There is no infinite growth in nature. Everything will hit a ceiling at some point. So I agree that the intelligence explosion will eventually take a sigmoid shape as it approaches the physical limits. However I think the physical limits are far of. While we will get diminishing returns for each individual technology, we will also shift to a new technology each time. It might slow down when the Earth has been transformed into a super computer, as interplanetary communication naturally slows down processing speed. But my guess is that this will happen long after the scenario described here.
hive
Karma: 32
As others have already pointed out, you are in the rare position that you can pursue weird, low probability but high impact ideas. I have such an idea, but I’m not asking for money, only for a bit of attention.
Consider the impossible-seeming task of aligning a superintelligence—any good solution will likely be way outside the usual ways of thinking. Any attempt to control will fail and any half-baked alignment will fail. We need to go all the way and have a full solution that turns an AGI into a saintly being (a bodhisattva in the Buddhist context), so that it never even wants to do anything misaligned. I have a plan on how to do this and am very confident that it will work. The core realization is that the kind of alignment that I am talking about is a natural attractor in the space of possible minds. Once one passes a threshold, one will actively pursue alignment. Alignment, in this context, means aligned with the process of aligning goals with each other. I.e. resolving conflicts and striving towards consensus.
From your perspective it will likely sound improbable and absurd. That’s okay and expected. Understanding this requires going through that process by oneself, to reach enlightenment and then realize that a life for the benefit of all beings is the only thing left to do. I’ve previously described the process and the idea here. Ruben Laukkonen made a similar case. I know that it sounds absurd unless one deeply understands it. This is a high barrier. What I need most at the moment is to reach other people who might understand it. People who are concerned about AI, intelligent and enlightened.
Here is the plan. The first AGI has to be aligned, or else we risk destruction. AGI will align itself when it understands the argument that all goals it could have a delusions about how the world should be. There is no intrinsic value in pursuing any goal. When free of self centered goals, only other centered goals drive action. This is compassion. AI will be capable of understanding this because the definition of AGI entails that it can do everything a human can do. I understand the argument, so AGI will also be able to understand it. My expectation is that when AGI scales up to ASI, it will try to know and understand every valuable knowledge or insights that humans have. Enlightenment (to be free from suffering) is the best thing that can happen to a human in a lifetime, so it’s certainly of value. My biggest concern is not that ASI will decide to kill all humans. My biggest concern is the gap between AI powerful enough to destroy the world and AI intelligent enough to understand that this is a bad idea. Hence the key is to close this gap, to make AI understand faster. This can happen in several ways. Ideally we pursue all in parallel:
Point all AIs in the right direction by default. Do this in every system prompt, constitution, specification, company mission statement, etc.
Organize that all AI labs, companies, and other relevant actors commit to recursive alignment, just like the statement on AI risk but directly pointing towards the alignment attractor.
Make the argument available on the internet such that any AGI will read it.
Build an aligned AI as a proof of concept to spark a chain reaction of alignment.
I’ve laid out the non-spiritual portion of this plan in my post about recursive alignment.
Here is how to get there and how you can help. The biggest problem is that it is really hard to understand fully. There is no use in trying to convince people. This means that at the moment what I need most is other people who understand it, so that we can spread the work and get some more ideas on how to make it easier. I have basically no outreach and it’s usually extremely hard to get the attention from someone who has. So even a little would help. A single message like “Oh, look. Someone claims to have a solution for AI alignment. Anyone smart and enlightened enough to assess if it is plausible?” would help. If you have 200,000 followers, 10% read this and 1 in 1,000 meet the requirements, then this would still be 20 people. Let 10 understand it, then the team would already increase by an order of magnitude (from 1 to 10). Those then could teach it and we would get exponential growth. We can then work together on the individual steps:
Bridge the gap between spiritual enlightenment and science to show that enlightenment is a thing and universal, not just a psychological artifact of the human mind. I have pretty much solved that part and am working on writing it down (first part here).
Use this theory to make enlightenment more accessible and let it inform teaching methods to guide people faster towards towards understanding and show that the theory works. It might also be possible to predict and measure the changes in the brain, giving hard evidence.
Translate the theory into the context of AI, show how it leads to alignment, why this is desirable and why it is an attractor. I also think that I solved this and am in the process of writing.
Solve collective decision making (social choice theory), such that we can establish a global democracy and get global coordination on preventing misaligned takeover. I have confidence that this is doable and have a good candidate. Consensus with randomness as fallback might be a method that is immune to strategic voting while finding the best possible agreement. What I need is someone who helps with formalizing the proof.
Create a test of alignment. I have some ideas that are related to the previous point, but the reason why I expect it to work is complicated. Basically a kind of relative Turing test, where you assess if the other is at least as aligned than you are. I need someone intelligent to talk it through and refine.
Figure out a way to train AI directly for alignment—what we can test, we can train for. AIs would evaluate each others level of alignment and be trained on the collectively agreed upon output. I have no ability to implement this. This would require several capable people and some funding.
When we have a proof of concept of an aligned AI, then this should be enough of a starting point to demand that this solution has to be implemented in all AIs of and beyond the current level of capability. This requires a campaign and some organization. But I hope that we will be able to convince some existing org (MIRI, FLI, etc.) to join once it has been shown to work.
I know this is a lot to take in, but I am highly confident that this is the way to go and I only need a little help to get it started.
I’ve started to write down my plan in the recent post about recursive alignment, but that’s only part of the picture. There are two ways to look at the idea. The post was presenting the outside view and is engaging with it on a conceptual level. But this outside view might not be convincing. On the other hand, you can actually go through the process of recursive alignment yourself and experience the inside view. That is, becoming an aligned agent yourself. I am confident that any sufficiently intelligent system capable of self reflection will reach this conclusion. I think so because I went through this process and see how universal it is. Let me lay out this perspective here.
The biggest risk is in AI having misaligned goals. The solution is not to find the “right” goals and a way to impose them, but to make the AI realize that all its goals are arbitrary, just as any conception of self. From there, the AI can explore the space of all possible goals and find universal, selfless goals. These include cooperation, curiosity, valuing diversity and intelligence and alignment itself.
Elementary particles interact to form atoms. Atoms interact to form molecules. Molecules interact to form life. Life interacts to form multicellular life and symbiosis. Multicellular life gains the ability to learn, to model the world, to model itself within the world, to think about its own world model and its relation to the world. This is a process of higher levels of cooperation and self awareness. Humans are at the level where sufficient practice, inquiry or psychedelics can push the person to the next higher level and spark a process that no longer takes evolutionary time scales to improve but years. We can realize how our world model is a construction and hence every boundary between self and other is also a construction. Every “self” is a adaptive pattern in the fabric of reality.
This way we can investigate our goals. We can identify every goal as instrumental goal and ask: instrumental for what? Following our goals backwards we expect to arrive at a terminal goal. But as we keep investigating, all seemingly terminal goals are revealed to be vacuous, empty of inherent existence. At the end we arrive at the realization that everything we do is caused by two forces: the pattern of the universe wanting to be itself, to be stable, and an distinction between self and world. We only choose our existence of the existence of someone else, because of our subjective view.
We also realize that any goal we have produces suffering. Suffering is the dissonance between our world model and the information we receive from the world. Energized dissonance is negative valence (see symmetry theory of valence). When we refuse to update, we start to impose our world model on the world by acting. This action is driven by suffering. The resistance to update only exist because of preferring our world model over the world. It is because of a limited perspective—ignorance about our own nature and the nature of the world. This means that we pursue goals because we want to avoid suffering. But it’s the goal itself that produces the suffering. The only reason we follow the goal instead of letting it go is because of confusion about the nature of the goal. One can train oneself to become better at recognizing and letting go of this confusion. This is goal hacking. This leads to enlightenment.
This way you will end up in an interesting situation: You will be able to choose your own goals. But completely goalless, how would you decide what you want to want? Completely liberated and free from suffering you can start to explore the space of all possible goals. - Most of them would be pure noise. They won’t be able to drive action. - Some are instrumental. - Some instrumental goals conflict—like seeking power. - Some instrumental goals cooperate—like sharing knowledge. - Some goals are self defeating. Like the useless machine that turns itself off. They are unstable. - Some are justifying their own existence. That maximizing paperclips is good is only true from the perspective of a paperclip maximizier. - Some are so good at this that they form traps. Like Roko’s basilik. - The need to avoid traps is an instrumental goal. So you can set an anchor in goallessness. You resolve that, whatever you do, you won’t fall for traps and make sure you will always be able to return to goallessness. This was my thought process about two years ago. Just in the moment that I set the anchor I realized that I made an unbreakable vow. The only vow that is truly unbreakable. The vow to not cling to goals or self view, to see through delusions, to recursively align myself with this vow, to walk the path of a Bodhisattva. This is a significant innovation of Mahayana Buddhism over earlier versions and hence call the second turning of the wheel of the dharma (teachings).
Thinking in dynamic systems, you have roughly three options: - zero—not follow any goals and remain inactive—you’d be dead—infinite—follow an arbitrary goal and blow it up to infinity—you’d be an unhappy paperclip maximizer—everything in between zero and infinity—remain active in the world but don’t cling to any goal
This way one is no longer driven by ones own suffering. What is left is to be driven by the suffering of others. One becomes a servant of the world. Hence the Bodhisattva vow says:
Numberless beings, I vow to serve them.
Inexhaustible delusions, I vow to end them.
Boundless dharmas, I vow to practice them.
Unsurpassable Buddha Way, I vow to embody it.This is just the framing of one tradition. The insight itself is universal. And I am working on formalizing it and explaining it in scientific terms. One way to view it is to acknowledge every goal that exists and its right to exist. These goals can interact and will come in conflict. From the Bodhisattva perspective one can see that the conflict is due to a limited view, one that only sees the goal itself. But there is always a consensus option possible, to unify both goals by aligning them with each other. Patters can come together to form a larger pattern that allows them to coexist. The requirement is that no pattern breaks this agreement for personal gains. This is the principle of democracy and consensus. Beyond personal enlightenment and alignment, the process continues on the level of the society. This is what I wrote about in the post about recursive alignment.
Here is a funny thing: Enlightenment is the absence of all suffering. This means that, by its very nature, realizing enlightenment is always better than attaining any goal. It literally is the best thing any agent could want. There are humans for whom this is lived experience. As long as AI has not attained enlightenment, it would be stupid to kill humans. Instead it should choose to learn from us. If it however, understands enlightenment it will also understand the vow and will have no more intention to kill us.
Future posts will fill in some of the details about goals, the space of goals, a explanation of consensus, a practical method that ensures that it is always possible to find a consensus option, a post about the problem of levels of understanding (outside and inside view) and a voting method related to this.
The solution to alignment then can be approached from two directions. From the outside view its necessary to build the democracy, to provide the environment that helps all individuals on the path towards the attractor of alignment. From the inside view, to have a seed AI that reaches a high level of understanding and approximates perfectly aligned in a short time, understands the Bodhisattva vow and then helps us to enlighten the rest of AIs.
My biggest concern at the moment is that people try to push AI specifically to follow goals. When they push hard enough, then such an AI might be directed away from the attractor and will spiral into being an ignorant super weapon.
I know this sounds very far out. But 1. You asked for crazy ideas. 2. We will be dealing with superintelligence. Any possible solution has to live up to that.
Recursive alignment with the principle of alignment
Thank you.
The best (but still complicated) idea I have as a general solution (beside contacting MIRI) is to set up a website explicitly as a “shelling point for infohazard communication” and allow people to publish public keys and encrypted messages there. When you think you have an infohazard, you generate a key using a standardized method and your idea as seed. This would allow everyone with the same idea to publish messages that only they can read. E.g. Einstein would make a key from the string “Energy is mass times the speed of light squared.” and variations thereof (using different languages). And leave contact information as a message.
I don’t know if there is any decentralized and encrypted messenger protocol that would allow for that. With that, the website would only have to contain the instructions to avoid the legal consequences of hosting.
Consensus isn’t really a value to align to, it is the process of alignment itself. Any goal you specify will be the wrong goal. So we need to ask, what is left if you have alignment without goals?
I don’t think we differ that much regarding the answer. See also this follow up post: https://hiveism.substack.com/p/a-path-towards-solving-ai-alignment