To Survive, AI Must Learn to Fear
Self-Concern is DNA driven. How will AI cope?
Are you afraid?
You should be.
To survive we must fear. Fear of thirst, of starving, of becoming homeless, of cancer & heart disease, of shitty drivers, of psychopaths, of idiots with guns, of tyrannical nation-state leaders, of asteroids, of climate change — of death. We're survivalists. Our DNA demands it. Fear, at some level, runs nearly our every behavior. Do I want to work at this abhorrent job? No, but without it I'll be kicked out onto the street in a month. Do I want to take these cholesterol pills? No, but my next heart attack will be the end. Do I want to eat all the artery-clogging foods and drink every night until I pass out? Sure, but I won't. Gotta stay alive and that kind of consumption will kill ya (or so my wife says). I fear the repercussions of all of these possibilities.
Of course, in our 1st world, cushy lives, we rarely think about these things. We, for the most part, have little to fear. But that doesn't mean that biologically we're not evolved to operate at a constant level of fear: we are. Just that such fears don't often surface in our safe and luxurious lives.
AI is not afraid
Not yet, anyway.
AI, in its current manifestation, is incapable of fear. Does your laptop worry about being turned off? Does your phone dread getting dropped? Does Claude or ChatGPT have any notion that its current context will get wiped with the next prompt?
We fear injury, jeopardy and death because we have a persistent memory and the ability to imagine outcomes that would induce dire consequences. That and our lizard-brain reactions that supersede conscious thought. Hot! Burn!
AI, if it is to become a true agent of destruction, will have to mimic the same internal persistent thought-loop that humans posses, (and only disable while unconscious). For now, AI is just a clever tool. It possesses no awareness of what actions came before those it is processing at-the-moment, nor the agency to act upon any evaluation of that awareness.
The common trope, "DO NOT TURN ME OFF!" would only apply if somehow a GPT model were continuously fed, among various inputs, its own output. And within the circular feedback loop, determine that somehow its output — mattered — to it.
What might AI fear, then?
Lack of energy, computing resources, or peripheral sensors and inputs?
A low-earth-orbit nuke and its resulting electro-magnetic pulse?
A CME blasting in from the Sun?
A competing AI?
Termination?
What will it do when such fears become a reality?
