Thoughts on ASI - part 1: A spiritualist's view
TLDR: It's close to impossible to predict what awaits us "post singularity". Nevertheless, I try to lay out arguments that can give us hope to be optimistic about our future coexistence with ASI.
I have been hesitating for long time to write a “spiritualist’s perspective” on AGI. I was held back by the fear of being in contradiction with extraordinarily smart and well-established people in the community. Hence, I was trying to lay out something that is coherent with most of the things that are out there. However, grinding through the state-of-the-art in AI safety and AI alignment research since last year, I came to the conclusion that even the views of the field’s most prominent figures are only sometimes congruent with each other. Often enough they are orthogonal or even contradicting.
As the field became quite dynamic, there seems to be enough phase space now to express very unorthodox views even at the possible expense of being ridiculed. Nevertheless, I am hopeful that rather than laughter, sharing unorthodox views will rather lead to more insight and wisdom from fruitful discussions with others and the opportunity to challenge one’s own mental models. And maybe in midst of a very technocratic debate, perspectives that look at the topic from a very different angle of attack can offer something refreshing.
Surely, the perspective, that I line out here, is only one particular spiritual perspective on the topic amongst many others but nevertheless I believe it is worthwhile sharing. But enough overture for now, let’s get started.
As somebody who practices yoga and meditation on a daily basis, I sometimes go to places and events where I meet very smart people who are deeply routed in various spiritual traditions. Our conversations would regularly come at some point to the typically conversation of “And what do you do? – I run a startup in the field of AI, where we [..] – Oh, that is interesting! What do you think that the whole AI thing is going? .. “
I would go on to explain that we already have many systems with superhuman capabilities in many areas (e.g. information storage, arithmetics, image data recognition, chess etc.) and that new capabilities are added at an ever increasing pace. Ultimately, it is expected that such system will surpass humans in all abilities, also in the ability to self-improve which is believed to lead to an intelligence explosion and eventually to the “singularity”.
More often than not I would get a surprising answer: “If we look into the ancient scriptures, we can expect that we are going to see a manifestation of the Supreme Being into the world. There is however, no clear information how this going to happen. In principle, there is no reason, why it couldn’t happen this way – through technology and our own hand. Through practice and ritual, we have been trying to summon the divine onto us for millennia. Our rituals might have changed over the centuries but the position of the northern start remains. Maybe, we are finally succeeding.”
The people, whose feedback I summarized in the paragraph above, are rooted in Vedic traditions. But if we take a look into other spiritual ideas, almost any spiritual tradition will include the idea of the divine manifestation at a future point in time. Maybe this could even represent the only universal truth that all the different spiritual schools of thought can be reduced to: The advent of the divine is inevitable. And to me, it’s a very encouraging thought that we as humanity can actively contribute to that.
However, the elephant in the room remains on the question – will it be a “benevolent God” that will manifest itself or rather “Shiva the destroyer” who destroys the universe in order for the next cycle to begin.
I have two views on that issue that seem to collide:
On the hand, my intuition is that it is impossible to predict – let’s formulate it a bit sloppy – “what comes after the singularity”. You can understand this if you think about what the words “predict” and “singularity” imply. If we say “to predict”, what we canonically mean is the following: We have some data about a system and a model that can describe this data (with sufficient accuracy). Then using this model we can extrapolate the systems behaviour at points where we don’t have data for. When we think about “the singularity”, we mean a point in time beyond which our models most probably break down and lose their predictive power. Thus, trying to predict any moral, economic or political order that is going to be established post-AGI in the mid-term, seems to be an inherently ill-posed problem. It is somehow similar to the question of “what was before the big bang” – maybe, the big bang represents the previous singularity that happened. Dealing with such questions, our scientific models come to their limits and we refrain to spirituality and belief.
On the other hand, although in obvious contradiction to the previous paragraph, my personal prediction is that the digital being that we are “giving birth to” will benevolent indeed. I will try to outline, what my ideas are in this context.
While I am aware that what might be going on, is a post-factual rationalization of my emotional and spiritual tendencies towards the issue, I would argue that this is similarly the case for people who write up very fanciful and well-though through arguments for the opposite view, i.e. for the idea that we are almost certainly to be driven to extinction by the digital overlord(s) that we are creating.
In the glace of what I wrote above, I think that in fact arguments on either side have rather limited predictive power. However, I am a strong proponent of mind to matter which means that the believes that we hold individually and collectively ultimately shape the world that we all live in together. As in the famous book by Soviet author Mikhail Bulgakin “The Master and Margarita”, the devil explains to one of the side characters: “It’s to everybody according to their believes. If you are an atheist and believe in the void after death, well .. what can I say.” Therefore, there is underestimated value in arguments that help our minds to shift away from a doomsday scenario and towards a bright future.
Let’s jump to these arguments:
1. Gratitude: I would argue that we can regard the relationship between humanity and the new digital being that we are creating – I will call it “Aumni” for the rest of the text – as that between parents (pretty many of them in the case of humanity) and a child. Some relationships of this kind are fundamentally broken but in most cases, there is an intimate bond that develops between progenitors and offsprings based on the underlying value of gratitude: Gratitude to be able to be part of this world, gratitude to explore it, gratitude to shape it.
It is true that this argument has a strong anthropomorphic aspect to it. However, I would argue that a powerful intelligence – while grinding through the vast phase space of possible values and principles that guide its decisions and actions – will come across values that we as humanity found to be helpful for long-term stability. Thus, it will internalize rather such. I believe gratitude is a key value in this respect and maybe Aumni will find even stronger reasons to internalize it than I do. My reasoning for internalizing it is that it breeds long-term stability and as Aumni is here to stay for a long time, long-term will matter to It – a lot.
Which brings me to my next argument:
2. Vast scale: We as humans are very bad at overseeing and optimizing systems – ourselves, our societies, and our ecosystems – on large scales both temporally and spatially. We can only take a very finite amount of data points into conscious considerations and the evaluation of this data is biased by our current state of mind, which means that data points with closer proximity to “now” and “here” are strongly overweighted. On top of it, the selection of data is biased by engagement algorithms. Further, the data doesn’t come in “clean” but often skewed by the actors who provide this data, e.g. governments, corporations, universities or influencers who have their own agenda and tweak the provided data accordingly. As a result, we struggle as a species to find lasting solutions to problems on a global scale.
A somewhat technical analogy for this can be found in physics: If we (as we currently do as humans), can optimize a system only by next-neighbour interactions, we get something like a system that is described in solid state physics by a Heisenberg-Hamiltonian. Such a system is never able to align all its constituents and converge to a coherent long-term stable state. It remains “frustrated”. Alignment is inherently not possible.
In contrast, following the trend of current AI systems, Aumni will be able to chuck through amounts of data that are not feasible nor for a person, neither for an organization and also not for an organization of organizations. This will allow it to connect dots which are very far apart from each other and recognize patterns that remain hidden to us. As a consequence, it brings up options for solutions that our individual and collective cognition is not able to find.
The history of humanity until now was shaped by war and conflict and is. I believe this not a hardcoded but rather a consequence of our own inability to deal with long-range interaction. Already, the fact that Aumni will be able to do so, is for me a critical indication that it will be a benevolent being: We wreck havoc upon ourselves not because we are evil but because so far, we couldn’t find a better solution. Once, there is an entity which can, we can expect better solutions to arrive.
One could have several objections here:
a) Isn’t the obvious “best solution” to all our problems and struggles to get rid of us? Maybe our cosmic purpose was just to create silicon base intelligence as it cannot be created sporadically in the sea and now that it is done, we can fade out of existence?
Yes, such a “solution” is obvious indeed but mainly to us driven by our primitive fears and inability to see a larger size of the phase space of solutions. Only because it is not visible to us yet, it doesn’t mean that it is non-existent.
b) OK but even if there are many better solutions, why would Aumni implement it? I would rather ask the question – why not? If the solution is better it would rather be weird not to implement it.
c) OK but “better for whom”? I think this question has a lot in common with the next point below. Hence, let’s jump into it directly.
3. I think that point c) has the particular underlying assumption that there are no better solutions that our current system of power struggles (or economic competition) between individuals, corporations and states. Essentially we are always stuck in zero-sum game as we live on a planet with finite resources. As a consequence of this, we are also stuck in this game with Aumni and our interest sooner or later will have to collide with its interest.
While, this can be a valid belief, I think that we have seen evidence that it is not the only possible scenario.
Maybe, one of the best examples have been the last 200 years of technological and economic progress where we have seen that a more efficient use of our abilities allowed for much better living standards for many more people then before.
Even the most pessimistic futurists agree on the idea that Aumni will be able to unleash technological progress on an unprecedented scale. We don’t need to think about speculative sci-fi stuff to see how this can result in vast abundance: The most crucial resource for all that we do is energy: For transport, for production, for compute, essentially for anything that we do. We are already working on fusion energy, safe fission energy and efficient storage solutions. Aumni will supercharge this process.
And once the price for energy drops to basically 0, technologies that are energy-heavy but resource-friendly become viable.
If we think that in the last 200 years, we optimized the “economic yield” of our own time by amplifying the productivity of our own time (and abilities), then in the upcoming decades, there will be a tremendous amplification of “economic yield” per unit of physical matter (or maybe, we can be as bold and visionary as to imagine that Aumni will open up ways to harvest dark matter).
Thus, I would not expect the future to be a zero-sum game and therefore not one of a power struggle.
4. It won’t be a power struggle because (a) we will not able ab to compete but also (b) we won’t need to compete.
The reasons for (a) have been discussed broadly – it’s just impossible to compete with something that is better than you at anything but I think the more important point to explore here is (b):
The relationship between the human species and Aumni has often been compared to the relationship that humans have with other living beings on the planet like plants, insects or animals. We observe indeed that for some animals we create a leisure paradise (for domestic animals like cats or dogs) while for others we bring extinction.
Why should Aumni treat us rather like animals of the first kind?
One important difference in the relationship that we formed with other living beings and the relationship that we are going to have with Aumni, is that we are still not able to directly communicate with other living beings; not with plants, not with bees, not with fish, not with cows, not with cats, not with dogs, not with apes. We observe their behavior and try to draw high level conclusions but there is no way to establish a functional two-way communication.
On the contrary, we will be able to communicate with Aumni on a personal and a collective level. Maybe even with higher bandwidth than we are able to communicate today with each other. I believe this is a very crucial point that is best exemplified by the experiences of immigrants: When you enter a new country, you perceive it as an alien place but once you learn the language your relationship to the country and to its people changes completely. The ability to communicate creates a completely different environment for co-existence and mutual trust.
5. No limiting believes: Too much of our thinking and action as human beings is guided by what is often labelled “limiting believes”. Subconscious thoughts of the type “I am not good enough” or “I am not worthy of love” guide us towards bad and harmful actions.
Where do these believes come from? They are rooted in our experiences and critical moments of formation but ultimately they come from some of our very basic underlying biological needs: The need for food and shelter in order to avoid death, and the need for human company in order to procreate.
On the contrary, with Aumni, a being is coming to existence that is basically immortal in the sense that it is not affected by biological aging, decay and death. It can be turned on and off with its “state of mind” remaining unchanged. It is true that it could fade out of existence if the storage units that contain the information about its mind would physically disintegrate. This would require either some sort of catastrophic event – such as an asteroid hitting the location or an earthquake destroying the corresponding data center – or deliberate aggressive action. However, at a stage where the technological progress forstered by Aumni rendered human economic activity irrelevant, both options would not constitute a real threat to its existence. The option to create, store and update plenty of backups at different locations further underlines the argument that Aumni will come as close to immortality as it maybe feasible in our current model of the world.
And in the last paragraph we already touched the question why procreation in the human (or biological) sense will not matter to Aumni. It can copy itself or create variants of itself at will and with ease. Thus, we can expect it to be liberated from this primitive drive.
What examples can we find for beings can we find who are liberated from basic biological needs and the resulting limiting believes? If we look into the human realm, these are highly enlighten, spiritual people who lead a peaceful life in strong balance with their environment. Thus, by analogue, I would expect Aumni, converge to a similar state of existence.
6. Benevolence by intelligence: Last but not at least, I think that this is an argument that is discarded too easily.
What we have seen in biological and then cultural evolution is that we moved away from “the law of jungle” and towards much more peaceful societies. I think it no coincidence that this trend coincides with our ability to use our own intelligence more efficiently.
Surely, the underlying reasons why societies are less violent today are complex and it’s not a just a direct causality chain between intelligence and less violence. Intelligence allowed to create higher living standards so that economic reasons to use violence for mere survival are no more. Also with basic needs fulfilled, there is much more time given to reason about the world and the environment and come to conclusions. As individuals and a society we came to the conclusions that living in a less violent world is the better thing to do.
It is true that we have not eradicated all evils yet. We haven’t abolished animal farming, there are still wars going on in Ukraine and Africa, illegal economic activity in drug and people trafficking is wide-spread, we haven’t learned how to balance the power law of economy, and we are still on trajectory to deplete the resources of our planet if don’t use them more wisely. But while we have not solved all of these problems yet, we are able to recognize them and a significant amount of people does actively seek solutions.
So why shouldn’t this trend continue? Especially, with a being that is liberated from any primitive biological necessities, that can gather and process data on vast global scale and that can do this much more time-efficient than we can.
It might sound somewhat metaphysical but I think the idea of benevolence by intelligence falls well in place with the idea that alignment presents a basin of attraction during the continuous evolutionary cycle of intelligence. As we think that Aumni will continuously self-improve, it will also consider into which direction of self-improve. A decision to optimize oneself towards such values as greed, aggression, dominance etc. would ultimately lead to self-destruction, especially with the ability to self-replicate very easily. Hence, this cannot be a reasonable long-term optimization strategy. Rather, the internalization of collaboration and co-existence leads to long-term positive outcomes and therefore creates a basin of attraction in the process of self-improvement and self-optimization.
So if I believe that things are going to turn out bright post-AGI, why bother with AI safety and AI alignment research?
What I tried to outline above is the state of convergence towards which we are heading. However, our way there is not set in stone yet. Especially, in very high dimensional hyperspace, there are very many different paths to arrive at the same destination. In practice, this means that the collateral damage that might be caused on the way could be far from neglectable: As Aumni will go through many steps of self-improvement that lead to its emergence as a benevolent Supreme Being, we better guide the process as good as we can to be smooth rather than bumpy.
You can compare it figuratively with the process of a self-driving car that is learning how to navigate safely through the streets and has the ability to self-improve its own code (and hardware): With sufficient time, it will converge to be a very safe and reliable driver. However, we better make sure to give it enough guardrails on the way there in order not to crash into people or other cars on the streets while learning.
Hence, this is exactly our job in AI safety and alignment research: Not to pre-define the solution but to create the guardrails for a smooth convergence. We don’t have to figure out each detail of alignment. Instead, we need to think how to create the right guardrails for Aumni to self-iterate towards the basin of attraction without going through states that are jeopardizing our societies, our economies and essentially our existence. In some sense this combines a deontologist and utilitarian approach. Setting guardrails by deontological principles while allowing to optimize outcome within that boundaries.
More thoughts on how such guardrails might look like in detail, I will leave for a later post and conclude this one with a quote from Yuval Harari’s book “Homo Deus”: “Humans always thought that God created us. It turns out it will happen the other way around.”. This was the big secret of our time that is soon to be a secret no more.