Any brain will store the necessary information it needs to recognize the difference in its perception that makes a difference in its actions. Not all information needs compression, so if you mean by compression, a lossy one, then it makes sense. Otherwise, it's total BS.
@ChrSzegedy@pwang@ESYudkowsky Lossless compression implies that the biological brain remembers everything it perceives. This is of course a myth. Therefore, the word 'compression' glosses over the notion that only what is relevant is remembered and recalled. Intelligence is about relevance realization.
@ChrSzegedy@pwang@ESYudkowsky And relevance realization is only possible relative to the subject that is performing the inference. That is, a living thing understands the world relative to the amortized knowledge that it has captured about the world.
@ChrSzegedy@pwang@ESYudkowsky Compression is an emergent property, but it is not an explanation. It's the same as entropy, an emergent property but not an explanation of how the universe evolves. It's like looking at the clouds and seeing a rabbit, it's not explanatory.
@ChrSzegedy@pwang@ESYudkowsky Furthermore, there's this magical thinking that associates compression with the simplicity of code that generates objects like Pi. Although it appears to be 'compression' in that infinite information is captured in a few symbols, it is actually not.
@ChrSzegedy@pwang@ESYudkowsky It's an artifact of reality that certain computations are reducible. It is related to computational undecidability. Certain behavior takes longer to compute (or prove) than other behavior. One can always argue that compression involvs finding short models.
@ChrSzegedy@pwang@ESYudkowsky Which is true, but finding short models to describe the world, albeit a very important ability that has led to our modern technological civilization is not the same as biological intelligence. The drive for simply explanations is not identical (but similar) as for survival.
@ChrSzegedy@pwang@ESYudkowsky A better explanation for intelligence resides in C.S.Peirce triad of induction,deduction and abduction. The notion of finding short models of the world requires all 3 inference methods. Compression is just a word that glosses over the complexities of intelligence.
@ChrSzegedy@pwang@ESYudkowsky It is no surprise then that we have not made much progress ever since this narrative of 'Intelligence is compression' was formulated.
@ChrSzegedy@pwang@ESYudkowsky In fact, a much better narrative is 'Intelligence is Error Correction'. Unfortunately, this is equally as bad for the same reasons.
• • •
Missing some Tweet in this thread? You can try to
force a refresh