Hacker News Reader

Comment by _1tem | Hacker News Reader

The problem with claiming that consciousness is just a complicated computation is not merely that we simply haven't advanced our study enough to understand it. The problem is that what we know so far about the nature of computation seems to directly contradict and falsify the notion that consciousness is just computation. To put it simply: our knowledge of computational complexity, algorithms, and general problem solving shows that our human intelligence performs far more sophisticated operations than are possible given the number of neurons that exist in our brains (unless there are some super-fast algorithms that we haven't discovered yet, which would have massive implications on the P=NP problem, and our understanding of physics).

Related video about why we don't have AI yet: https://www.youtube.com/watch?v=c3of7xYoMQM

View on HN

> our knowledge of computational complexity, algorithms, and general problem solving shows that our human intelligence performs far more sophisticated operations than are possible given the number of neurons that exist in our brains

This is not true.

I wish I could respond with something more interesting, but that's all there is to it. You are just saying something that is incorrect. Our brains are information-theoretically and complexity-theoretically entirely reasonable.

> Our brains are information-theoretically and complexity-theoretically entirely reasonable.

I know this has been the traditional view of most information theorists. But many philosophers (and a few computer scientists) challenge that view. See Fodor's frame problem[1]. The basic idea is this: yes, our brains are complexity-theoretically reasonable after the problem is defined given a set of inputs, outputs, and problem states. However the real issue is how our brains formulate the problem in the first place - how to sort through the combinatorially explosive amount of information in order to narrow down what is relevant to a problem. This objection has never been satisfactorily answered by information theorists in my opinion. The answers given by info theorists always seems to presuppose the existence of a problem formulation before solving it.

[1] http://plato.stanford.edu/entries/frame-problem/

Everything in the brain is natively probabilistic. Representations which have low to zero mutual information with the current sensory input are simply not computed-with. You could say that almost all representations will have nonzero mutual information with almost all real-world sensory input, but the sensory input is precision-weighted, so it will take both a sufficiently high degree of mutual information and sufficiently high precision of the activating sense-data to make a given branch of the hierarchy (the whole model is hierarchical, and very large) relevant enough to load it into working memory and use it for prediction.

> Representations which have low to zero mutual information with the current sensory input are simply not computed-with.

Who decides that a particular representation in the brain has zero mutual information with current sensory input? This requires a comparison of some sort (i.e. computation).

It's a distributed computation that has been mostly precomputed, ie: brain connectivity. And a very popular theory of the brain says it does calculate second-order (precision/entropy) statistics about what it models.

The combinatorially explosive amount of information that we deal with at every single moment is too large to be dismissed as "precomputed by brain connectivity". If such a large amount of computation could be "precomputed", it has massive implications on our understanding of complexity-theory. Besides it contradicts our subjective feeling of free will, which is why it is not a satisfactory answer for most people.

>combinatorially explosive amount of information that we deal with at every single moment

What are you referring to? The sensory input to the brain takes, as a ridiculously high upper bound, perhaps terabits per second. In reality it's probably a few megabits. It's really not much. Vision is, I suspect, by far the highest bandwidth input to our brain, and computer vision has met or exceeded human vision in many tasks.

> our human intelligence performs far more sophisticated operations than are possible given the number of neurons that exist in our brains

This only holds if you presume those operations are being performed in the same way as a computer, rather than the incredibly imprecise rule-of-thumb heuristics they actually use.

>To put it simply: our knowledge of computational complexity, algorithms, and general problem solving shows that our human intelligence performs far more sophisticated operations than are possible given the number of neurons that exist in our brains

I'd really like to see a source for this, especially taking into account that neurons are natively probabilistic.

Reply on news.ycombinator.com