On Sun, Nov 23, 2014 at 2:29 PM, Tim Tyler via AGI <[email protected]> wrote: > Knowledge gets into brains in three main ways: from genes, > from social learning (memes) and from individual learning.
I estimate 10^9 bits of genetic knowledge, 10^9 bits of social knowledge (written down or known by someone else), and 10^7 bits of individual knowledge. It is the individual knowledge that will be the most expensive component of AGI once Moore's Law drops the cost of hardware. It takes about a year to collect. Collecting all of this knowledge (like how to do your job) from everyone in the world will cost about 1 year of global GDP. But right now it's the hardware: 10^26 OPS and 10^24 bits of memory to simulate 10^10 human brains. Global computing capacity is around 10^19 OPS and 10^22 bits of storage now. It will take 35 years of Moore's Law to reach 10^26 OPS assuming we don't run into any obstacles, like our inability to make transistors smaller than atoms or inability to achieve a 10^5 fold energy efficiency improvement to match human brains. Clock speeds have already stalled several years ago, and this was not predicted. > It is things like astronomical > observations and particle accelerators that will slow machines down. > You can't easily rush those kinds of observational learning. Medical knowledge has the same problem. It takes decades to do experiments in longevity research, so we have no idea what works and what doesn't. For example, all we know about calorie restriction is that it worked in one 20 year long experiment on monkeys but not in another. Nobody knows which drugs will add 10 years to your life if you start taking them as a child. The rate of increase of life expectancy peaked at 0.2 years per year and has been declining globally since about 1990 and in developed countries since about 1970. It would seem that computational models of the human body might solve this problem, but you can't model even simple chemistry without a quantum computer because otherwise the solution to Schrodinger's equation becomes intractable. That is why there is no software that can input a formula like H2O and output properties like the freezing point of water. Quantum computers won't speed this up much because a quantum computer is really just a physical system that behaves according to the same equation. The fastest way to model a quantum system is just to build it. What this means is there will always be jobs testing experimental drugs and interventions after every other job is automated by machines. -- -- 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
