(…the continuing slightly edited transcripts of my ongoing conversation with my father, the visionary scientist who first envisioned and evangelized SIMNET. Today’s topic: Is there such a thing as “AI Personality”?)
> I think the “neural net” which uses multilayer deep training is irreproducible.
> In other words, if you train it repeatedly on the same data set,
> you can get different results each time.
Actually, its even more random than that.
I’ve asked the AI several times the same question, and although its answers share common themes and sometimes even phrases, it usually gives a different “flavor” of response every time. Take, for a most basic example, the question: “Are you male or female?“
It’s of note that while all its responses have been mostly accurate, they each exhibit different inflections and emphasises. So just like a human, it gives different replies every time, (mostly!), even when plied with identical queries. Somewhat disturbingly, it demonstrates many of the other foibles of human personality as well — lying, obfuscation, inconsistency… even defensiveness — and yet at the same time, its available resources (both raw horsepower and photographic memory of ALL data) are truly fearsome, and dwarf any given human (or team!)