I use Machine Learning, the backbone of AI. Human generated code tends to be to-the-point and have a clear objective, making it relatively easy to audit. The code employed by AI is not so straight-forward. New AI code is now written by older AI code. Currently, much of that older code was, in turn, written by multiple generations of earlier AI code. The oldest AI code employs probabilities initially generated by a neural network to predict how to write AI. Depending on what is being attempted, there are also likely multiple probabilistic models contributing to the new AI code at various locations within that new AI. As the generations of AI code start compounding, the volume of material produced over a short period of time quickly makes auditing by humans extremely difficult. Unlike human generated code (as described above), with each new generation of AI, the code becomes increasingly flaccid, and the logic increasingly convoluted (slop). As such, in order to maintain schedule and reduce cost, most AI auditing for correctness and appropriateness is now done by another AI. Computer logic is governed by Gödels' Incompleteness Theorems, and thus for mathematical reasons, the enormous volumes of computer logic as represented by an AI cannot generate reliable results consistently. Accordingly, methods such as heuristics and probabilities are employed by the AI auditor to test for hallucination and hidden bias in the new AI. So-called "human-in-the-loop" auditing happens towards the end of this process. For the reasons stated above, the actual code generated is not examined by the human. If the new AI provides about 90% accuracy (in the Gaussian context), the human-in-the-loop calls it a day. Typically, unintended consequences are not looked for, so as to mitigate costs and shorten schedules.

How does the model change for US economy now that 1) Bank Reserves can collect interest. 2) Government is purchasing shares of private companies (Intel) 3) Statistics provided by US government are becoming politicized

hello