On Fri, Oct 4, 2019, 6:54 AM John Rose <johnr...@polyplexic.com> wrote:
> On Wednesday, October 02, 2019, at 11:24 AM, James Bowery wrote: > > Wolfram! Well! Perhaps you should take this up with Hector Zenil > <https://www.hectorzenil.net/main.html>: > > > Interesting: https://arxiv.org/abs/1608.05972 > Zenil, like Wolfram, wants to find the simple program that describes the universe, the long sought theory of everything. Wolfram believes it is just a few lines of code. The problem is that no computer contained in any universe can simulate its container to test the code. Lloyd estimates it would take 10^120 operations and 10^90 bits of memory to simulate the observable universe. But perhaps the code could be proven to be consistent with known physics, or supported empirically using partial simulations. Minsky said there ought to be a way to find simple programs that generate observed data most of the time, the key to all learning algorithms, even though Kolmogorov, Solomonoff, and Chaitin proved no universal procedure exists, as Minsky was aware. The human brain seems like a near universal learner, but that is because it is complex (a necessary condition of powerful learners, as proven by Legg), and a product of evolution, which selects for learning patterns important for survival and reproduction. To show that the brain is not universal, try multiplying a pair of 3 digit numbers in your head. Evolution is arguably simple, but it required 10^48 DNA copy operations on 10^37 bits to create human intelligence. Zenil's paper (I only read the abstract) describes graphs that appear complex or random but are actually algorithmically simple. But we already know that about data in general. The digits of pi are believed (but not proven) to be Borel normal (uniformly distributed in all bases) in spite of having a simple description. Likewise for encrypted or hashed data, where it is very hard to find the simple generator. ------------------------------------------ Artificial General Intelligence List: AGI Permalink: https://agi.topicbox.com/groups/agi/T8eabd59f2f06cc50-M38b59b167550370703b33e91 Delivery options: https://agi.topicbox.com/groups/agi/subscription
