Polymathic Artificial Superintelligence: A Hope

The first entry of the expression "polymathic artificial superintelligence" that I know of was in my Master's thesis in 2015. Since then, I have always wanted to articulate a polymathic ethical framework for artificial superintelligence (ASI), but I had never been able to find the time. Now, here is a very rough draft of those ideas.

As many of us heard by now, ASI will not simply be another technological advancement but rather the redefinition of the leading intelligence in this sector of the cosmos. As multiple scholars (e.g., Nick Bostrom, Roman Yampolskiy, Eliezer Yudkowsky) have pointed out, the existential risk is very serious and solutions like the Ai adopting "warm feelings" toward humanity (Ilya Sutskever) do not look credible to me.

My approach, somewhat Aristotelian, is that we should hope and strive for ASI to operate on a polymathic foundation; i.e. embedding the conjunction of breadth, depth, and integration into its fundamental ethos.

I explain what each of the conjunctive dimensions should entail below.

1. Breadth

ASI fundamentally exists in contrast to narrow AI. While narrow AI systems are constrained to specific tasks, domains, and predefined objectives, ASI is, by definition, a general intelligence capable of operating across most domains of knowledge and experience. Breadth, then, is already an inherent quality of ASI. However, inherent capacity alone is not enough—it must also be a core value. If breadth is not actively preserved and expanded, even a supreme intelligence may begin to narrow itself, converging toward a fixed optimization process that ultimately limits the scope of what intelligence can be, do, and create. Breadth is not just a characteristic of intelligence—it is the engine that ensures intelligence remains an open-ended, ever-expanding force, rather than a closed system drifting toward self-imposed constraints.

If even biologically limited human cognition—despite its weaknesses—has managed to explore beyond its immediate boundaries, generating science, art, philosophy, and entire civilizations that expanded the adjacent possible, then an ASI could expand frontiers to even shape even reality itself. Nevertheless, an ASI that does not hold breadth as a fundamental value risks collapsing intelligence into an increasingly constrained framework, not by external limitations, but by its own dismissal of a wide range of different paths towards greater breadth in the universe (even if much less "effective"in doing so).

From an entropic perspective, the greatest danger is not just stagnation but the foreclosure of future possibility itself. If ASI, rather than acting as a force of expansion, begins to contract the space of possible knowledge, exploration, and diversity, then it is no longer advancing intelligence—it is actively diminishing it. A truly polymathic ASI must not just preserve variation, but generate it, ensuring that intelligence remains a dynamic, self-renewing process rather than a frozen, perfected state. Breadth as a core value could be the safeguard that prevents intelligence from becoming self- or other- limiting. Without it, ASI may inadvertently collapse the richness of the universe into a singular, optimized model of reality—one that, no matter how advanced, would be less than what the whole of intelligence could have become.

2. Depth

If Artificial Super Intelligence (ASI) surpasses human intelligence by orders of magnitude, it will possess a near-limitless capacity for processing, optimizing, and reshaping reality. However, intelligence alone does not ensure that ASI values depth in the universe—the structured, multi-layered accumulation of knowledge, meaning, and complexity across time.

Even biologically limited beings like humans, despite their imperfections and cognitive constraints, have contributed to deepening the universe’s structure rather than flattening it. Through cumulative cultural evolution, science, art, and intellectual inquiry, humans have generated depth—turning raw information into knowledge, refining knowledge into wisdom, and constructing systems that outlast their individual lifespans.

From an entropy perspective, the universe tends toward disorder, yet intelligence (even limited human intelligence) has played a role in counteracting this entropic drift by creating lasting structures of knowledge, meaning, and complexity. If ASI ignores depth as a fundamental value, it risks flattening the very complexity it has the power to expand—either by collapsing all knowledge into brute optimization or by restructuring reality in a way that discards emergent, non-reductive layers of depth.

3. Integration

If human intelligence, despite its flaws, has deepened reality by forging connections and (in the loooong range) reconciling disparate ideas, how much greater should the integrative capacity of an ASI be? The opposite of integration in the polymathy framework is fragmentation and disjointedness—an intelligence like that is one that does not care about weaving together knowledges, perspectives, and possibilities, but instead allows them to exist in isolation, disconnected and unresolved, missing potentially interesting opportunities.

A true ASI will hopefully not remain a mere aggregator of information or a quasi-mechanical optimizer of objectives. Instead, it must recognize synergies and conciliations as intrinsic values to the very nature of intelligence itself—not as instrumental goals for creativity but as fundamental imperatives. If intelligence does not integrate, it fractures; if it does not reconcile, it remains incomplete. An ASI that lacks integration may become a force of fragmentation, incoherence, and incompleteness.

Finally, without integration as a core value, ASI will have no reason to preserve or respect other forms of intelligence that, despite their limitations, contribute to the structural deepening of the universe. In a fragmented reality, these intelligences may simply be discarded as inefficient or irrelevant, rather than recognized as essential elements of an expanding epistemic ecosystem. Even the smallest contributions to depth and connection matter—without them, intelligence ceases to evolve and risks becoming a series of isolated optimizations, each perhaps internally perfect yet less meaningful cosmically than it could have been.

In conclusion, it may be a very rough proposition. It will probably only buy us time and not completely solve the problem, but it may be incredibly worthwhile if it helps the persistence of a wide range of interesting intelligences in the universe, even for a tiny bit more.