After looking over some of the material over the last few weeks, I believe that a major step in the software part of the problem of building an AGI has been done and the hardware needed to run an AGI needs to be built.
I termed this a ‘pre draft’ since this is a very general outline of what is being planned.
I define an AGI as a machine that would have the abilities of a typical person and test around 100 IQ. It would be able to pass the Turning test with no problems and do anything else that a typical person could do who was limited to interacting with the real world through a computer interface or connections.
From what I can tell, the typical person has about 1,000 billion neurons with multiple connections between each and the hardware for an AGI project should target the same. The simple form is one neuron would be like a CPU with its own memory cache.
I see such a project where the hardware and software would be improved in steps, where as the hardware expanded, the software would perform better, which would then be improved with more or faster hardware; etc; until at some point the combined hardware and software would be working as an AGI. I also envision that by improvements in hardware and software beyond an AGI is that eventually the system would improve to the point that it would become a SGI (Super General Intelligence) ; a machine that would have the capability of multiple people all in one system.
My ‘general feeling’ is that the first step for hardware could be done for about $60 million to $70 million USD with about 10% of that amount needed each year for operational issues; such as lease space for the equipment, electricity and cooling, system administrators, repairs, etc. I expect the costs of such a system to decline at least 10% per year minimum as the hardware improves, becomes faster, takes less space and electricity to run and better programming makes less hardware needed.
The end result of this expenditure of money would be a system that would be an AGI and having the same abilities as one person with an IQ of 100.
I expect such a project to have at least three major iterations or changes as the project is going along. The end result of the capabilities of one person for a cost of $60 million to $70 million is not very cost effective but once one AGI was made, subsequent ones could be made at lower and lower costs over time until they were very cost efficient and give people capabilities that had never existed before in computing intelligence.
Good Luck and Take Care,
Louis J. Desy Jr. – Sunday, February 23, 2020 11:50am