In a recent paper, artist Tanya Klowden and mathematician Terence Tao suggest we adopt a Copernican view of intelligence. We must accept that human intelligence is not the “immobile center” of the universe, but rather one node in a much broader cognitive system. Tao also sat down for an interview with Dwarkesh Patel and argued that AI excels at scaled intelligence (e.g. synthesizing vast amounts of knowledge) while humans are better at trial and error. He went on to discuss new methods of scientific inquiry and research that could leverage, and derisk, the strengths of AI.
Another Dwarkesh episode featured Dylan Patel, CEO of SemiAnalysis, and what stood out was the careful discussion of how microchip production is likely to be the bottleneck in AI growth within a few years (Energy and memory are the key bottlenecks at the moment). Dylan explained that a single gigawatt of AI compute can drive millions of extreme ultraviolet (EUV) lithography passes per year. This level of activity requires the full attention of 3.5 ASML’s $300-million EUV machines. And ASML only manufactures about 60 to 70 of these machines annually.
So, it seems natural that AI, as the fast-emerging intelligence partner of humans and as the primary accelerant of economic activity today, will increasingly be pointed at relieving that bottleneck. Whether this application will be focused on improving chip yields, accelerating EUV production, or developing alternative methods of synthesizing chips, etc., the image that comes to mind for me is of a vortex: AI used to unlock AI.
I am drawing that vortical image from The Last Economy by Emad Mostaque. In that book, Mostaque argues that there are three types of flows that characterize economic (or any?) activity: Circular Flow (the vortex), Gradient Flow (the high-pressure movement of capital and effort toward extreme bottlenecks—see above), and Harmonic Flow (The Human Structure, which answers the question of how we coexist with economic forces). As I interpret the argument, the vortex is about the self-reinforcing accelerant that drives the economy.
In an earlier piece on Hamilton and the founding of the United States, I essentially argued that Hamilton helped to initiate a US-centric vortex, based on economic potential, confidence in government, and debt. Government debt, which could seem like a weakness, was used to drive attachment of individuals to the government and the dollar, and to drive their economic activity. Hamilton created a vortex of American government, capital investment, and enterprising activity which has been one of the defining phenomena of world history since that time.
In my view, AI is creating a new vortex which is disrupting the American and nation-state centric paradigm that has been prevalent. Whether one vortex subsumes the other in the near-term, or the two exist side by side, in the near-term, remains unclear. But the combination of a decentered human being and a new, self-propelling vortex are the defining features of our age. The other chaos that we witness around us is, in some sense, a consequence of this fundamental disturbance. The digital era of the internet and social media may have initiated some of the patterns of destabilization that are now accelerating, but AI is the conceptual culmination of that disrupting force.
We are so concerned about whether AI has latent intent, but we should be more concerned about how we clarify and direct our own intent, given the enormous power of AI. The new vortex, which has certain inhuman capabilities, disturbs our sense of ourselves and of our own history. But it needn’t distract us from our own capability for form an intent, and to follow through with it.
If we have truly discovered a machine that can either solve or cause an unimaginable set of problems, we had better invest ourselves more heavily in determining which of that enormous range of options we will dedicate ourselves to pursuing. If we are flustered or distracted by the power of the tool, we will lose the opportunity and our nerve.
What this all points to is the need to codify intent. This trend has already emerged in AI-forward companies: What this looks like is system prompts, shared “AI skills”, and agent management systems. But what does codified intent look like at the social level?
It points to metaprompts, shared in a common library, organized by projects and /or teams. That will be the subject of my next post.