Why non-deterministic is not enough
Right now there are a lot of discussions going around about how AI can’t be used in critical workflows because it is non deterministic, meaning for the same input it may give a different output. This is not acceptable in most situations where computers are involved, however the counter arguments are that people are non deterministic (and are already critical part of workflows), and I think that is completely true.
However that make me think, would we want a human for each and every single component of an airplane, manually pressing buttons and doing the process rather than a full deterministic system?
Because if we are equating AI Agents (LLM’s) to having human like performance, the whole benefit of AI is that it can be easily scaled. And if taken the wrong way, it may be used as a hammer to solve every problem and stick faster and smaller (Yet still non deterministic) humans in every single nook and cranny. Which is not a flight I would like to fly.