Content Analysis computer programs still have a ways to go…
Although our computer coding skills are progressing in leaps and bounds, these content analysis computer programs are still not suited to interpret content. Interpreting (connotative) content is subjective and requires the bias mind of a human who uses their experience (and sometimes “gut” feelings) to come to conclusions. Computers on the other hand are programmed to do x when y. These are literal black and white commands boiled down to core 1s and 0s. And this coding relies on logic. Designing a computer program which deciphers language (video, photo and sound are even more difficult) becomes infinitely more complicated when the language is illogical or, at the very least, muddled. The current programs are good at accounting for instances of keywords and semantic composition because these can be easily coded (i.e. because they are logical). Even if a codebook is developed and followed to the letter there are still judgement calls which a human makes on the fly – this spontaneous thinking process cannot be replicated by a computer (yet – see Alan Turing and his famous test and claim of an impending “singularity”).
I have issue with the assumption that one can compare a modern computer program with a human in the first place (even if the computer-assisted team had the luxury of seeing the other answers first). The investigation should not be between the human and the machine but within the machine itself. What is the validity of the code and/or algorithm within this context? Have you branched the logic to account for all possible scenarios? Was the program coded for that type of content analysis? For example, if I were to claim that MS Word is terrible because it does not do sophisticated mathematical operations I would be perfectly correct. However, mathematical operations are not the purpose of Word like they are in Excel. My conclusion is correct but my expectation was a fallacy from the start.
In the end, a program is only as smart as the programmer(s) who creates it. The branching logic has to be, well, logical. The rules to branch have to be logical and, in a perfect world, should be mutually exclusive. If a researcher could account for EVERY possible way that he/she could interpret the data and the program is coded to decipher the content that way then the expectation that the computer would outperform the human could be warranted.
I’m not holding my breath for computers to outperform humans by becoming sentient but I do believe that programmers may be able to eventually code for all possiblities…and that will explode the method exponentially as well.