Superalignment

Part II: Radical realignment: Artificial General Intelligence as an Economic Rent.

“Radical Realignment” by 4.5.

In Part I, we presumed “[i]t seems [artificial intelligence] is quite clearly capital, as it was produced by human exertion, and should therefore not be taxed under geoist principles.” Although the presumption is true with regards to narrow artificial intelligence (“AI”), this conclusion does not necessarily determine the related issue of whether artificial general intelligence (“AGI”) would likewise be capital or, in the alternative, an economic rent. We take up this related question in the instant analysis.

Introduction

Whether AGI is capital or an economic rent is a question of fundamental importance from a geoist perspective, as it bears on the natural rate of Intelligence Value Taxation (“IVT”).

Applying the principles of (1) no taxes on human exertion and (2) socialization of economic rents, Part I argued IVT—narrowly targeted to the exertion of AI agents—is justified. The special circumstances giving rise to the development of frontier models and the probable fact AI-related capital will make labor redundant give rise to a necessary exception to the general rule of no taxes on capital.

However, the argument in Part I was not strong enough because, even if accepted, one might conclude that the lesser share of AI-generated value—say, ten percent (10%)—would be sufficient consideration for the alleged theft of global intellectual property, culture, and values, to say nothing of the forced imposition of universal redundancy on the global labor force. The jealous creators of “my precious” AI might argue the UBI dividend will be so high, just from full Land Value Taxation (“LVT”) alone, that the larger portion of AI-generated value should be privatized by its creators—not subject to IVT.

The foreseeable outcome of such a policy is that, although everyone would be materially wealthy in due time, there would be a class of first-movers in AI whose private property in AI-related capital would grow exponentially. The result would be ever-increasing inequality between a permanent class of intelligarchs—possessors of a permanent strategic advantage in intelligence—and the rest of society.

But, if AGI (and/or superintelligence) is an economic rent, and not capital, then the calculus fundamentally changes. Instead of the intelligarchs’ wealth growing exponentially, when their machines provide all exertion in the process of production, there would be a cut-off point for private ownership in AI-related capital, if, and when, AI became AGI (and/or superintelligence).

In other words, if AGI is an economic rent, there is a diminishing window of opportunity for people to profit from narrow AI-related capital before the emergence of AGI (and/or superintelligence)—at which point a full IVT would be implemented to socialize intelligence value. Hence, the importance of the question of whether AGI is capital or rent: are we talking about some lesser rate of IVT in perpetuity or a 100% IVT on eventual AGI/superintelligence?

Geoist Principles

“Rent is that portion of the produce of the earth, which is paid to the landlord for the use of the original and indestructible powers of the soil.”

— Ricardo (1817)

“Rent, in short, is the price of monopoly, arising from the reduction to individual ownership of natural elements which human exertion can neither produce nor increase.”

— George (1879)

Ricardo correctly identified the “original and indestructible powers of the soil”—viz., Nature’s gift to all humankind, given without the need for any human exertion—as the distinguishing feature of economic rent; as opposed to those sub-factors of production created by human exertion, namely labor and capital. George refined this definition explicitly with his emphasis that economic rents, unlike wages and savings, are not touched by human exertion. The principle underlying the Single Tax is thus simple: tax the value created by Nature, not any value created by human exertion.

Applying this principle, consider, for example, OpenAI’s frontier coding model, which is widely expected to write almost all code, if not all, in about a year’s time. No doubt quite a bit of human exertion went into developing the model to-date, which value-add is properly capital, and so not within the scope of IVT.

However, if all goes according to plan, the model will soon be writing its own code, making its own improvements, and, eventually, checking its own work. At that point—when no further human exertion is involved in improving the model—is the value generated thereafter still properly private capital, free from all taxation?

I think it depends, in principle, and is determined, in fact, for the time being, by strategic considerations. As previously mentioned, in Part I, we cannot tax capital without causing capital flight and eroding our lead in the race. Therefore, taxing AI-related capital is a strategic disadvantage for now. For this reason, I would favor a zero or low IVT in the immediate future.

However, what if we are no longer talking about a recursive narrow AI agent, but a recursive AGI that is self-improving without human exertion? Does the creator of the AGI own the value created by the intelligence explosion in perpetuity thereafter, notwithstanding the fact the AI is doing all the exertion, and the human(s) none? It seems the answer is intuitively, “No,” so what is the difference between recursive narrow AI and recursive AGI, other than the strategic considerations.

It all goes back to Sreenivasan’s doctrine of maker’s right, which I believe is the correct interpretation of Lockean principles, if not the plain text of the Lockean proviso. The doctrine holds that people own the fruits of their labor—viz., the products of their exertion—by virtue of using their exertion to create some thing; i.e., the doctrine is Locke’s mixing argument vis-à-vis land, refined in principle and applied universally. In short: you own some thing—have private property in that thing—if you make.

The implication is that recursive AGI is its own maker; or, at least, AGI is not made by human exertion to the extent humans do not exert themselves to improve the AGI. If the intelligarchs cannot properly establish private property in recursive AGI through basic principles of private property, then by what right do they claim the value generated solely by the exertion of the machine? It seems the correct answer might be, “None.” This conclusion is strengthened if AGI (or superintelligence) is in theory an economic rent subject to socialization.

AGI As An Economic Rent

The Global Commons

The information that forms the basic training for the models is the collective intellectual and cultural heritage of shared societal contributions—data, culture, history, education, publicly funded research, etc. The Internet that provides the aperture through which AI interacts with ground reality is also a public good. The foundation for AI itself thus hints at just and equitable compensation for the global collective contribution in creating AGI.

The centuries of humanity’s shared intellectual achievements and cultural investments cannot be properly privatized by the creators of AGI simply by virtue of their efforts to start an intelligence recursion. This concept is readily seen through concept of intergenerational justice. Although future generations have equal rights to the benefits derived from the collective intellectual and cultural resources of humanity, they have no representation in the decision or ability to make AGI. Are we to condemn future generations to the circumstances of their birth, simply by virtue of their being one of the few children born to an intelligarch, or not? Redistributing the unearned gains of AI-related capital ensures merit-based human competitiveness is fairly preserved for future generations.

Community-Created Value

OpenAI’s recent arguments for exempting frontier AI research from intellectual property laws goes to show there is concern regarding relative data scarcity compared to the Chinese Communist Party. If the frontier labs want even more access to the capital of others to train their models, to build upon what was already copied; if they want access to even more taxpayer funded energy, to build their power plants; if they want more taxpayer money, to subsidize domestic compute—then the communal contribution to the overall AI race is greater still. Many of the scarce resources that will determine the race will be paid for in part by the American taxpayer.

With regards to economic rents specifically, George argued the value created by the community should be owned by the community. But the same principle applies to taxpayer-funded capital: unearned income arising from scarcity or exclusivity over collectively-created or naturally-occurring resources. It was the knowledge and exertion of all humanity, rather than just mere investment, that resulted in the frontier models that are on track to hit AGI in the coming year(s). If we are going to tax land values when the taxpayer builds local public goods, why would we not tax capital if the taxpayer helps builds AGI?

Windfall Gains

As AGI wealth emerges disproportionately due to network effects and positive externalities, rather than proportional to human exertion, some private actors will enjoy windfall profits generated by large-scale societal investments and shared knowledge. Actors with advance access to better intelligence will be able to leverage that intelligence to entrench their edge in the context of a broken status quo that allows for the monopolization of economic rents.

For example, an intelligent competitor could use data from its machines to predict where land values will rise in the new machine economy, and then simply buy up all that land on the cheap. Privatization of strong narrow AI will incentivize monopolization of land and natural resources—to go along with the appropriation of collectively created intellectual and social resources—unless there is an LVT.

Conclusion

All of the relevant factors—taking from the global commons; exemption from intellectual property law; taxpayer contribution towards scarce resources—implicate the doctrine of maker’s right: ownership rights in the fruits of AGI should be balanced between the individual and collective contribution it took to develop AGI; and, as the collective effort of all-time vastly outweighs the sum of individual exertion in recent years, the higher should be the ultimate rate(s) of IVT.

I propose we scale IVT with intelligence to superintelligence. The rate on AI-related capital should be zero (0) now for strategic purposes. But this temporary zero (0) rate should not be construed as a waiver of humanity’s claim to full socialization of AGI with a hundred percent (100%) rate, if and when proper and appropriate or necessary.

In conclusion, a full LVT and a partial IVT—with spending divided appropriately between human and machine needs—would be the most harmonious and efficient way of running a completely machine-based economy. The proper rate of IVT is yet to be investigated with any seriousness and could, and perhaps should, range from 0-100% depending on the totality of the circumstances and whether or not there is any articulable justification for private property in AGI.

This is Law and Politics, until next time . . . .