[SIGCIS-Members] Firsts, the IAS computer, and Turing.

Thomas Haigh thaigh at computer.org
Wed Feb 25 12:06:34 PST 2015


So collecting my thoughts on the WP article and the discussion in this
thread...

 

The striking meta level point is maybe that people continue to care about
this stuff. The scholarly history of computing community largely moved on
from the 1940s some time ago, but with Isaacson, Dyson, Smiley etc. the
histories that reach a broad audience are still obsessed with the earliest
computers, invention, and the question of "the first computer." That, in
turn, reflects to a large extent the legacy of the ENIAC patent.

 

Another observation is that the various institutions and communities
involved do not have much interest in the subtleties of the unofficial
"truce" whereby the first generation of history of computing scholars
agreed, circa 1980, which string of adjectives went with which “first
computer.” Thus we saw repeated attempts by the UK National Museum of
Computing to call EDSAC the "first practical electronic computer." Now
Princeton CS chair Appel, perhaps confused by (George) Dyson's book, is
quoted as saying that "Turing invented computer science and the idea of the
computer, and John von Neumann built the first stored-program computer." The
usual disclaimer of course applies: people did not necessarily say exactly
what journalists quote them as saying.

 

I'll write a little below about the specifics of the two claims. The bigger
point, however, is that our attempts to insert various kinds of nuance into
the demarcation of firsts simply haven't proved compelling to the broader
audiences looking to be entertained or to celebrate the accomplishments of
particular people and institutions. That may have something to do with the
inherently pedantic nature of inserting limiting adjectives between "first"
and "computer" but it also has a lot to do with lack of respect for
historical expertise. (See
http://www.wsj.com/articles/book-review-to-explain-the-world-by-steven-weinb
erg-1423863226).  In the world of computing, getting a technical claim wrong
carries much greater consequences than getting a historical claim wrong.
Indeed, not knowing what one is talking about with respect to the history of
computing doesn’t seem to be a problem at all.

 

Specifics re von Neumann and the IAS Computer. Dyson's book, Turing's
Cathedral, does put a huge emphasis on the materiality of the IAS computer
as a watershed moment in human history. This idea was widely propagated by
reviewers who didn’t know much about the topic. For example, the NY Review
of Books review essay on the book, “How Computers Exploded,”
(http://www.nybooks.com/articles/archives/2012/jun/07/how-computers-exploded
/) began

 

The digital universe came into existence, physically speaking, late in 1950,
in Princeton, New Jersey, at the end of Olden Lane. That was when and where
the first genuine computer—a high-speed, stored-program, all-purpose
digital-reckoning device—stirred into action. It had been wired together,
largely out of military surplus components, in a one-story cement-block
building that the Institute for Advanced Study had constructed for the
purpose.

 

Here is how I explained the issue in my essay on Dyson’s book
(http://www.siam.org/news/news.php?id=2054)

 

His insistence on the IAS computer and its "fully electronic random access
storage matrix" as "as close to a point source" for the origin of the
"digital universe" as "any approximation can get" reflects an urge to
explain a particular episode as the singular origin of something vast.
Pinpointing beginnings is a primal driver of storytelling---consider the
Book of Genesis. But historians have spent decades trying to move beyond
partisan advocacy for one or another great man as the true inventor of the
computer. Looking for a point source leads to history as viewed through a
fisheye lens.

 

When discussing the influence of von Neumann on computing, historians
traditionally focus on the 1945 "First Draft Report on the EDVAC" circulated
under his name. His personal responsibility for many of the ideas set forth
in the document has frequently been disputed, but its huge influence on the
computer projects initiated over the next few years has not. Historians who
have looked more closely at the era also credit an early description of the
planned design for the IAS computer, circulated in 1946, and its early
revisions as an important influence on many of these projects. The physical,
functional computer was much less influential, in part because engineering
delays resulted in its completion only after at least one of the machines
modeled on its detailed design was already operational. One of Dyson's
idiosyncrasies is to write as if these three achievements could not be
separated, commenting relatively little on the 1945 "First Draft." He places
the full burden of universe-changing historical importance on the physical
IAS computer, which ran its first program in 1951, rather than on ideas that
many others had already embraced, and indeed extended, years earlier. 

 

Dyson boosts the historical importance of the IAS computer by omitting or
downplaying information on developments elsewhere before or during 1951,
with the exception of a tiny 1948 prototype computer at the University of
Manchester from which von Neumann's team took the memory technology. Dyson's
evidence is truthful, but startlingly incomplete. For example, while he
concedes in his introduction that "the IAS machine was not the first
computer," he never mentions EDSAC, operational at the University of
Cambridge in 1949, which historians have almost universally recognized as
the first useful computer built on the model described by von Neumann in
1945. It was also in 1949 that the Manchester team got its memory technology
working in a full-scale computer. In 1951 UNIVAC provided the first
commercially manufactured computer to the U.S. Census Bureau, while in the
U.K. J. Lyons and Company, best known for its chain of teashops, completed
its own computer and applied it to business automation. Other computers were
already operational in the Soviet Union and Australia. All those milestones
pass unmentioned by Dyson.

 

Dyson asks in conclusion, "How did the von Neumann vector manage to
outdistance all the other groups trying to build a practical implementation
of Turing's Universal Machine in 1946?" Even an attentive reader might
assume that this "outdistancing" involved winning a race rather than losing
it by several years. Dyson's implication that the various teams of computer
builders inspired by von Neumann's proposed design all saw themselves as
trying to implement the computational model described by Turing in his
now-celebrated 1936 paper "On Computable Numbers, with an Application to the
Entscheidungsproblem" is also likely to raise howls of protest from
historians who have looked at early computing.

 

Specifics re Turing

 

The idea that Turing’s 1936 paper somehow provided a crucial “stored program
concept” which von Neumann appropriated for the 1945 “First Draft of a
Report on the EDVAC” is being actively promoted and surfaces with increasing
frequency. I personally do not find it at all convincing. Here’s what I said
about it in “Actually, Turing Did Not Invent the Computer” in CACM last
year.
(http://www.tomandmaria.com/tom/Writing/CACMActuallyTuringDidNotInventTheCom
puter.pdf) 

 

Where one might leap into fantasy is by asserting the cluster of ideas
contained in von Neumann's 1945 "First Draft" are merely a restatement, or
at most an elaboration, of Turing's earlier work on computability. Judge for
yourself, by placing side by side Turing's 1936 "On Computable Numbers..."
and "First Draft of a Report on the EDVAC." They are easy to find with
Google, though you might want to pour yourself a fortifying beverage first
as neither is particularly easy reading.

 

The former is a paper on mathematical logic. It describes a thought
experiment, like Schrödinger's famous 1935 description of a trapped cat
shifting between life and death in response to the behavior of a single
atom. Schrödinger was not trying to advance the state of the art of feline
euthanasia. Neither was Turing proposing the construction of a new kind of
calculating machine. As the title of his paper suggested, Turing designed
his ingenious imaginary machines to address a question about the fundamental
limits of mathematical proof. They were structured for simplicity, and had
little in common with the approaches taken by people designing actual
machines.

 

Von Neumann's report said nothing explicitly about mathematical logic. It
described the architecture of an actual planned computer and the
technologies by which it could be realized, and was written to guide the
team that had already won a contract to develop the EDVAC. Von Neumann does
abstract away from details of the hardware, both to focus instead on what we
would now call "architecture" and because the computer projects under way at
the Moore School were still classified in 1945. His letters from that period
are full of discussion of engineering details, such as sketches of
particular vacuum tube models and their performance characteristics.

 

The phrase "stored program concept" has sometimes been used to encapsulate
the content of the "First Draft" report, but this underplays its actual
impact by implying it held just one big idea. In fact it provided a wealth
of intertwined ideas and details. In my current work with Mark Priestley and
Crispin Rope I have found it useful to separate these into three main areas.
The first, the "EDVAC Hardware Paradigm" described an all-electronic binary
computer with a much larger memory than anything ever built previously. The
second, the "von Neumann Architecture Paradigm," set out the basic structure
of the modern computer: special-purpose registers on which all operations
were performed and from which data was exchanged with main memory,
separation of arithmetic functions from control functions from memory units,
only one action performed at a time, and so on. The third, the "Modern Code
Paradigm," concerns the nature and capabilities of its instructions. For
example, instructions were expressed as through a small vocabulary of
operation codes followed by argument or address fields. These were held in
the same numbered memory cells as data. While executed by default in a
particular sequence, the machine could jump out of sequence and the
destination of this jump could be modified as the program ran based on the
state of the computation.

 

Taken together, von Neumann's cluster of ideas guided the construction of
computers that were much cheaper, smaller, more reliable, and more flexible
than their predecessors. ENIAC, the first general-purpose electronic digital
computer, used almost 18,000 vacuum tubes. The more tubes a machine held the
more expensive it was to build and, as they eventually burn out, the less
reliable. Its immediate successors held 1,000 or 2,000 tubes yet could
handle problems of greater logical complexity and were easier to program.
This efficiency made possible the construction of computers in cash-strapped
Britain following the war, and made computers affordable and useful enough
that they were rapidly turned into commercial products and applied to
business tasks as well as scientific computations.

 

According to Copeland, "the fundamental conception of the stored-program
universal computer" was Turing's. Von Neumann merely "wrote the first paper
explaining how to convert Turing's ideas into electronic form." But what
actually would have been different about von Neumann's "First Draft" report
if Turing had never written his now famous paper? My answer to that question
is: nothing (with the possible exception of the neuron notation he
appropriated to describe logic gates, whose creators cited Turing).

 

Copeland has gone so far as to argue the basic idea of a single machine that
could do different jobs when fed different instructions can be traced to
Turing. But Charles Babbage had that idea long before, and as mentioned
earlier, several computers controlled by sequential instruction tapes had
already been constructed with no influence from Turing and were well known
to von Neumann before he wrote his report. EDVAC went far beyond this to
store a program in addressable internal memory rather than on a sequential
instruction tape. To suggest this advance came from Turing is odd, as the
machine Turing described had no internal writable memory and took its
instructions from a tape. Von Neumann brought a concern with logic and
preference for minimal, general-purpose mechanisms to the design of EDVAC
but he did not need Turing to teach him that. He was a mathematic genius
with a deep pragmatic streak and an astonishing track record of productive
collaborations across a huge range of fields.

 

Turing's 1936 paper lacks many novel and fundamental features found in the
"First Draft" such as addressable memory locations. Neither did Turing
describe instruction codes followed by arguments, the building blocks of
computer programs. The suggestion that the EDVAC design was merely a
conversion of Turing's paper implies these features are trivial, and the
single important idea in each document is that code and data should be
treated interchangeably so programs can modify themselves. Yet while
Turing's paper showed one machine could, in modern terms, emulate the
functioning of another it never described a machine altering its own
instructions. Furthermore, at the very end of the "First Draft" von Neumann
expressly forbade EDVAC from overwriting the operation fields in its
instructions, even though he relied on modifications to their address fields
to accomplish basic operations such as conditional branching. This address
modification was a very influential idea in the "First Draft," but was, of
course, absent from Turing's paper as his machines did not use addresses. In
other words, the capability for unrestricted self-modifying code von Neumann
is said to have copied from Turing is something Turing did not describe and
von Neumann's design explicitly prohibited.

 

So my challenge for anyone who wants to argue that Turing’s 1936 paper was a
prerequisite for the invention of modern computer architecture is to find
specific and essential features of the modern computer which

 

1)      Are present in the 1945 “First Draft of a Report on the EDVAC.”

2)      Are present in Turing’s 1936 paper.

3)      Are not present in the designs of the Harvard Mark I, ENIAC, or the
Bell Labs relay computers, all were entirely familiar to von Neumann by
early 1945 and were designed in complete ignorance of Turing’s paper.

4)      Are not present in the work of Babbage (though I cannot remember
seeing any discussion of whether von Neumann was familiar with the specifics
of Babbage’s ideas for the analytical engine, and it seems quite likely that
he wasn’t).

5)      Are not generic features of mathematical logic with which von
Neumann would in any event have been familiar with.

 

Simply waving one’s hands vigorously while invoking a “stored program
concept” will not do.

 

How about the claim that Turing founded computer science? That’s not true in
any literal way either. As I addressed that idea in the CACM article:

 

Turing provided a crucial part of the foundation of theoretical computer
science. There was no such thing as computer science during the early 1950s.
That is to say there were no departments of computer science, no journals,
no textbooks, and no community of self-identified computer scientists. An
increasing number of university faculty and staff were building their
careers around computers, whether in teams creating one-off computers or in
campus computer centers serving users from different scientific disciplines.
However, these people had backgrounds and appointments in disciplines such
as electrical engineering, mathematics, and physics. When they published
articles, supervised dissertations, or sought grants they had to be fit
within the priorities and cultures of established disciplines. The study of
computing always had to be justified as a means, not as an end in itself.

 

Ambitious computer specialists were not all willing to make that compromise
and sought to build a new discipline. It was eventually called computer
science in the U.S., though other names were proposed and sometimes adopted.
To win respectability in elite research universities the new discipline
needed its own body of theory. The minutiae of electronic hardware remained
the province of engineering. Applied mathematics and numerical analysis were
tied too closely to the computer center tradition of service work in support
of physicists and engineers. Thus, the new field needed a body of rigorous
theory unique to computation and abstracted from engineering and applied
mathematics.

 

Turing was not, in any literal sense, one of the builders of the new
discipline. He was not involved with ACM or other early professional groups,
did not found or edit any journal, and did not direct the dissertations of a
large cohort of future computer scientists. He never built up a laboratory,
set up a degree program, or won a major grant to develop research in the
area. His name does not appear as the organizer of any of the early symposia
for computing researchers, and by the time of his death his interests had
already drifted away from the central concerns of the nascent discipline.

 

When building a house the foundation goes in first. The foundations of a new
discipline are constructed rather later in the process. Turing's 1936 paper
was excavated by others from the tradition of mathematical logic in which it
was originally embedded and moved underneath the developing new field. In
several papers historian Michael S. Mahoney sketched the process by which
this body of theory was assembled, using pieces scavenged from formerly
separate mathematical and scientific traditions. The creators of computer
science drew on earlier work from mathematical logic, formal language
theory, coding theory, electrical engineering, and various other fields.
Techniques and results from different scientific fields, many of which had
formerly been of purely intellectual interest, were now reinterpreted within
the emerging framework of computer science. Historians who have looked at
Turing's influence on the development of computer science have shown the
relevance of his work to actual computers was not widely understood in the
1940s.

 

Turing's 1936 paper was one of the most important fragments assembled during
the 1950s to build this new intellectual mosaic. While Turing himself did
see the conceptual connection he did not make a concerted push to popularize
this theoretical model to those interested in computers. However, the
usefulness of his work as a model of computation was, by the end of the
1950s, widely appreciated within large parts of the emerging computer
science community. Edgar Daylight has suggested that Turing's rise in
prominence owed much to the embrace of his work by a small group of
theorists, including Saul Gorn, John W. Carr, and Alan J. Perlis, who shared
a particular interest in the theory of programming languages. His
intellectual prominence has been increasing ever since, a status both
reflected in and reinforced by ACM's 1965 decision to name its premier award
after him.

 

By the way, “The Imitation Game” (which I have so far avoided watching)
apparently brings whole new levels of nonsense to the discussion. This time
the NYRB has taken a rather more critical stance, branding the film
“monstrous hogwash”:
http://www.nybooks.com/blogs/nyrblog/2014/dec/19/poor-imitation-alan-turing/
. 

 

Best wishes,


Tom

 

-----Original Message-----
From: members-bounces at sigcis.org [mailto:members-bounces at sigcis.org] On
Behalf Of Dag Spicer
Sent: Wednesday, February 25, 2015 9:47 AM
To: PeterEckstein at comcast.net; John Impagliazzo; Computer, SIG
Subject: Re: [SIGCIS-Members] Washington Post article about Turing

 

BTW, Merton’s original Mathew Effect paper is available here:
<http://www.garfield.library.upenn.edu/merton/matthew1.pdf>
http://www.garfield.library.upenn.edu/merton/matthew1.pdf

 

Dag

 

 

 

On Feb 24, 2015, at 5:44 PM,
<mailto:PeterEckstein at comcast.net%3cmailto:PeterEckstein at comcast.net>
PeterEckstein at comcast.net<mailto:PeterEckstein at comcast.net> wrote:

 

I consider this interview and the article to be nothing more than PR puffery
on the part of Princeton. Isaacson is quoted as saying that Turing did not
invent the automatic electronic digital computer, but somehow the article
goes on to give as much glory as possible to Princeton but never mention the
names of those Isaacson said were the real inventors--John Mauchly and
Presper Eckert. They built ENIAC before there was any clear technology in
which to store a program, and, after  the ENIAC design was frozen, they and
several others at Penn set out to develop the architecture of a
next-generation stored-program computer that they named EDVAC. At some point
von Neumann, who had never seen a computer before, joined the group and
contributed to its thinking, and he volunteered to write up their
conclusions, which were issued in a typed (and widely distributed) paper
that bore his name alone. Von Neumann was, as I believe the article calls
him, a protean genius, but attributing the stored-program concept to him is,
indeed, an example of the Matthew Effect gone wild.

 

________________________________

From: "John Impagliazzo" <
<mailto:John.Impagliazzo at Hofstra.edu%3cmailto:John.Impagliazzo at Hofstra.edu>
John.Impagliazzo at Hofstra.edu<mailto:John.Impagliazzo at Hofstra.edu>>

To: "Paul Ceruzzi" < <mailto:CeruzziP at si.edu%3cmailto:CeruzziP at si.edu>
CeruzziP at si.edu<mailto:CeruzziP at si.edu>>, "sigcis" <
<mailto:members at sigcis.org%3cmailto:members at sigcis.org>
members at sigcis.org<mailto:members at sigcis.org>>

Sent: Tuesday, February 24, 2015 2:02:42 AM

Subject: Re: [SIGCIS-Members] Washington Post article about Turing

 

Thanks, Paul.

 

Below is the link to the article for those who do not have access to it.

 
<http://www.washingtonpost.com/national/health-science/what-imitation-game-d
idnt-tell-you-about-alan-turings-greatest-triumph/2015/02/20/ffd210b6-b606-1
1e4-9423-f3d0a1ec335c_story.html>
http://www.washingtonpost.com/national/health-science/what-imitation-game-di
dnt-tell-you-about-alan-turings-greatest-triumph/2015/02/20/ffd210b6-b606-11
e4-9423-f3d0a1ec335c_story.html

 

John

 

John Impagliazzo, Ph.D.

Professor Emeritus, Hofstra University

IEEE Life Fellow

ACM Distinguished Educator

Editor-in-Chief, ACM Inroads

 

From:
<mailto:members-bounces at sigcis.org%3cmailto:members-bounces at sigcis.org>
members-bounces at sigcis.org<mailto:members-bounces at sigcis.org> [
<mailto:members-bounces at sigcis.org> mailto:members-bounces at sigcis.org] On
Behalf Of Ceruzzi, Paul

Sent: Saturday, 21 February, 2015 08:47

To: sigcis

Subject: [SIGCIS-Members] Washington Post article about Turing

 

 

On the front page of today's Washington Post is an article by Joel Achenbach
about Turing's 1936 paper and its influence on computer science. All well
and good, except later on he quotes the Chair of the Computer Science
Department at Princeton as saying "...Turing invented computer science and
John von Neumann built the first stored-program computer." An example of The
Matthew Effect ("them that's got shall have; them that's not shall lose").

 

Overall, Achenbach has written an very good summary of Turing's
contributions. He also gets one thing right (unless I am mistaken): we
really don't know to what extent von Neumann and Turing discussed these
concepts when both were at Princeton.

 

_______________________________________________

This email is relayed from
<mailto:members at sigcis.org%3cmailto:members at sigcis.org>
members at sigcis.org<mailto:members at sigcis.org>, the email discussion list of
SHOT SIGCIS. Opinions expressed here are those of the member posting and are
not reviewed, edited, or endorsed by SIGCIS. The list archives are at
<http://sigcis.org/pipermail/members/> http://sigcis.org/pipermail/members/
and you can change your subscription options at
<http://sigcis.org/mailman/listinfo/members>
http://sigcis.org/mailman/listinfo/members

 

_______________________________________________

This email is relayed from
<mailto:members at sigcis.org%3cmailto:members at sigcis.org>
members at sigcis.org<mailto:members at sigcis.org>, the email discussion list of
SHOT SIGCIS. Opinions expressed here are those of the member posting and are
not reviewed, edited, or endorsed by SIGCIS. The list archives are at
<http://sigcis.org/pipermail/members/> http://sigcis.org/pipermail/members/
and you can change your subscription options at
<http://sigcis.org/mailman/listinfo/members>
http://sigcis.org/mailman/listinfo/members

 

_______________________________________________

This email is relayed from  <mailto:members at sigcis.org> members at sigcis.org,
the email discussion list of SHOT SIGCIS. Opinions expressed here are those
of the member posting and are not reviewed, edited, or endorsed by SIGCIS.
The list archives are at  <http://sigcis.org/pipermail/members/>
http://sigcis.org/pipermail/members/ and you can change your subscription
options at  <http://sigcis.org/mailman/listinfo/members>
http://sigcis.org/mailman/listinfo/members

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.sigcis.org/pipermail/members-sigcis.org/attachments/20150225/88e0be11/attachment-0001.htm>


More information about the Members mailing list