A couple of updates to my paper on the cost of AI. https://docs.google.com/document/d/1Z0kr3XDoM6cr5TgHH0GXQTjyikr7WpCkpWFn9IglW3o/edit#
1. I refined the hardware requirements by finding better estimates of the number of synapses in the brain. The cerebral cortex has 16 billion neurons with 7000 synapses each, for 112 trillion. The cerebellum has most of the brain's 86 billion neurons, but most of these are granule cells (50 billion) with 80-100 synapses each, for a total of 4-5 trillion synapses. This means that 1 petaflop and 100 terabytes of memory may be sufficient, about 1/10 of my previous estimate. (The neuron count was done by liquefying a brain and counting nuclei). Keep in mind that a 3 year old child may have several times more synapses than an adult, so the solution might not be so simple. 2. I ran more compression tests on the human genome (one lasting 40 hours) and on a large source code collection to improve both results. I estimated new upper bounds of 4.58 x 10^9 bits for the genome and 16 bits per line of code. This raises the cost estimate of the software slightly from $25 billion to $30 billion. The details are complex, so I moved them into an appendix. -- -- 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
