http://malus.exotica.org.uk/~buzz/byte/pdf/BYTE Vol 03-01 1978-01 The Brains of Men and Machines.pdf
After reading the brain article you can easily get the idea that simulating a brain with 0's and 1's is a really bad bad idea. Despite the article being from 1976, it gets so much stuff right.
What a great article! I think today's so-called AI "experts" should go and read this article. Programming supercomputers based on 0's and 1's that use on/off transistors is a majorly-faulty approach to achieving any form of intelligence; no matter how rudimentary. All you can really accomplish is little more than a huge truth table as output.
The type of circuit element needed for true AI applications has not yet been invented. Nor has it been simulated with today's digital logic gates. The component needed to handle analog and digital signals in time/space/intensity dimensions, AND in varying quantities of each aspect. Sometimes it might be analog or sometimes digital in nature. Perhaps anywhere in-between the two qualities. It might have a varying frequency or amplitude, and it might take place here or there x,y,z. Moreover, the circuit component we need should be *generally* consistent but yet at the same time it must be required to deviate and make "mistakes". And, depending the behavior of its connections it should be able to make and break those connections seemingly at random. Or at least turn them on and off or strengthen and weaken them. It should also be able to connect itself to other circuit elements from time to time without outside intervention, most of the times. Many of the circuit elements in a single "block" may operate at slightly different speeds, from time to time, and they may vary their output levels, from time to time. Each element must be able to also be re-purposed by itself, or at the direction of other circuits. And last but not least, it should be able to change its own internal logic "truth-table" and sensitivity and power output slightly in most cases, but sometimes radically; all aspects being affected by the environment withint which the element resides, its own internal "bios" code, and under direction from outside signaling. The transistors sitting in your core i7 chip are nowhere near any of these requirements. In fact, modern manufacturing processes go to great lengths to ensure there is no variation between processor units, what comes out of one CPU will be exactly the same as another. Zero difference. Even an FPGA with DSP peripherals don't meet all the requirements.
As far as programming it, you wouldn't need to. Not in the sense of the traditional computer with ram and hard disk. There is no sequence of instructions that are to be followed. No way! You might do some "bios" low level core programming to get some of the regulatory stuff up and running stable. But beyond that, no. The conglomeration of circuit elements would "program" and arrange themselves over time, by themselves. How they would connect and remember and process things would be a result of the initial layout or schematic of the circuit, but only vaguely, generally, for a little while. After time, one network of elements could develop a radically different configuration. None of our current computer systems allow for that flexibility.
On another note: It is also refreshing to read articles that are not constantly quoting other work, the bibliography for this article had only like a few references. That's fantastic! Many papers I read today have a bibliography section equal in length to the original article! Fer'Chrissakes! Talk about cookie-cutter degrees being handed out.. Where's the originality?
After reading the brain article you can easily get the idea that simulating a brain with 0's and 1's is a really bad bad idea. Despite the article being from 1976, it gets so much stuff right.
What a great article! I think today's so-called AI "experts" should go and read this article. Programming supercomputers based on 0's and 1's that use on/off transistors is a majorly-faulty approach to achieving any form of intelligence; no matter how rudimentary. All you can really accomplish is little more than a huge truth table as output.
The type of circuit element needed for true AI applications has not yet been invented. Nor has it been simulated with today's digital logic gates. The component needed to handle analog and digital signals in time/space/intensity dimensions, AND in varying quantities of each aspect. Sometimes it might be analog or sometimes digital in nature. Perhaps anywhere in-between the two qualities. It might have a varying frequency or amplitude, and it might take place here or there x,y,z. Moreover, the circuit component we need should be *generally* consistent but yet at the same time it must be required to deviate and make "mistakes". And, depending the behavior of its connections it should be able to make and break those connections seemingly at random. Or at least turn them on and off or strengthen and weaken them. It should also be able to connect itself to other circuit elements from time to time without outside intervention, most of the times. Many of the circuit elements in a single "block" may operate at slightly different speeds, from time to time, and they may vary their output levels, from time to time. Each element must be able to also be re-purposed by itself, or at the direction of other circuits. And last but not least, it should be able to change its own internal logic "truth-table" and sensitivity and power output slightly in most cases, but sometimes radically; all aspects being affected by the environment withint which the element resides, its own internal "bios" code, and under direction from outside signaling. The transistors sitting in your core i7 chip are nowhere near any of these requirements. In fact, modern manufacturing processes go to great lengths to ensure there is no variation between processor units, what comes out of one CPU will be exactly the same as another. Zero difference. Even an FPGA with DSP peripherals don't meet all the requirements.
As far as programming it, you wouldn't need to. Not in the sense of the traditional computer with ram and hard disk. There is no sequence of instructions that are to be followed. No way! You might do some "bios" low level core programming to get some of the regulatory stuff up and running stable. But beyond that, no. The conglomeration of circuit elements would "program" and arrange themselves over time, by themselves. How they would connect and remember and process things would be a result of the initial layout or schematic of the circuit, but only vaguely, generally, for a little while. After time, one network of elements could develop a radically different configuration. None of our current computer systems allow for that flexibility.
On another note: It is also refreshing to read articles that are not constantly quoting other work, the bibliography for this article had only like a few references. That's fantastic! Many papers I read today have a bibliography section equal in length to the original article! Fer'Chrissakes! Talk about cookie-cutter degrees being handed out.. Where's the originality?
Last edited:
