THIS!! Whenever this technology is used in an accuracy-critical setting, it is a recipe for disaster. These "AI systems" produce the MOST PREDICTED output whe

THIS!! Whenever this technology is used in an accuracy-critical setting, it is a recipe for disaster. These "AI systems" produce the MOST PREDICTED output whenever there is any ambiguity, and this REPLACES the ACTUAL UNEXPECTED value/ITEM. In information system terms, they are literally stripping out content with the highest information value. I saw this first hand giving a deposition that was supposedly transcribed by an "Artificial Intelligence" system. The output looked beautiful, until I started checking it for errors. Normally an errata sheet has a dozen lines for errors. Mine had pages. And they were substantial and very meaningful. The most obvious site was a specific name that sounded vaguely similar (started with a "J" and "K" sounds) was replaced by "Jack Kennedy". It also replaced "you have" with "I have" and omitted "not" several times, literally reversing the meaning of the statement. There were 85+ errors that I caught, over an order of magnitude more than normal. These things are basically fine as glorified type-ahead systems where it doesn't matter much and a human is in the loop. But using them in any situation where the results are critical is, at this point, basically fraud. They do not do what the proponents claim.