2 Comments
User's avatar
⭠ Return to thread
DC Reade's avatar

The logical fallacy that "correlation = cause" also has a long history as a tool of manipulation in the sciences, medicine, and psychology. The more orders of inference required in order to confidently assert a conclusion--and the more threadbare the foundation of factual knowledge--the more vulnerable the extant raw data is to manipulation. Correlations are closely aligned with causation in structural engineering, say, or electronics design; it's fairly easy to establish the limits of "tolerance" in those endeavors. Once the topic turns to (for example) AI, that reliable linkage is upset. Consider the "large-language model" of AI; it's entirely correlation-based! which explains how hilarious some of the declarations of AI programs can get. The more "advanced" large-language models aren't really improved in terms of their cognition; they're merely programmed to demur, deflect, and deny more often, about their ability to give an answer. The programmers of the algos at least have to be given credit for finally acknowledging how badly wrong a correlation-reliant model can potentially get, even with a stupendously massive data set to draw on.

Numbers provide an illusion of Objective Authority, because they're entirely denotative. But the most important thing is not the numbers; it's the nature of the question(s) being asked, and the limitations of the ability of the data set(s) to provide cogent answers. Ask bad, clueless, woefully incomplete, biased questions, and the accuracy of a mathematical result doesn't matter, as far as its worth as an answer.

Expand full comment
Jason Brain's avatar

Super astute and perfectly stated!

Expand full comment