2023-06-17
Artificial Intelligence, or Automated Intelligence as I would prefer to call it, is programmed by humans.
The nature of humanity is to make errors.
Therefore Automated Intelligence makes errors.
The problem of course is that in most traditional IT systems the specific reasoning process is explicitly coded and can therefore be tested. It can even be exhaustively tested (although from experience I'd guess that's not too common due to marketing pressures!).
In AI, the reasoning process is hyper-generalised and whilst it can in theory be tested, it's not always so easy to exhaustively test (unless you have the capacity to run it in "verbose" mode and then manually follow through all the myriad of logical deductions that it relied upon - although I suppose you might be tempted to use another AI to do this ... ).
Add in the fact that in this day and age the people who can afford to write the AI software are those who also seem to expend much effort in conducting psychological operations, censoring information they don't like, and gaslighting us with all the falsehoods they want us to believe - one begins to suspect that AI may not be all it's cracked up to be.
Harvey Risch writing for the Brownstone Institute explores the situation.
As always, be careful whom (or what!) you trust.
