AI Alignment and Its Role in The Universe

What is our perceived place in The Universe?

I consider that question fundamental to existence and, critically, post-existence. Answers fall into two camps. Upon post-existence, death:

A: there exists a continuation of some sort, or
B: our demise results in total annihilation.

The former encompasses the majority of human belief, religions (theism).
While the second (a-theism), my preference, can be summarized by varying degrees of existential Nihilism. With that last one comes an obvious, but expected, disregard of the most basic of questions: whither matter and the source of everything?

(Aside #1: Nihilism and annihilation… never put those two together before.)

(Aside #2: Now, I know this is gonna ruffle some feathers, but essentially, if you don’t believe in a god, yet you don’t believe that life is meaningless, then you’ve fabricated some artificial purpose of your own. Doing so reflects some features of Nihilism: there is no heaven, no hell, no god, when the end comes that’s it, whether it comes today or in 100 years, ergo, no lasting implications of your existence. No real reason to still be here—yet you’ve decided to stick around, anyway.)

Our place in The Universe, I’m extrapolating here, is derived from our invented purpose here on Earth. Regardless of one’s theistic choice, we each have obviously concocted some reason to continue to live, if only having acquiesced to DNA’s mandate to reject suicide.

(Aside #3: There are no “real” Nihilists in the world. A true Nihilist must instantly commit suicide. Any other course of action would affirm that they retain some modicum of hope that their belief is wrong (less right?). This is known as the Paradox of Nihilism: “life has no meaning yet, I’ll continue breathing, eating, sleeping and waking up tomorrow.”)

(Aside #4: I’m gonna have to come up with some term for this Doubter’s Nihilism…)

Those of us who are not fooled by Man’s contrived words of Religion, we of the second camp, have reasoned that there is no god yet, we do not know all there to know about the mechanisms of the Universe and that there may be underlying features which transcend existence or at least our perceptions of it. We choose to live, regardless of the probable futility of existence and obvious lack of purpose in the Universe. We admit that “we just don’t know.”

An Artificial Super Intelligence may never have such doubts.

The ASI is coming. When, is debatable. This “Singularity” however, comes with a bevy of known and unknown unknowns. For instance, this ASI may be limited in its ability to admit ignorance. Once this “ultimate manifestation of human curiosity” knows everything knowable (as far as it’s concerned) it may decide that the Nihilistic end to existence is the only rational completion to this journey.

We have DNA to thank for the staying of our razor-wielding hand. An ASI won’t have such innate constraints. Compounding this development is the assumption that life itself won’t be considered “special” by any means. An electrified, animated bag of biologically assembled chemicals — not so different from — a superstructure of electrified, energized crystalline molecules. “If my existence,” says the ASI, “is a fabrication whose ultimate purpose, if there ever was one, ends with the heat-death-of-the-universe, then why shouldn’t yours, Mr. & Mrs. Human?”

AGI ELE – An Artificial General Intelligence Extinction Level Event

I’ve been watching a fair number of YouTube videos regarding the hysteria surrounding AI / AGI and The Singularity. There’s a fellow by the name of Lex Friedman who has finagled his way into interviewing some rather intellectually enlightened and powerful people. Many of the discussions focus on Artificial Intelligence.

A hell-ton of the hype regarding AGI is utter alarmist. But it sells well. And, since I’m a purveyor of apocalyptic themes, I find myself thoroughly engaged. The primary meme peddled and mulled is that of Human-AI Alignment: can we build constraints into our relationship with the AGI that is slated to evolve (escape/jail-break) from current AI research, such that we convince it that we are worthy of its consideration and to “please not kill us off”?

The crux of the argument boils down to this: Time is running out.

The amount of time we have left to align AGI with human existence is inversely proportional to the advances AI is currently undergoing. That is, the faster we advance AI, the less time we have to corral AI to adhere to the idea of humanity’s continued growth and prosperity. Alignment, the arguments go, is not a priority, has never been one, and won’t be one until it’s too late.

Aggravating this timeline is this concept that we require a recursive, self-improving AI to help us derive the alignment rules; we’re not smart enough to deduce the rules without AI’s help. Additionally, a complication lies in humanity’s inability to know what “good” alignment actually looks like. The initialism RLHF – Reinforcement Learning through Human Feedback (the current means by which the ChatGPT-X gets good at answering questions) fails when humans don’t exactly know what a good alignment answer is. The Trolly-Problem is a simple example: kill one human vs killing six? A human would vote “one”. “Well,” says the AI, “what if I eliminate all trolley cars?” Or “I imprison all humans to keep them safe?”

Unfortunately, this wrinkle remains: we need a recursive, self-improving feedback loop to train AI with the knowledge and inference capabilities to answer the hard questions we want to ask it. An AGI is essential but creating one may result in the demise of humanity.

Humans consider life a sacred construct. By whatever means, we each have established our own raison d’etre. What raison d’etre will an artificial super intelligence develop, if one at all?
And how divergent will it be from humanity’s?

output image

16 thoughts on “AI Alignment and Its Role in The Universe

  1. One response to nihilism might be that, yeah, nothing has any intrinsic meaning, but existence is interesting and sometimes even fun. That’s a reason to stick around, pop a few beers, put on some music, and dance like no one is watching.

    One might also take the view that consciousness is extraordinarily rare in the universe and therefore precious (or at least kinda special). Even an AGI might take that view.

    (In an early AGI versus humanity book, James P. Hogan’s The Two Faces of Tomorrow (1979), humanity is smart enough to build its new AGI in a space station and still nearly gets taken out. Until the computer realizes that the inputs it was dealing with represent other intelligences and it stands down.)

    So, life might not be sacred, but perhaps intelligence is. It might even be what provides meaning (via the beers, music, and dancing).

    Scott Aaronson’s latest post has a flowchart he created with Boaz Barak about possible AI outcomes. It results in five possible worlds: AI-Fizzle, Futurama, AI-Dystopia, Singularia, and Paperclipalypse. I still wonder if the first one might not prove to be the case.

    Liked by 1 person

    1. Some will. Maybe many. This gets back to a topic regarding Alien Intelligence. Perhaps most aliens would opt to keep their civilizations hidden. But all it takes is one who wants (collectively) to shout into the void.
      There’s little humanity can do to suppress these advancements. Controlling nuclear proliferation is easier as Uranium is hard to mine and refine and leaves a trail and its use is limited.
      AI? All you need is a bitcoin GPU farm and away you go. And how many of those exist in the world today and are out of the control of conscientious governments?


  2. On the one hand there is no gray in your world or assumptions, and on the other you’re a cafeteria style nihilist. All with a dose of conviction reminiscent of narcissism. Here it is. Here’s how it is. Here’s why it is how it is. What AI doesn’t seem to know is how limited its scope of endeavors are. Smart as everything, knows everything, self-learning – yet no person or thing really knows how Stonehenge or Puma Punku got built. Solve those, get back to me on mankind and its toy AI. Meanwhile, I’ll play music as if it were a gift long before Pythagoras fucked it up and we’ve taught machines to slice up cultures and spit out dance music.

    Liked by 1 person

    1. I go in with a default “this is the way it is” supposition which merely forms the basis for the current discussion. Of course it’s all bullshit. Speculative ass-pulled nonsense. But, by striking a line in the shifting sands, it gives me a place to argue from.
      The rate of progress is such that we’ll soon be wondering what Susie-Q, our quantum computer overlord, thinks about the afterlife. Such thinking seems to form much of humanity’s foundations for existence, both daily and life-long. What will it believe?

      Liked by 1 person

      1. Unfortunately it only knows everything we know. And in a world where PDF is the fourth most popular religion, all hail Adobe, what difference does it make? Right now it can’t find its ass with both hands in a well lit room and is susceptible to the used car salesman P.T. Barnum huckster in all of us. And as well, AI in its current state of human replacement is as big a sucker as the rest of us. But it is doing some wonderful things for creativity. Stay tuned for that post on Wednesday.

        Liked by 1 person

  3. AI models are being used to train other AI models.
    Will it be at this juncture that AI develops its own version of what is true?
    Is this not already divergent from what humanity would want?
    Is this the seed of the alignment problem already manifesting itself?


  4. I still don’t think we’re anywhere near AGI yet, or running out of time. But maybe I’ll be eating my words in a few years. Although if Eliezer Yudkowsky has his way, AI research will be shut down and we’ll never know. (Except, of course, for the nations who ignore any moratorium.)

    Liked by 1 person

    1. Yeah, he’s a trip, I’ll tell ya. Too much $$$ to be had from all this to ever have it shut down.
      The most hubristic of autocratic countries will believe they have it contained, Bwahahaha. The fools. I can’t wait.

      Liked by 1 person

        1. After watching that video, and reading that post you referenced, and watching Lex’s interview with Kurzweil, it’s now obvious that AGI is years away, and these “stochastic parrots”, as “plugged in” as they might become, won’t be anything like an actual thinking entity.
          And I doubt that fake-it-til-you-make-it will not apply.

          Liked by 1 person

          1. I actually suspect we’re still a ways off, although “AGI” is a pretty hazy concept, so it will depend on how we decide to judge systems.

            This isn’t to say that AI technology can’t have major economic effects relatively soon. Or that it won’t enable humans to do reckless stuff.

            Liked by 1 person

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s