On Fri, Mar 29, 2013 at 10:30 AM, just camel <[email protected]> wrote: > Why do you always argue via evolution and DNA? DNA requires a vast amount of > overhead and comes with tons of evolutionary baggage that is irrelevant to > our intelligence but was/is required in order to make the system > evolve/work.
Large software projects also have a lot of useless and redundant code, test code, code for features that nobody uses, code that doesn't work, etc. When you measure cost and lines of code, all that code still counts. You can throw out all the code you don't need, but you still paid for it. Living organisms are the same way. Both humans and the microscopic bacteria eating roundworm C. elegans have the same number of genes, about 20,000. However, the human genome is 30 times larger, 3 billion bases vs. 100 million. The difference is that the exome (protein encoding regions) makes up 2% of the human genome but 70% of C. elegans. Let's assume that humans are more complex than C. elegans. Then that extra information must lie outside the exome. This is not all junk DNA. Some of it encodes binding sites for gene regulatory proteins. Some of it encodes genes that are no longer used, but could be activated by mutations. Many of our genes have multiple copies whose count can vary. C. elegans' densely packed genome, just like densely packed optimized code, is much harder to modify without breaking something. That is why it is not as highly evolved, if we can make that comparison. There is a wide range of genome sizes among related species, mostly due to copying. But in general, the lower bound increases for "higher" organisms. For mammals, the smallest genome is 2 billion bases. http://www.genomesize.com/statistics.php Human DNA has the same information content as 300 million lines of code. That does not mean we will solve AGI by writing a program that big. The way we now automate work is to write lots of little programs to do specific tasks. So it could mean automating 100,000 different tasks with 3000 lines of code each. You might point out that Watson is far simpler than my estimate. Watson does something very close to passing the Turing test. Its development was a 30 person-year effort, suggesting about 60 K lines, and certainly no more than 300 K. However Watson cannot see, hear, move, or make copies of itself. It has a good language model, as long as the input has no spelling or grammar errors. Watson cannot use discourse or nonverbal context to aid understanding as humans can. I question whether it could have defeated its human opponents without the huge advantage of an 8 ms response time to buzz in. > Also we do not have a lot of the restraints that evolution had > ... like being extremely energy efficient or caring that much about heat > dissipation or compactness and mobility. > > The power all of our brains consume is 0.4TW (50Watt * 8billion people). > The total power consumption of the human world in 2010 was 16 TW. A human brain sized neural network requires 10 MW on a 10 petaflop computer. Running 10 billion of those would raise the Earth's temperature by 12.5% (from 59 F to 123 F). Reducing transistor sizes won't fully solve the problem because feature sizes are already down to about 100 atoms wide. It will probably require a fundamentally new computer architecture, something other than silicon. Right now we have no idea what that might be. > Clearly we have way more options than evolution in terms of energy > constraints, in terms of vehicle/body constraints, in terms of a priory > intelligence. We have already discovered many solutions that improve on evolution. No bird can fly into space or carry passengers at 600 MPH. A mechanical adding machine is better at arithmetic than the brain. I can't prove that there aren't more computationally efficient ways to do everything that humans can do, just that both evolution and 60 years of AI research, both with enormous incentives, have failed to find them. -- -- Matt Mahoney, [email protected] ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657 Powered by Listbox: http://www.listbox.com
