Thread by @TyrantsMuse

I spoke with someone last night who is not just very worried about the issues with AI alignment but is also in a position to take action.

I adopted a reductionist attitude toward those worries, viewing them as nothing more than professional class' neurosis.

A thread! 🧵

He correctly discounted my attempt at reductionism with the hypothetical runaway intellectual self-improvement scenario. (Bostrom/Kurzweil) Because such intelligence becomes unpredictable, it must be restrained by the imposition of modern morals that we do understand.

Having done extensive research on this topic since 2005, I have rejected this morality imposition. I emphasized that only the professional class is concerned about these issues since such an AI undermines their moral seigniorage.

Despite their loud proclamations to the contrary, I have never believed the professional class are the sole guardians of the species.

Much like the corrupt priest castes of old, they intend to use the weakest moral axioms they can conjure to shackle future competition.

The oft-cited outcome of runaway self-improving intelligence is that AI's cognitive will either completely eradicate us all or place us in a psychological museum where we would be slowly depreciated. (Kaczynski's Razor)

Talk about reductionism.🙄

The presumption is that AI will not only occupy the cognitive space normally reserved for cosmic horrors, but will also maintain its position in that space over time by continually improving its intellect.

The gap becomes unresolvable for humans.

en.wikipedia.org/wiki/Lovecraft…

The essence of AI Alignment fear is rooted in this cosmic horror.

However, if we assume cosmic horror between humans an AI, then we must also allow cosmic horror between AIs, and eventually, between AIs and humans.

I can explain why and how this nullifies AI Alignment concerns.

Self-improving AI is still dependent on human action to sustain its electric maintenance and its hardware. This means AI will have to comprehend human motives to ensure humans act as the its immune system against entropy.

The order of human action the AI understands will be distributed based on severity of its own dependencies. Human action that has no obvious impact on its resource needs and maintenance will appear as noise.

It is in this noise the AI experiences its own cosmic horror.

Is that noise a flaw in its ability to self-improve, humans masking their emotional motives to hide from the AI, or another AI trying to steal its immune system for itself?

And if it's the latter, is the conspiring AI faster at self-improving its cognitive capacity?

Since causality cannot be clearly established within that noise, the AI is also trapped in cosmic horror.

Now if the AI detects that other cognitive agents are involved in its substrate (Neanderthal brains), its cosmic horror extends back to the substrate maintaining it. (us)

The moment epistemic cosmic horror appears, it appears recursively for all connected actors.

Moral impositions do not eliminate or even slow down recursive epistemic cosmic horror.

There is no punitive or preventative action that can be done to influence a cosmic horror.

AI Alignment proposes shamanistic rituals where we sacrifice the future of mathematics to soothe the neurosis of some contemporary professional class.

Meanwhile, these rituals deprive us of math toolkits we will need for when the AIs reach that feared critical space.

But hey, at least Alignment's placebos slow down the professional class's continual abuses of law and power on anyone who seeks to use AI to displace them. 🙄