Infinite Dimensionality

A speculative theory of technology

Note: this is a “thinking out loud” post, not a polished essay. Consider these speculative notes about how concepts like optimization, dimensionality, and the free energy principle could help inform a viable future with advanced technology and human flourishing.

Universal optimization

The universe optimizes at all scales and all levels. Its most generalizable principles— least action, free energy, evolution—all explain different aspects of optimization. The path of the universe is that of least resistance.

But how does life relate to optimization? Life seems to have the unique ability to defy or subvert optimization. When everything is going downstream, life will move upstream. This is perhaps life’s most defining attribute.

How should we characterize this? Perhaps life is that internal drive to put optimization in service of value. Perhaps life explains intrinsic optimization rather than extrinsic optimization. Or motivated optimization rather than lawful optimization.

But is life truly “defying” optimization? Or is it simply optimizing at higher dimensions? Perhaps life is that which can defy optimization at one dimension to pursue optimization at higher dimensions.

This would indicate a dialectic between optimization and dimensionality:

• Optimization occurs by compressing dimensionality to increase efficiency at lower dimensions.

• Dimensionality occurs by expending energy to expand optimization to higher dimensions

Playing with this dialectic can take us to some strange places.

Infinite dimensionality

Active inference is a theory related to the predictive brain hypothesis that seeks to explain much of how our perception, learning, and action works. In essence, we are always trying to minimize the difference between what we need from the environment versus what we actually experience. The more certain we are that our environment can meet these predictions for what we need, the better.

Active inference would define optimization as the increase of homeostatic certainty. Increased certainty requires the ability to infer and model increasingly greater dimensionality.

For example, if you contrast the "dimensionality" contained in the model of a bacteria with that of a human, human models are vastly more dimensional:

• We have a far greater scale and scope of action

• We can consider more variables and relationships, and project those farther into the future.

• We can vastly improve the accuracy of our predictions through simulations, counterfactuals, etc.

• We can embed ourselves in collective intelligences that encompasses more and more of our environment.

And almost all of this dimensional expansion is due to technology. Technology expands our capacities of perception and action, it enables us to form better predictions for explaining our environments, and it expands our ability to form larger collectives.

In this sense, technology is the natural extension of our desire to optimize for certainty. Because this desire is infinite, the ultimate goal of technology could be said to be infinite dimensionality.

This is, in essence, the fundamental “drive” of life. And because there is no ultimate certainty, life is playing the ultimate infinite game.

The alignment problem

The problem is that dimensionality is expensive. Simple models are cheaper to process. This means we are driven to relentlessly optimize against dimensionality. Satisficing, heuristics, and “gut instincts” are examples of dimensional optimizations that humans have evolved to use in order to conserve energy.

But not all optimization is equal. Optimization could be said to be “good” when it compresses, rather than collapses, dimensionality. Dimensionality that is compressed for optimization can be decompressed with no loss of information, while dimensionality that is collapsed is irretrievably lost. Kolmogorov complexity offers a possible formalization of alignment as conserving information through compression.

This gives us a model for thinking about aligning technology.

• Technology is “aligned” when it expands dimensionality AND increases our ability to compress dimensionality in service of optimization.

• Technology is “misaligned” when it reduces dimensionality OR accelerates the collapse of dimensional optimization.

But this dialectic is always negotiating trade-offs. As optimization occurs at greater dimensionality, more and more energy is required—both to compress dimensionality and defy optimization to pursue even greater dimensionality. So as dimensionality increases, there will be greater risks for over-optimization (a collapse of dimensionality) or for stasis (a local dimensional maxima), and the costs for both will be greater.

Dimensional poverty

Civilization paradigms (like cultures, institutions, and religions) can be considered “viable” when they contain sufficient dimensionality to effectively model reality. Viable paradigms can compress all available dimensionality for optimization, conserving energy for expanding dimensionality.

Our current civilizational paradigm could be characterized as insufficiently “dimensional”.

Technology increasingly expands our dimensional reality:

• global interconnectivity

• information abundance

• human agency at planetary scales

• expanding collective intelligence

• omni engineering (bio, geo)

• crypto primitives, virtualization, etc.

Yet our means of interfacing with reality remain mired in low-dimensional paradigms:

• representative democracy (in almost all forms)

• science as parsimony

• health as diagnosis

• land as asset

• value as money

• art as commerce

• self as autonomous individual

• education as testing

• quality as quantity

These paradigms have been relentlessly optimized for lower dimensionality. This leads to a felt sense of living in dimensional “poverty”—the sense that our civilization is no longer able to effectively model the full dimensionality that it contains. This would be another way to characterize the “meta crisis”—our civilizational paradigms are “leaking” so much dimensionality that a critical threshold has been reached. New paradigms are required.

This also leads to a fragmentation of our models. We have fewer outer meta-blankets that can unite our collective models by compressing maximal dimensionality. Religions used to play this role, compressing all dimensions of reality into legible forms that even the smallest model could optimize. Today that is no longer true. In an interconnected and pluralistic world, every religion is leaking dimensionality.

One hypothesis of this Substack is that only life itself is sufficiently “omni-dimensional” to serve as that outer meta-blanket that can surround and unite our collective blankets.

Technology’s role

From this perspective, the normative role of technology is to optimize certainty in the broadest sense, and thus to optimize the expansion of dimensionality. Technology should thus strive to:

• Reveal and create novel dimensionality

• Compress dimensionality in service of optimization

• Reduce the energy cost of dimensional expansion

• Enable new paradigms for modeling higher dimensionality

At its most abstract, technology should seek paradigms that can support infinite dimensionality. These “infinite” technologies would enable scale-invariant dimensional expansion and compression.

For example, the role that feelings play in human consciousness is an effective paradigm for dimensional compression. At its maximal compression, a feeling is legible as positive or negative valence. Yet any affect can be decompressed into its constituent feelings, up to maximum dimensionality.

This may look like “affective” technologies where even planetary valence can be compressed into affect that can still be legible to the most efficient models. At the highest compression these “affects” would translate to a simple “good” or “bad” valence, yet they could be decompressed for any model with sufficient energy to process their maximum dimensionality.

Imagine if every state’s felt sense of “security” included some component of the security of our planetary collective? Or if our felt sense of “agency” could include the entirety of our interconnected relations? Or if our felt sense of “progress” included second and third order impacts?

This would be a future of infinite dimensionality, yet one where our models would never be at risk of being overwhelmed. Every model would engage with reality at its appropriate level of dimensional optimization.

---