Comments on: 06-08-25 http://www.skeptic.com/eskeptic/06-08-25/ Promoting Science and Critical Thinking Fri, 10 Feb 2012 09:58:15 +0000 hourly 1 http://wordpress.org/?v=3.3.1 By: Matt L. http://www.skeptic.com/eskeptic/06-08-25/#comment-887 Matt L. Fri, 20 Nov 2009 19:47:21 +0000 http://www.skeptic.com/eSkeptic/?p=397#comment-887 In paragraphs 12 - 14 of Connectionism, the author glosses over an important point. By utilizing object-orientation where each neuron is has the same code with different values, the code length decreases dramatically. Thus, when he points out that "A software program large enough to simulate the human brain would contain about 20 trillion errors", he glosses over the fact that "the program itself might be relatively simple, with all the complexity in the data for each neuron and each synapse." What this means is that, for testing purposes, the code could conceivably be less 2000 lines of code. Current neural network code can be found at less than 400 lines, with all the complexity contained in the values for the threshold, connection strength and the connections. Using a generous estimate of 10,000 lines of code, this translates to 200 errors. Significant, but not insurmountable. This does not necessarily mean that connectionism is feasible. Currently, we do not have information about the threshold, connections and firing intensities. But, I feel it is safe to assume that if we determine a way of reading this information from a human brain, it will involve computers. And computers are already designed to copy and manipulate data without error. In paragraphs 12 – 14 of Connectionism, the author glosses over an important point. By utilizing object-orientation where each neuron is has the same code with different values, the code length decreases dramatically. Thus, when he points out that “A software program large enough to simulate the human brain would contain about 20 trillion errors”, he glosses over the fact that “the program itself might be relatively simple, with all the complexity in the data for each neuron and each synapse.”
What this means is that, for testing purposes, the code could conceivably be less 2000 lines of code. Current neural network code can be found at less than 400 lines, with all the complexity contained in the values for the threshold, connection strength and the connections. Using a generous estimate of 10,000 lines of code, this translates to 200 errors. Significant, but not insurmountable.

This does not necessarily mean that connectionism is feasible. Currently, we do not have information about the threshold, connections and firing intensities. But, I feel it is safe to assume that if we determine a way of reading this information from a human brain, it will involve computers. And computers are already designed to copy and manipulate data without error.

]]>