The Blind Pi!ERROR
10.05.2005
Consider the following image:
This is the sign for a local drinkery and rathskeller, The Blind Pig. Notice the character 'g' on the word 'Pig'. It has been adjusted vertically to fit on the solid wooden beam; if it hung over, the tail of the 'g' would likely snap off for whatever reason (although the door hangs above has been sealed shut). In other words, from all outward appearances, the placement of the 'g' in this situation was based entirely off pragmatic, and not aesthetic, considerations.
Notice also that the placement of the 'g' does not change its syntactic role in the sign. It is not suddenly a capital 'G' due to its placement.
However, consider the following: a computer, designed to read characters off a sign, would come across this character in this sign and might stumble on transcription. The new character encountered does not seem to fit the expected character set. This theoretical computer expects regular (that is, digital) input, and upon encountering a deviation from that input, fails.
Such a computer is obviously impractical, for precisely that reason. But consider this: the characters before 'g' do not all fit the mold of an 'ideal' symbol. There are undoubtedly small variations in the shape and texture of the symbol that, on a more sensitive machine, might likewise produce an unexpected syntax error. Our hypothetical machine, however, is blind to these smaller deviations. Still, there are some structural properties it deems necessary for transcription, and the 'g' in this sign case violates those constraints.
To construct a machine to read signs like this, we might assume that all signs have (more or less) regular font and syntax; the novel placement of the 'g' is not something we expect to encounter, and would require an explicit rule in our machine to cover such cases. In humans, however, no such rule is required, and it is obvious to us that the assumed regularity of signs is but an artifact of the contemporary technological age; 100 years ago encountering such regularity would have been a far more surprising phenomena.
Discussion questions:
1) Suppose we construct machines that can read indefinite variations on syntax regularity- it can read signs of any form and character, including those with novel variations. It can also read nearly any example of handwriting, no matter how sloppy (within limits). I am not suggesting it can read any text and understand its meaning; rather, it simply can transcribe any instance of syntax into regular Times New Roman font. Do we consider this machine intelligent?
2) Suppose we construct another machine that can read any text and understand it completely: it can provide summaries and outlines and critical analyses of anything from Job to Joyce. However, it can only read text in Courier New font size 8, and all texts must be transcribed into that format in order to be read. Furthermore, it is extremely picky about the text itself; any smears, smudges, underlines, creases, or any other deformation of the structural properties of the text renders it totally unreadable to the machine. Do we consider this machine more or less intelligent that the machine described in 1?
3) Which do you consider the more useful machine? Which machine would you rather have? If these seem like separate questions, explain.
4) Is The Blind Pig really a rathskeller, or is it too high brow?
