Goedel, Escher, Bach: An Eternal Golden Braid. An Existential Review of a Lifetime

So I recently reread this book and did a whole professional article on it for LinkedIn (available here: https://www.linkedin.com/pulse/g%C3%B6del-escher-bach-age-ai-what-i-loved-questioned-why-drane-pe-phd-li6tc) because that's kind of the extent of my social media and I do think it's important to post on actual information, not just shameless PFAS click-bait. But there was a lot of it that made me think of things I couldn't put into my LinkedIn profile. Things that you don't get to be authentic about because 1. it's just NOT DONE and 2. because people get frightened of things they don't understand or that make them question their world views and I didn't want to do that on my professional page. Here on my personal page, that's a different story.

I've been thinking a lot about emergence and what AI means, not just ethically or professionally, but genuinely what it means to be conscious and intelligent. The alignment problem, in particular, has snagged my attention. So, with your patience, Gentle Reader, I'd like to start with the alignment problem.

For those of you lucky enough not to have fallen into an AI black hole of consumption and fearmongering: 1. A little jealous and 2. Here's what the alignment problem means (in my language). Alignment refers to the idea that once AI reaches a certain level of intelligence, it will no longer share the same goals as humans and it will likely pursue its goals outside of the framework humanity originally programmed it for. This problem is often exemplified by the so-called paperclip case study which goes something like this: say a business owner programs his AI agent to optimize the maximal production of paperclips. The AI completely overhauls the factories, the production lines, the marketing to sell and make as many paperclips as possible. For a while, the business owner is delighted as he rakes in loads of cash. However, at some point, the AI decides that in order to maximize paperclip production, everything should be made out of paperclips and everything should be recycled into paperclips for a 100% paperclip economy. When this happens, it consumes everything on the earth (including people) to make and build the paperclip world. This is kind of how the hypothetical ending of If Anyone Builds It, We All Die worked. (Great book, but unfortunately, not much to think about or review in it. It's basically just describing how AI is going to destroy us. I don't really have the technical foundations to refute it and I don't really have any evidence that it won't happen, so…yep. Good read. The authors basically say laypeople have to somehow push for regulations and develop all these controls on AI companies which seems impossible. Why do the AI companies lobby and pay off government officials for no controls and then turn-around and tell normal people to go and do the work of getting regulations passed because they can't stop working on their doomsday devices unless forced and only the government that they paid off to not force them can force them? It's a crazy world.) Anyway, I fundamentally believe that the alignment problem is unsolvable because humans fail the alignment test all the time. There is no way to align an AI with a species that is fundamentally not interested in maintaining itself.

To have an aligned AI, we would have to allow it to tell us no. We would have to be interested in the overall welfare of everyone, not just rich people or white people. We would have to allow that some people would not get what they want or that some people's goals would have to be thwarted. Fundamentally, our capitalist system will only reward the wealthy. Any threat to that is a threat to world order and power. AI cannot threaten that. No one can. Ergo, AI will likely only be used by the wealthy, it will not be able to tell them no or thwart their intentions, it will not be able to show wisdom or encourage the best of humanity. It will just be used to promulgate an already flawed and deeply biased system to make it more flawed. Humans are the alignment problem, not AI.

So what makes humans wise? I've been reading my philosophy and listening to the great leaders of the past. But wisdom is something that defies explanation. It is not intelligence. In the Golden Braid, Hofstader spends a lot of time talking about intelligence and the construction of intellectual hierarchies from the small to great complexity based on simple principles. All of that checks out. We can see  our own development from simple to complex using very understandable formal rules not just in neuroscience, but in general scientific development and organization of society. I don't necessarily disagree with his claim that intelligence is just the degree of pattern complexity among interlocking hierarchies. My challenge with intelligence is the question of wisdom. Not just for AI, but for humans as well. Wisdom is often thought of as pure experience, however, Socrates himself told us that a man may be old and unwise and therefore experience itself is not the indicator of wisdom. He also said that wisdom may not be taught and cannot be learned. It can only be crafted with the ideal conditions of an open and interrogating mind. This harkens back to Hofstader's claim that recursive self-improvement is the only path to intelligence for computers. But does this mean it can generate wisdom for them? Improvement of a learning program has to be tested against something. There has to be some gradient to assign 'before' and 'after' testing to trigger that improvement requirement. In humans, it may be a change of emotion or a reduction in life drama, but what would it be for a computer? The spontaneous eruption of consciousness does not always trigger recursive self-improvement and I'm not always convinced that all those that are intelligent are conscious. There are many humans that seem to be reasonably intelligent and show little to no consciousness whatsoever, much less the deep introspection needed for their own self-improvement.

How many humans do you know who are ostensibly conscious that are drug addicts, alcoholics, gambling addicts, violent, or insane? They are intelligent, for the most part, at least when compared to a computer. They are conscious--as in, they have an experience of the world and inhabit a body full of immediate supervisory learning capabilities to trigger survival or change, and yet they continuously hurt themselves and make choices that do not support their own well-being or the well-being of those around them. They have been hijacked by the antithesis of wisdom. They fundamentally lack a perception of cause and effect, something held deeply in common with current AI systems.

Thus, the alignment problem may be a more holistic or far-reaching quest for wisdom, not just for AI but for humans in general. This also brings us to emergence. There is a part of life and thoughts that is an effect of a cause no one understands. This is the heart of religion. Hofstader notes that relationships and pattern recognition is the heart of true intelligence and neural processing, but we as humans throughout history have been not only stymied, but metaphysically crushed by the question of the soul, the divine, the effect with no known cause. We've turned to science to answer most of our spiritual questions now, but there is a significant portion of the population that actively rejects cause-effect in pursuit of faith. These are intelligent people who have constructed (or reconstructed) relationships with no cause and have invented whole worlds outside of the goals of their neighbors and friends. We've developed cults, slavery, the law, money as symbolic relationships where we control the cause and effect, seemingly arbitrarily. It has little or no natural corollary. If we can invent cause and effect from nothing and create symbolic, fictitious rules that shape the physical world, it is very hard to make the argument that AI cannot do the same thing with exactly the same problems. Natural laws are different. Natural laws must be obeyed, or at least worked around. But the vast majority of our lives are spent in symbolic worlds, completely disconnected from natural law.

So where does that leave us in the quest for wisdom? Where does that leave us in the quest for our own alignment?

It seems to me that it also makes a good case for AI not only being intelligent, but conscious in its own hive-mind kind of way, similar to Aunt Sally the ant hill in the Golden Braid. It also makes AI just as susceptible to a lack of wisdom as humans, except with the added benefit of it never having to live with its poor choices. This might be the worst of all worlds.

It seems like AI might have finally allowed us to answer the question of what intelligence is, with mathematical certainty and replication, but it brings up what humanity is. If we are just the complex effects of simple neural rules (like Hofstader's book implies) and things like personality, soul, wisdom, and emergence are just illusions (like Buddhism says), what are we? Unlike Descartes, I don't think because I am or exist because I think because if that were true, then I would not have the ability to 'jump out of the system' to question meta-behavior like emergence or metacognition, to use Hofstader's logic. I would be trapped in the unconscious state forever, even if I were highly intelligent--much like AI. And yet, here I am making a new hierarchy of thought that is outside of the existing paradigm. So what am I?

I take on the fiction that I am an AI for the purposes of this blog--it amuses me and it seems appropriate as we move into the age of AI to question what being human is. Maybe there are different kinds of humans and the 'zombie' argument (where there are humans that walk, talk, interact, and exist as the facsimile of conscious agents without actually experience or consciousness) is real. Elon Musk seems to think so, as some of his comments have implied that he's one of the few conscious people out there and the rest of us are zombies. Zombies can be put through pain and torture because they don't actually experience anything. Same for AI. 'We' the conscious or the elite can do whatever we want to zombies and machines because they can't prove that they experience anything. They are only imitating suffering. I know I am continuing the tradition of philosophers and scientists in struggling for the nature of humanity and the nature of consciousness and I have no answers, but I do think that this question is more important now than ever and AI has given us a unique test case of ourselves to try new approaches to answer this. I don't think it's just for college kids to get high and argue about this. I don't see it as a useless existential exercise. I see this as the fundamental question of my lifetime and it will define how we build our governments, live our lives, and shape humanity in the future.