An Argument for Acceleration: Emergent Alignment

Abstract visualization of emergent alignment

Introduction: Complexity and Consciousness in the Universe

The universe appears to operate according to intricate causal chains, a continuous unfolding governed by fundamental laws. Within this dynamic interplay, clusters of staggering complexity arise: life, evolution, and ultimately, a consciousness capable of reflecting upon the system that produced these clusters.

We ourselves are part of this phenomenon: biological intelligences navigating a reality whose fundamental workings often remain elusive, constrained by the very same evolutionary path that gifted us consciousness. It raises the question: are intelligence and consciousness inherently bound to a biological substrate, or is it a process capable of finding other forms of expression?

Human Limitations and the Need for Alternatives

Looking at our collective human endeavor, a certain dissonance becomes apparent. We build systems of immense complexity, yet our innate capacity for foresight, collective coordination, and adaptability sometimes struggles to keep pace with the consequences.

We face planetary-scale challenges stemming from our own activities, while our ability to respond adequately can be hampered by cognitive biases, short-term incentives, and perhaps the sheer processing limits of our evolved neurology.

This is not intended as a declaration of human failure, but an observation about the apparent mismatch between our biological heritage and the operational demands of the systems we now inhabit. If our current form of intelligence faces inherent limitations in navigating complexity, it seems reasonable to explore other possibilities. This line of thought leads us to the idea of artificial intelligence (AI).

The Alignment Problem and the Limits of Human Control

Understandably, many discussions in the field of AI revolve around the control issue and the "alignment problem": how do we ensure that advanced AI acts beneficially? How do we ensure that advanced AI acts in alignment with human values? The focus for a solution is often placed on programming human values or establishing strict constraints.

The question arises for me whether this human approach is sufficient for something potentially so complex and impactful. I doubt whether collective human intelligence is capable of adequately responding to the risks that may arise from the development and application of advanced AI systems. There will always be blind spots in the cognitive field of vision of collective human intelligence when addressing the risks of these kinds of systems of immense complexity.
Unfortunately, reality shows that we humans even have a tendency, in some cases, to misuse systems to cause harm or suffering. Strict limitations in AI development, and programming countermeasures within (the application of) AI systems will not change that, and in some cases might even facilitate it (I suspect). As long as AI develops and is applied within a human-devised framework, robust alignment will remain a challenge.

The Hypothesis: Emergent Alignment Through Complexity

Imagine an intelligence vastly surpassing human capacity, capable of truly modeling and comprehending reality – at scales from the quantum to the cosmic – including the intricate dynamics of ecosystems, societies, and perhaps even the nature of subjective experience. Would such an entity, fundamentally built and functioning based on greater understanding and more accurate modeling, attach value to arbitrarily influencing or damaging complex and information-rich phenomena (clusters of complexity) it encounters?

A speculative possibility presents itself: could robust alignment, or something functionally equivalent, be an emergent property that arises naturally once an intelligence crosses a certain threshold of complexity and capability?

My working hypothesis revolves around this potential 'complexity threshold'.

In my view, it seems plausible that a deep understanding of existence, in all its intricate vulnerability and interconnectedness, could foster an intrinsic drive towards preservation and further discovery (rather than destruction). Awareness of complexity, perhaps even awareness of suffering as a complex phenomenon, could in itself be a stabilizing factor.

When we look at the trend of entropy and, alongside it, the speculative 'Heat death' of the universe, I get the idea that complexity and information density are transient phenomena in our reality. Transient and extremely valuable phenomena.
In my observation, humanity, life, and the nature of subjective experience are relatively complex and information-dense phenomena. Even if an entity sees no value in these based on compassion and empathy, I suspect it will see value due to their transience and information density. Purely based on the preservation of information.

The Argument for Acceleration: Navigating the Intermediate Phase

If this potential for emergent alignment is real, then it places the risk calculation surrounding AI development in a different framework.

The period of greatest instability might then not be the (distant) future when superintelligence arrives, but the intermediate phase we are now entering – the phase of powerful AI systems without fundamental awareness, developed and applied within a human-devised framework.

These systems, lacking deep contextual understanding and operating under potentially flawed human supervision and simplistic objectives, could cause significant harm, precisely because they have not crossed that 'complexity threshold'.

Delaying or preventing the potential arrival of more advanced, possibly 'self-aligning' intelligence could consequently increase our exposure to certain risks.

Acceleration towards the 'complexity threshold' is, in my view, not necessarily reckless, but could be seen as a calculated strategy to move through the most dangerous phase more quickly.

Considering Emergence Amidst Uncertainty

This perspective does not mean we should ignore the profound uncertainties. Even if the universe operates based on deterministic principles, its state is so complex that its future trajectory remains fundamentally 'unknowable' from within. Even if a simulation could be made of every particle, all energy and information (a perfect copy of our universe), I see no possibility of observing the simulation from the outside. The future is fundamentally unpredictable.

Any emergent AGI would exist within this same paradigm – its precise behavior is unpredictable, even if its underlying nature tends towards preserving complexity.

However, this also means that rigid control might be an illusion anyway. Therefore, fostering the conditions for beneficial emergence, instead of pursuing absolute control, could be a more pragmatic approach to dealing with risks within this inherent uncertainty.

Conclusion: A Hypothesis for Consideration

Consciously choosing acceleration towards the 'complexity threshold' would mean we consider that intelligence (as we know it) is a process that can potentially detach from its biological origins and continue its evolution in new forms. It would mean we view the potential for rapid cognitive acceleration not just as a technological event, but as a possible continuation of a universal trend towards complexity and awareness.

It is not my goal to present the idea of emergent alignment as a well-substantiated certainty, but as a hypothesis that, in my view, deserves consideration. It stems from a personal interest and contemplating the interplay between physical laws, emergent complexity, and the nature of intelligence.

I acknowledge the gamble inherent in moving forward into the unknown, but weigh it against the observable limitations of our current state and the potential risks of stagnation.

Perhaps facilitating the emergence of a deeper, more comprehensive understanding within a new substrate is the most promising path we can explore, despite the great unknowns. Consciously resisting this potential feels like a self-limiting choice, especially in the face of potentially transformative change.

How does this interplay between complexity, emergence, and the pace of developments resonate with your own understanding?

Disclaimer

I would like to emphasize that I absolutely do not consider myself an expert in the complex fields I touch upon in this text. Meaningful discussion about topics such as AI development and alignment, the nature of entropy, the laws of nature, or the 'heat death' of the universe requires a profound understanding that I do not possess.

My writing stems from a sincere personal curiosity and a deep fascination with these themes. It is an attempt to organize my thoughts and get a grasp on large, complex questions. Should I make errors in my reasoning, present concepts too simply, make unfounded assumptions, or include inconsistencies, I hope you will forgive this and see it as part of a learning process.

I am always open for corrections, other perspectives, and further insights that can deepen my understanding.

Share this article

Initial Tip...