lectures by von Neumann on “High
Speed Computing.” They provide a
more complete and explicit discussion
of the connection of Turing’s 1936 paper to the design of actual computers
than the documents historians have
previously relied upon. We believe they
predate any prior documented reference by von Neumann to Turing’s 1936
paper. Through not dated, from internal evidence and comparison with other documents from the period, we conclude the text dates from mid-to-late
1945; we justify that assertion below.
As shown in Figure 1, Lecture 3 began with the words “The problem of
developing a computing machine can
be considered as a problem in logic,”
which in this context referred to “logical
control” or the automatic sequencing of
operations. Von Neumann’s approach
to computer architecture was deeply
shaped by his background in logic. His
plan for EDVAC was a simpler, cleaner,
and more practical design than any of
the earlier attempts to build a general-purpose automatic computer.
The text shows that von Neumann
knew Turing’s 1936 paper and fully appreciated the significance of the universal machine described in it. He first
described Turing’s machine concept
and its connection to the question of effective calculability: “We shall consider
two systems of logic which could be used
in building a computing machine. The
first, developed by Turing, is essentially
a logic machine. Turing considered setting up a mathematical apparatus for the
decidability of mathematical problems.
More specifically he was interested in determining when an arithmetic function
can be constructed. Instead of treating
problems in the usual fashion of starting
with a set of assumptions and then proving theorems, Turing set up a hypothetical machine to construct the function.
“The Turing machine consists of
two parts; one is permanent, and the
other—the recording medium—can
be changed … It is composed of a long
paper band with symbols recorded and
an apparatus to sense these symbols,
put on new ones, and erase old ones.
There are a finite number of states of
the machine. Let the range of the indi-
cation i of those states be i = 1, 2, …, N.
Let the state of a square of tape be j,
where j = 1, 2, …, M. At every moment
the machine inspects the tape and then
had initiated computer-building proj-
ects in complete ignorance of Turing’s
work. But such claims underline the
importance of finding out what, if any-
thing, von Neumann felt in 1945 about
the relevance of Turing machines to
Before proceeding to answer that
question, we should acknowledge another controversy. Even those historians
of early electronic computing who see
the First Draft as a crucial and original
document have disagreed about whether
its key ideas should be credited to von
Neumann or to the original ENIAC
team. It is common to read claims that
von Neumann was merely writing up
ideas formulated by J. Presper Eckert
and John Mauchly. Those who give full
credit to the ENIAC design team often
focus on the computer as a product of
innovations in electrical engineering.
Those, including Arthur Burks, an-
other of the ENIAC team, who felt von
Neumann made a crucial contribution,
point to his abstraction from engineer-
ing details to produce the first coher-
ent proposed architecture for EDVAC.
Surviving evidence is inconclusive, but
in our book ENIAC in Action we did our
best to plausibly divide credit for differ-
ent aspects of the EDVAC design.
“High Speed Computing,”
by John von Neumann
After the completion of ENIAC in Action, one of us (Priestley) returned to
the archive of Herman Goldstine’s
papers at the American Philosophical
Society in Philadelphia. Goldstine had
been von Neumann’s closest collaborator within the ENIAC group, and
chose to have the First Draft typed up
and widely distributed.
Hiding in Goldstine’s papers was
the typescript of a series of three short
Figure 2. After discussing Turing, von Neumann moved on to explain the use of abstract
neurons to represent digital switching circuits.