Published in Datamation sometime in the early 1970s
Prior to now, little or no valid work has been done on the psychological aspects of computer programming. The problem is, not supringly, economic. Even using that least expensive of research subjects, the graduate student, acheiving adequate sample sizes is prohibitively expensive. Moreover, the end of the military draft has sharply decreased the number of graduate students available for study. Recognizing this problem, my colleagues and I set out to study the feasibility of investigating the psychology of computer programming using nonhuman (and therefore cheaper) subjects.
It quickly became clear that a modicum of intelligence is required for the process of computer programming. A few simple tests quickly ruled out parmecia, flatworms, fruit flies and most other invertebrates as possible subjects for study. (At least for computer programming -- several very interesting experiments are underway in software project management and configuration control using both fruit flies and flatworms).
For a while, we hoped that the social insects -- ants, bees, termites -- might make suitable subjects. We were eventually forced to give up on this by our discovery (which we termed the Civil Service Effect) that the apparent structure of such insect colonies consists almost entirely of individuals working at cross purposes at that there is no discernable method to input either data or direction into the insect society.
A few experiments were made with cephalopod (octopi). with somewhat amibiguous results. Their habit of hiding in a cloud of ink whenever anything went wrong, plus their general ill nature (they bite) and the highly corrosive characteristics of their salt water environment proved to be more than we could cope with. This phase of the experiment was terminated when our prize subject -- a four foor specimen of Argonauta argo named Millicent -- removed the author's left left little finger to the first joint after 15 consecutive unsuccessful attempts to input a wet papertape through a rusty papertape reader.
We therefore elected to continue our study using that old standby, the white rat. Our initial experiments proved promising. Several of the rats were taught COBOL, which they learned (or at least, appeared to learn) with surprising ease. They quickly mastered the art of punching cards by jumping over the keyboard, although the long jump from the "3" key to the "Z" presented some problems for the smaller specimens. It was necessary to devise a latch for the numeric shift key since there is no way for a 6-inch rat to hold down a shift key which is 14 inches from the number keys which must be depressed simultaneously. We gave some thought to designing and installing a better rodent engineered keyboard, but we never got around to it.
Our initial success with the white rats proved illusory. Rats are rather short lived creatures and COBOL is, to understate the case, anything but a concise language. None of our rats lived long enough to complete even the shortest and simplest of COBOL programs. Most were still working on the DATA SECTION when they expired of old age. Attempts to teach them other programming languages were an utter failure although one rat did produce a rather extensive program in what our colleagues in the mathematics department describe as "an ALGOL-like language."
At this point, it appeared that our project had come to an unsuccessful conclusion. However, just as we started to wrap it up, I was contacted by a gentleman from an unidentified government agency. This man, known to us only as "Charlie," expressed an interest in our project. After a series of meetings in the back booths of obscure road houses, a long, if chilly, consultation on a popular East Coast resort beach (in February) and an 'accidental" encounter on a Vermont ski lift, a substantial amount of money in small, used, un-marked bills was made available to us. The nature of "Charlie's" interest in our project was never revealed.
Given a fresh infusion of money, we decided to try monkeys as subjects. It should be noted, that at this point the project had gained a momentum of its own and had lost reference to its original goals. Laboratory monkeys were not cheaper than graduate students since the ASPCA and other organizations make sure that the former are adequately fed and housed while the latter can be left to shift for themselves.
A group of half a dozen monkeys was procured and was trained in COBOL, FORTRAN, and PLl. All six proved able to write programs in any of the three languages provided that they were given detailed, definitive specifications. Unfortunately, much of our time was consumed because of the primates' practice, apparently instinctive, of eating any page from the specifications which was erroneous or ambiguous. The monkeys would then sulk until one of us found the ambiguity or error and corrected it. In reviewing our work later it was recognized that we had stumbled upon a solution to the program documentation quality problem, but a majority of us felt that it was unlikely that the solution would find favor with either software designers or programmers.
Things were going well, and we had prepared a series of quite interesting experiments when disaster struck. Tired of setting up the monkeys' job decks for them (and, to be honest, piqued by their habit of jumping up and down, pointing and laughing whenever a job came back with a Job Control Language error, which happened with distressing frequency) we attempted to teach them JCL.
One monkey, a two year old female named Sarah, went into a catatonic trance from which we have been unable to rouse her. Another, who was in critical condition for four days after eating an entire JCL manual (Acco fastener and all) in a fit of rage, has recovered, but refuses to have anything to do with computers. He is currently making good progress in a mail order course in radio-tv repair.
A third monkey, Mike, has apparently concluded that JCL preparation is a random process. Although he uses the JCL manual in the initial preparation of his decks, his procedure for correcting errors consists of closing his eyes, whirling his chair around, and -- with his eyes closed -- striking random keys with his toes. Surprisingly, perhaps, this process works, but it is our subjective impression that it takes Mike at least twice as many runs as the average human requires to get a JCL deck debugged.
Two of the remaining monkeys studied the JCL manual for a few hours, and apparently decided that it was some sort of joke -- and in rather poor taste. They adamantly refused either to write their own JCL decks or to use ours. We were forced to cut them from the project. I understand that they have since formed their own company and are doing quite well on programming fixed priced contracts.
The remaining monkey not only learned JCL, he took to it so enthusiastically that he refused to work on routine programming tasks. He was able to master not only JCL, but the operating system's purported utilities which had, to that time, remained a complete mystery not only to ourselves, but to the computer center staff. For a time, we regarded his progress with admiration and pride, but he soon became a substantial annoyance reorganizing our disk structure "for greater efficiency"; "correcting" our JCL decks whenever he could get his hands on them (the corrections didn't always work, but he always had a good explanation why they should have worked); and "improving" our operating system, with unpredictable results.
Eventually we were able to control his activities by keeping our job decks under lock and key, and making it clear to him that he would be barred from the computer center if he continued to make unauthorized changes to the operating system. Nonetheless, his constant begging for small change to feed the coffee and cigarette habits he had acquired was a nuisance. None of us was really heartbroken when he expired from malnutrition, apparently as a result of his attempt to live exclusively on the food available in the computer room work area candy machines.
That summarizes our progress to date. The results of our current, and final, we feel, attempt at teaching non-humans to program. are not yet in. We have overcome the technical problems of interfacing dolphins and standard computer peripherals, largely because of the recent availability of waterproof cards. The dolphins have had no trouble learning to program, but after an initial spurt of high productivity they have shown little desire to actually do any programming.
We are considering various methods of motivating them. A consensus seems to be forming around the concept of using a conventional motivational technique. We will establish an impossible schedule, and subtly give the impression that their failure to meet the schedule is due to their own inadequacies. We will back this up with vague threats of cutting off their salaries (paid in fresh fish) if they fail to come through.
We confidently expect to be reporting on the success of this experiment within six months or so.