Guest Post: More Answers to the Dwarkesh AI Essay Competition

Answers from a Philosophy PhD

I’m very lucky that I have a lot of extremely smart friends who like to think about hard problems. When Dwarkesh published his essay competition, I sent the questions over to a few folks just to see what they thought. One of them is an old college friend and roommate, Nikhil Dominic. Nikhil is currently doing a PhD in Philosophy at Cornell, previously did a Masters in Philosophy from NYU, and studied Philosophy and Econ at Columbia. This is a man who has spent a lot of time thinking about a lot of things. I thought he had great answers, so I convinced him to write them up a bit more formally so I could publish them on the blog.

---

With OpenAI’s new raise at an $852B valuation, OpenAI Foundation’s stake is now worth $180B. Anthropic’s cofounders have pledged to donate 80% of their wealth. Nobody seems to have a concrete idea of how to deploy 100s of billions (soon trillions) of wealth productively to “make AI go well”. If you were in charge of the OpenAI Foundation right now, what exactly would you do? And when? It’s not enough to identify a cause you think is important, because that doesn’t answer the fundamental problem of how you convert money to impact. Identify the concrete strategy you recommend pursuing.

What should countries which are not currently in the AI production chain (semis, energy, frontier models, robotics) do in order to not get totally sidestepped by transformative AI? If you’re the leader of India or Nigeria, what do you do right now?

I say that this is an answer to prompt three, but it’s really about three and four at the same time.

AI is like oil. The combined economic output of America’s AI hyperscalers already contributes something like 75% of GDP growth. This is, we are told, just the handle on the hockey-stick: we await Final AI, the AI that can scalably replace humans on any arbitrary (non-fine-motor-skills-based) task. Let’s make some assumptions. If Final AI’s impact is more like the sewing machine, then all of this special pleading is unnecessary; let’s instead assume that Final AI won’t open up new jobs for humans. Competition between AI labs would be good for human opportunity, so as a worst-case scenario let’s say that Final AI is winner-take-all. Final AI doesn’t presuppose Final Robotics, so there may still be place in the supply chain for human manufacturers and primary resource-extractors.

If all of this is true, AI will be a kind of eternal oil fountain. It will endlessly generate money and drive output, and may well require some amount of labor in the margin, but it will also be a trap: it will funnel massive gains into one sector of the economy while atrophying everything else. The heads of AI labs may currently gesture towards their future generosity, but this is no guarantee that they won’t eventually decide to hoard their gains. This is how resource-rich nations fail, and how nightmares of a perpetual underclass are realized.

How do successful countries escape the resource trap? Oil-wealthy nations like Norway aren’t good places to live because everyone’s employed on an oil rig. They are able to fund their generous social services because they hold their oil wealth in a sovereign wealth fund. The US, too, now has a sovereign wealth fund, perhaps the one far-sighted act of Trump 2. The terms of the question presume a model where AI gains flow largely to private owners, and through them to philanthropy. (This is, admittedly, the most likely outcome.) But unless the future of private philanthropy is essentially private governance, it cannot replace the structural transformations enabled by strong state capacity to invest in infrastructure. Local knowledge matters. American philanthropy could not build Chinese high-speed rail. The field in general is waking up to these facts: OpenAI’s “Industrial Policy for the Intelligence Age” doc proposes a SWF, and even the taxes required to fund it. If Sam Altman really wants his UBI, this is how to get it.

The point here is that there is already a model for success and a model for failure for single-resource-based economies. The difficulties are not conceptual but political and practical: how quickly do we transition from private to public ownership of AI labs? Does the government begin acquiring small stakes now or do they await Final AI and deploy eminent domain? Should the state be part of the board? How much of a SWF should go to dividends and how much to improving state capacity? These are difficult questions to answer, but there’s at least a playbook one can consult. But this playbook only tells us what to do when the domain of single-resource dependence is one country. Final AI doesn’t just affect one country. So far only China has been able to provide a competitor model, and that too only by distilling American frontier output. There is no obvious future here where every country gets its own model, especially if Final AI really is winner-take-all.

Whatever political difficulties there may be in nationalizing or at least acquiring shares in the “winner,” these will be exponentially worse in sharing those gains internationally. If you imagine the future, picture Americans living off their UBI like the consoomers in Wall-E, while middle-income countries manufacture and low-income countries mine and/or starve. Work itself will be something only non-Americans have to experience, and that too only in its most malignant and back-breaking forms.

If I were in charge of India, then, I would begin trying to lay the political foundation for the belief that Final AI ought to belong to all humankind, not just the citizens of the country of the lab in which it is achieved. This will strike some as parasitic. But I think this is not just what’s needed for AI to “go well” for all mankind: it’s also philosophically true. If Final AI is achieved, Congolese miners will have contributed to its creation, however distally.

Practically, rather than some kind of massive global wealth fund, the most efficient approach here might be some kind of nation-based AI protectionism. In China, you use the Chinese model. In Nigeria, you use the Nigerian model. These models might, under the hood, be exactly the same, but when you use the Nigerian model, Nigerians get a payout. Poorer countries will still lag behind in terms of physical infrastructure, but they will at least have access to whatever information layer is needed to organize their resources most efficiently and hopefully build out quality-converging infrastructure long-term. (It is unclear what entrepreneurship will look like under Final AI: are there still human entrepreneurs who have to bring goods to market, or do AI-CEOs compete amongst themselves, hiring the occasional human where needed?) If these countries are able to implement local AI-backed UBIs, the global relationship to work also changes. Manufacturing and resource-extraction shift from being extremely low-wage jobs with a labor surplus of roughly every person in Asia and Africa to premium work that has to pay commensurate to its difficulty.

These bold predictions all depend, of course, on just how vertical the blade of the hockey stick turns out to be. A global UBI that pays 13 cents a month would be a farce. I remain somewhat skeptical that the Final AI described at the top of this post will really arrive so soon. That said, it’s still eminently valuable to consider the limit case and start laying the groundwork for adaptation now rather than when we’re all underwater.

PS: You may also like my answers, below.