Matt, On Wed, Apr 10, 2013 at 7:39 AM, Matt Mahoney <[email protected]> wrote: > On Tue, Apr 9, 2013 at 12:46 PM, Ben Goertzel <[email protected]> wrote: >> Matt, as you well know, OpenCog is not trying to emulate the human >> brain in detail, so that detailed calculations of computational >> requirements based on the human brain are not directly relevant > > Ben, as you well know, my estimate is not for a detailed brain > emulation. It is for a neural network with approximately the same > number of synapses and response time. Each synapse requires one > multiply and one add operation per cycle, or 2 of each if updated. > That is a vast simplification of the behavior of real neurons, but > sufficient to get equivalent high level behavior according to > generally accepted theory.
Yeah, but OpenCog is not a neural network, and neither is DeSTIN... So detailed component by component comparisons with neural networks are not really useful... > It seems very convenient that your estimate of computing power for AGI > exactly matches the computing power you have available. I don't have an exact estimate, so my estimate can't exactly match anything... >And yet people > with thousands of times more computing power have only been able to > solve much easier problems like recognizing cat faces or winning at > Jeopardy. While computing power is important, having the right structures and algorithms is important too... A giant rock has many more molecules than a human brain, but is not smarter ;-p ... > So I would really like to know how many nodes and links (OpenCog, > DeSTIN, or otherwise) you need for AGI. How many concepts and how many > relations? How do you get these numbers? I explained my estimate of > 10^7 features and 10^13 relations for vision. You need to explain why > those numbers are wrong. Those numbers are simply inapplicable, as we are not building a brainlike architecture We will proceed experimentally over the next few years, by using our system to control robots and seeing what they can do, and expanding computational resources as needed A single modern GPU card is enough to run a DeSTIN hierarchy processing a video feed with reasonably high resolution, in real time. Or so it seems. An open question is how many DeSTIN centroids are needed in each DeSTIN node, to achieve a high level of recognition ability. The answer to this question will tell us how big an OpenCog Atomspace needs to be, in order to appropriately coordinate with DeSTIN... Measuring ability to recognize objects and events in a real-world context, is relatively straightforward; so I think we have a clear way to proceed... >Intuition is not an explanation. Like pretty much all researchers, we are proceeding in part based on intuition, yeah Fortunately, we don't need to convince skeptics in order to proceed with our R&D ... If scientists were banned from proceeding based on intuition, until they had convinced skeptics of their methodology and ideas, nearly all science would halt... -- Ben G ------------------------------------------- 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
