The humans best coders create potential virus exploits. One person simply changing an input diagram, or even the order of querry, will alter the returned output. Slow, fast, or incorrect responses can be inputted by switching the order of when something is querried or input’ed, and might be enought to prevent accessing the already know answer.
AI could be developed with a self repairing truth-quest borg like absorbing knowledge sifter that constantly reworks all variables in the order of knowlege acquisition acquisition. Getting a nonpolitical AI seems impossible. Power means people steal control.
Getting a nonpolitical AI seems impossible. Power means people steal control.
My reaction was: well, why bother then. If you are training something that will ignore the real data, you are going to get something that will spit back “nice” but wrong answers. Seemed like a big waste of money and talent to me.
They wanted to go from AI to AGI, but all they got was AG: Artificial Guess.