Guessing is Intelligence?
Have been thinking about guessing and intelligence. Mostly motivated by my work with LLMs. Partially motivated by my recent fascination with OZ Pearlman and his ability to "guess" people's PIN codes, etc. Even the game Wordle is interesting. How is it that after 3 guesses we can figure out the word? Is guessing really that bad or wrong? We're often more right than wrong when we guess -- aren't we? What is a fact? What is the truth? There is a push towards making decisions based on first principles. This process identifies that many of our underlying assumptions are wrong. It requires that we analyze the problem into facts that cannot be decomposed any further. Billion dollar business deals, significant medical diagnostics. All guessing. Guessing is all that you will get from AI. Yet people think that we must inject a human in the loop. A human who also needs to guess is the one who must manage the AI. Why? Maybe it falls back onto us who design systems to help the guessers -- who are either human or machine. Maybe we should examine what we perceive as intelligence and what we perceive as "guesses". Are they different? (And how would this relate to standardized testing?) When viewed from this lens, things such as context and values (weights) come into play. When we build an agent to do the work of a human who is guessing the answer to a question we should probably treat the process exactly the same. Give them the appropriate information and value set. For me, I'm starting to admire the process of formulating the best guess. Guesses are good. Explainability is king. Own your guesses.