* This site is based on internet standards not supported by IE 6.
wJones
EC Container 5

Background: Stored-Program Computers

Background: Stored-Program Computer

Limitations of Stored-Program Computers >>

Citation

Warren Jones, Lana Rubalsky (2010) "Background: Stored-Program Computers", wJones Research, January 18, 2010
The Digital Computer at age Sixty-five
Its design, called stored-program1 was developed between 1936 and 1946. Alan Turing defined binary program logic in 1936 in a thought experiment and designed the Automatic Computing Engine2 ten years later. Presper Eckert and John Mauchly designed and built the first working stored-program systems, ENIAC and EDVAC between ’43 and ’46. John von Neumann documented and structured Eckert and Mauchly’s work3 in his 1945 First Draft Report on the EDVAC computer. His draft quickly became known as the von Neumann Architecture.

Stored-program
was a limited design, a beginning. It provided a facility to execute fixed binary program instructions. In 1943, the limitations of stored-program design did not keep computers from meeting expectations4. The ability to load and rapidly execute any code was a major leap forward5, particularly with the limited memory available in machines of the era6.

In 2010, the von Neumann Architecture remained the de facto standard for every computer made7 (i.e. supercomputer, server, laptop, smartphone) despite a two million fold increase in memory8 and comparable improvements in general machine capacity.

Turing and von Neumann would probably not have expected this. Each saw stored-program as the beginning of an evolutionary process that would lead to computers with capabilities comparable to the human brain. By 1948, just two years after ACE report, Turing would write broadly about the issues of programming machines to work more like a human brain9. von Neumann’s First Draft Report included frequent reference10 to brain-like functions. Before his death, he would prepare a series of lectures titled the Computer and the Brain11.

The main limitation of stored-program design was the system’s complete reliance on static programs. The von Neumann Architecture was designed to run on extremely limited hardware with no processing ability beyond the executing program(s). This meant the stored-program design included no facilities above the program, i.e. no ability to understand:

  1. its users’ (clients) goals,
  2. the purpose of a running program or
  3. the results from use of its resources.

Thus, the stored-program computer design included no facility to generally assist a computer user. Such computers simply ran code, unable to determine the difference between calculating the trajectory of an artillery shell12 or the decay rate of a nuclear particle. It depended completely upon a computer user, at every step along the way, to invoke the very method of its assistance.

The document you are reading was edited in a word processing program on a computer that also had a program to search the web and a program to organize files. If the computer was not constrained by the limitations of stored-program, it could understand the intentions of its client (the authors). Thus it would understand that “T”, typed in the web browser’s search13 would more likely refer to “Turing” because it is a current topic of this document, and not prompt “tornado”, the last topic searched when in the web browser.
Vision Not Realized
Having established that both Turing and von Neumann envisioned machines that would think more like people14, it is also clear that as of this date, the vision has not been realized and is seen by many as very distant. Kurzweil, director of the Singularity Institute, predicts intelligent machines won’t be a reality until 202915. Possible reasons there has not yet been an intelligent computer include:

  • An effort by Turing or von Neumann to attempt such a machine would have been fatally constrained by limitations in hardware available to them in the 1950‘s16.
  • By 1957, both Turing and von Neumann would decease.
  • Since the 1950’s, man’s ability to construct better computing machines has increased two hundred fifty million-fold17, but there’s been no comprehensive successor to stored-program logic. Efforts to build an intelligent computer were undertaken, but probably took a wrong turn in the beginning. Rather than develop a general model of intelligence and then design a computer that could run it, successors followed what they believed was Turing and von Neumann’s initial lead … to copy the human brain.
  • The brain was and largely is a mystery to science, so it was and is very difficult to take a comprehensive approach to replicating its functions.
  • Efforts to design and build artificially intelligent machines were about as successful as an effort to develop artificial flight or artificial birth control would have been. An optimal technology implementation for an intelligent machine would have been based upon fundamental principles of knowledge and understanding18.

________________________

1 also known as programmable computers
2 presented on Feb 19, 1946 to the UK National Physical Laboratory
3 copies of the Report sponsored under US Army Ordnance Department contract W-670-ORD-4926 were distributed widely on June 30, 1945, without mentioning either of the EDVAC inventors
4 such as ballistic missile calculations, the US Army’s primary goal for ENIAC
5 in the 1940‘s the term computer more widely referred to a person (typically a women) who performed mathematical calculations for the military
6 the 1945 EDVAC design stored 1024 44-bit words, or about 1KB in modern 32/64 bit systems
7 “von Neumann would recognize every machine currently in sight -- from PalmPilot organizers to supercomputers, .. as a further instance of his original architecture.” pp xii, Churchland, Churchland (2000) forward to The Computer and the Brain
8 two billion bytes today vs. one thousand bytes in 1945
9 Turing’s idea was that an initially random network can be organized to perform a specified task by means of what he describes as 'interfering training' similar to neurons in a brain (1948, p.16).
Turing, A.M. 1948. 'Intelligent Machinery'. National Physical Laboratory Report. In Meltzer, B., Michie, D. (eds) 1969. Machine Intelligence 5. Edinburgh: Edinburgh University Press, pp.3-23
In 1950, Turing further explored ideas on machine intelligence in more philosophical directions in the article Computing Machinery and Intelligence (Mind 49: pages 433-460)
10 e.g. “4.2 It is worth mentioning, that the neurons of the higher animals are definitely elements in the above sense.”
11 von Neumann prepared material for Yale’s Silliman lectures 1956. He died in 1957. His work was published posthumously as The Computer and the Brain
12 ENIAC’s first use for it’s sponsor, the US Army Ordnance Department
13 Firefox, like most programs remembers only what tasks were performed within the context of the program
14 I propose to consider the question, “Can machines think?”, Computing Machinery and Intelligence, page 1, A. M. Turing
It is an approach toward the understanding of the nervous system … mathematical techniques, the logical and the statistical aspects will be in the foreground. Furthermore, logics and statistics should be primarily, although not exclusively, viewed as the basic tools of “information theory.” Also, that body of experience which has grown up around the planning, evaluating, and coding of complicated logical and mathematical automata will be the focus of much of this information theory. The most typical, but not the only, such automata are, of course, the large electronic computing machines., The Computer and the Brain, page 1-2, John Von Neumann, Paul M. Churchland, Patricia Smith Churchland
15 Kurzweil (2005), “The Singularity is Near”
16 at the time Turing wrote, the only electronic stored-program computer in existence on either side of the Atlantic was a tiny pilot version of the Manchester Mark I.6 (The pilot version of the ACE did not run its first program until 1950.) By the time he did have access to real computing power (at the University of Manchester) his interests had shifted and he devoted his time to modeling biological growth. It was not until the year of Turing's death that Farley and Clark, working at MIT, succeeded in running the first computer simulation of a small neural network (Farley and Clark, 1954)17 a typical computer memory capacity in a large enterprise machine today is two hundred fifty billion bytes vs. the one thousand byte memory of the state of the art EDVAC in 1945
18 author’s opinion

Limitations of Stored-Program Computers >>




EC Container 6