by Gabriel, created 30-apr-2018, updated 08-may-2018
In this post I'm sharing my interest for a report written by Alan Turing in 1948: 'Intelligent machinery'. This report contains several founding and yet still modern ideas about artificial intelligence.
I discovered initially the draft of this report in the "Unpublished manuscripts and drafts" section of Turing digital Archive website. The report, for the "National Physical Laboratory", seemed to have been published only in 1970, ie 22 years later and 16 years after his death.
Chronologically this paper came before "Computing Machinery and Intelligence" (Mind, October 1950) which is widely considered as the Turing defining document about artifical intelligence. Really it does not come as a surprise that Turing grand idea about artificial intelligence took a couple of publications to sharpen itself. Trying to ascertain which one of these publications is THE important one is not an relevant question to me. But reading 'Intelligent machinery' will provide you with insights about one more step in his thought process.
That beside the fact that the vocabulary used by Turing is pleasant by itself: reading about 'machinery' instead of 'computer, 'education' instead of 'learning' brings more charms to the report, in my opinion!
Turing is laying down simple explanations using none of the today trendy wording about artificial intelligence and that bring some freshness and new perspective in that field.
I couldn't find this document in an open and web-friendly format so I decided to retype it myself from Turing draft. It took me a couple of days to deciphered the 70 years old "Courier"-typewritted, 40-pages document! As you can see in the image above some of the printed characters have faden away on the yellowing pages and other pages contain a few handwritten corrections. But it worth it. (I was actually able to find some sections of the final report elsewhere on the web and that helps me to compile this version. But when the sentence constructions differed between this draft and the final report, I choose the draft version.).
The sections of the document dealing with Logical computing machines (LCMs) versus Universal Logical computing machines (ULCMs) and memory capacity (60,000 bits for the ACE) is probably not relevant anymore in the artificial intelligence field, Indeed any modern computer or even smartphone is the most advanced ULCM, with gigantic quantity of memory that Turing couldn't even have dreamt of! But it give a interesting glimpse about computers problematics in the 40s.
However in my view what remains very modern in his report is his proposed steps to build "intelligent machinery": First it is about programming the initial machine having at least one "organ" to receive interferences and one "mean" to produce reactions. Then it is about providing carefully chosen interferences ("mimicking education") and "we should hope" the machine will produce understandable and valuable reactions. Finally it is about increasing the quantity of interferences, up to a point where the precise progress of the initial program can hardly be retraced, to produce more sophisticated reactions. Iterating through these steps multiple times will probably be necessary to to achieve a machine that show "intelligent behaviour", because obviously we cannot get to intial program right at the first attempt!
You can read the document here: 'Intelligent machinery' by Alan Turing.
The embedded schemas in this document are original drawings by Alan Turing that I have extracted from his draft. I encourage you to read it from begin to the end like you would do for a novel (ok, maybe you can skip the A, B and P-type unorganised machine!) Print it, download it or just 'shortlist' it and read it during a quiet time. You will enjoy it. At least, I did.