I was at Cambridge as an undergraduate reading Mechanical Sciences for just one year. I didn't much care for the engineering but was taken with the IBM-1130 which they had installed in the Engineering Laboratories, partly for teaching undergraduates.
We were taught how to program in FORTRAN II, and prohibited from using assembly language (this was not a multiprogramming system and there was no limit to the damage a user program might do). Somehow I figured out how to assemble a program into a FORTRAN array and then contrive (by using negative indexing in the array I think) to overwrite the FORTRAN program so that on exit it jumped into the array. By this means I was able to try out interesting things which couldn't be done in FORTRAN, like getting red ink on the printer. This was a paper tape machine and the program and data tapes were punched up using these things called ``flexowriters''. Sometimes they were in such great demand that the computer itself would be sitting idle but all the flexowriters were in use. So I wrote a FORTRAN program which turned the computer into a flexowriter. However, since it was in FORTRAN when a tape was read, tabs were converted into spaces, but when a tape was punched the reverse effect did not take place. When the computer was used as a paper tape editor, the paper tape got much larger, because all the tabs were converted into multiple spaces.
This provoked the first thoughts which I can recall about AI. The editor was introducing redundancy, and I thought about eliminating redundancy. I decided that if you could write a program which would take a large set of true statements and code the documents up in the least redundant way possible, so that the paper tapes seemed completely random, then by feeding random numbers into the inverse of the compression function, you would get out random true statements.
OK so this is a completely and hopelessly naff idea, I never tried to implement it, and it obviously wouldn't work. Fortunately, I don't think it is representative of the character of my subsequent thoughts about AI, though thoughts about AI don't usually actually work, some of them are not quite so easily dismissed.
Roger Bishop Jones 2016-01-07