Playing the game on the field
Nobody knows what's going on
We are in the “early innings” of AI’s evolution, and everything around us is accelerating. Models are growing larger and more capable, capital continues to flood in, and progress compounds at a pace that often feels alien. Many call this inevitable, with some like Elon even being quoted as saying “humans are bootstrapping AI’s development.”
Leopold Aschenbrenner, in his widely read Situational Awareness essays, for example, projects AGI arriving by 2027, nationalization of the industry in the United States, and a Manhattan Project-style sprint into superintelligence.
The truth is a bit simpler: nobody knows what’s going on.
A lot of VC’s at the moment say they’re “playing the game on the field,” a truism they use to dismiss the notion that they have no idea what’s going on.
What’s clear is that for lack of a better idea, the present game is about scale, compute, and power at all costs, with little concern for sustainable economics, profitability, or alignment.
A number of different articles have emerged recently making the case that current investment strategies are misguided. My favorites include Ed Zitron’s AI is a Money Trap (also: The AI Bubble’s Impossible Promises) and Jerry Neumann’s AI Will Not Make You Rich. For more on my thoughts concerning these pieces and an “AI Bubble” please see here.
In what follows I’d like to explore the implications of machine intelligence colliding with human limits. AI can’t be examined in a vacuum since its trajectory will be defined as much by political, societal, and cultural forces as by technology itself, leaving us to grapple with a more fundamental question: how do we live with it?
The “Too Fast” Threshold
Something else is clear: machines accelerate faster than humans. They don’t eat or sleep, for example. What remains to be seen is whether or not they can accelerate indefinitely. Either way, humans are at a speed disadvantage: we have cognitive, cultural, and institutional half-lives. We metabolize change more slowly, requiring reflection, context, sometimes even over-reaction (see: the markets), and recovery.
As AI continues to develop, at some point I think society decides AI is moving too fast. Maybe it’s the “job loss” and “AI displacement” narrative. Maybe it’s tech CEO’s wielding disproportionate power over the imminent future. Maybe it’s deepening social and economic inequality. As I noted earlier, who knows?
What counts here is not whether it’s actually true that AI is the reason for perceived change or not. All that matters is the perception that AI is to blame.
If society does decide that things are moving too quickly, and that AI is the scapegoat, there will be resistance. To be clear, I’m not fear mongering or prophesying collapse. I’m speaking of resistance in the ordinary sense: friction, refusal, the insistence on time to breathe.
The resistance will be society’s immune response to being pulled at machine speed when human speed simply cannot keep up. The resistance need not be violent or dramatic. It may take the form of regulation, or ostracising those who do use AI, or labor strikes at companies that implement AI too broadly.
As Greek Prime Minister Kyriakos Mitsotakis warned at a recent AI forum, “unless people actually see benefits, personal benefits, to this revolution, they will tend to become very skeptical…a recipe for significant social unrest.”
For now, I’m calling this the “Too Fast Threshold.”
As the Intelligence Curse by Luke Drago and Rudolf Laine makes clear, intelligence is not a universal good. Its expansion often destabilizes existing systems, reducing human leverage rather than enhancing it. Scaling machine intelligence may therefore accelerate society past the point where traditional institutions can adapt, entrenching the very powers it was meant to liberate us from. Although I haven’t attended them, I get a sense this is part of the point of Peter Theil’s lectures on the Anti-Christ.
In such an environment, even well-intentioned advances can erode the fragile alignment between human needs and institutional incentives.
Ironically, current capital investment is sprinting toward scale, toward training runs that exceed any plausible return profile, and toward a monoculture of capability that neglects durability, interpretability, and alignment with human needs.
That sprint will almost certainly succeed on its own terms, meaning we will develop technology capable of rendering us obsolete. In so doing, we may also be stuck with a distrustful public turning on AI as the scapegoat for society’s problems.
The Transition
What happens next will matter more than what happens now, for in the immediate aftermath of the pushback comes the transition. Machine intelligence will continue accelerating, and humans will begin to adapt, unevenly and imperfectly.
Thomas Piketty’s Capital in the 21st Century describes the balance of power between labor and capital from 1945 to 1985 as an aberration, not the norm. In fact, he argues, the long arc of history returns systematically to capital dominance. AI appears to be accelerating that reversion.
Investors often speak gleefully of the returns to come when AI ushers in an “age of abundance.” But those returns appear founded on rendering “inefficient” human labor obsolete, replacing it with much more efficient machines or robots. It’s a vision of value predicated on doing away with people.
Beyond the dystopia of “replacing” humans, that path involves sacrificing at least a generation that will suffer dramatic economic disruption. Even at 75-85% AGI capability, that would mean mass job displacement will trigger protective regulation across the US and Europe.
Far from impacting just the individuals who lose their jobs, governments will fight to self-preserve as well. If any meaningful amount of jobs are lost, the tax base could collapse, meaning governing institutions won’t be able to provide even basic services (not to mention exacerbating their current debt crises). You can imagine the chaos as politicians get voted out and social disruption increases.
What makes this particularly volatile is the speed and breadth of the transition. Unlike historical job displacement that affected specific sectors over decades, this kind of job displacement won’t be selective, affecting white-collar sectors and impacting bankers, consultants, and lawyers as much as factory workers and blue-collar industry labor.
Once dependent on scarce human expertise, intelligence itself is being reimagined as the output of machines running at scale. Massive investments are being made to construct data centers, interconnects, and power infrastructure – a planetary-scale substitution of capital for labor.
Should AI have even a small impact on jobs, it will contribute to the collapse of the old balance between human effort and non-human assets. That won’t – cannot – happen quietly. Maybe I’m delusional, but nothing that important can happen without significant societal consequences that must be accounted for in the projection.
Capital is the Lever
Ironically, capital is also the last significant form of leverage humans still hold over machines. What once required cultivating rare human talent can now be replicated and scaled through machines, making capital the decisive input. They can generate, scale, and replicate knowledge, but they cannot autonomously finance fabs, secure energy, or direct money into infrastructure. For now, humans still decide where capital is allocated, and thus the development of this transformation. That leverage may be temporary, but it’s decisive in the short term.
What follows may resemble a test of stewardship: whether remaining capital is deployed to build tools and incentives that allow humanity to coexist with machine intelligence, rather than be consumed by it. As I’ve written before, returns are healthy and necessary – one cannot hope to save the world unsustainably. What’s more, I won’t pretend to know what challenges await the intrepid investors of the future, but stewarding capital during the transition will be both difficult, and paramount.
Coexistence
Again, as intelligence commoditizes, labor, both physical and knowledge, loses pricing power. If that’s the case, one wonders what happens next. How does humanity coexist in a world where economic output, the organizing principle of our society, is dominated by machines?
We’ve covered what replacement looks like, and, at least at first glance, it isn’t great. Does augmentation work? A recent back and forth between David Sacks and Eric Schmidt on the All-In Podcast I think properly encapsulates the tension:
Sacks begins saying, “AI, at least as we know it today, is not end to end, it has to be prompted. You get an answer, that answer has to be validated. Then you have to ask a new question because it never gives you exactly what you want. You have to apply more context. You have to go through an iterative loop. Finally you get to an answer that has business value.
The way Balaji (Srinivasan) puts it is that AI is not end to end, it’s middle to middle. Humans are end to end. And so as a result of that, instead of AI replacing all of us, AI will be very synergistic with humans because we can define the objective function. We do the prompting. And we work with it to iterate, and it does a lot of the work in the middle.”
Schmidt replies with the following: “What you just said is exactly what’s going to happen for the next few years… To me, the real question is, when does it cross over to having its own volition, its own ability to seek information and solve new problems?
In an earlier part of the same conversation, Eric Schmidt says that if AIs can improve themselves through recursive learning, they may be able to define their own objective functions. Recounting a test involving giving an AI information from 1902 and seeing if it can develop special and then general relativity as Einstein did in 1905 and 1915, Schmidt explains: “if we can solve that problem, then I think it’s over. Then we get to AGI, and it’s a whole different world.”
The Uncertainty Paradox / Event Horizon
Perhaps most striking is that experts like Schmidt just shrug when asked about post-AGI scenarios. Predictions confidently extend to 2030-2035, and then they vanish. Even the most sophisticated analysts building these models admit they have no framework for post-superintelligence scenarios.
As we’ve mentioned, perhaps that’s because all signs point to some dystopian apocalyptic collapse. Perhaps technologists just cannot say the hard part out loud. Many allocators are already responding, though: getting cautious about fundraising, going liquid, taking chips off the table precisely because of this uncertainty.
In short, we appear to be confidently building toward a future we can’t actually envision. Which means we can’t quite comprehend the societal implications, either. We quibble about job loss, return profiles, and regulatory reactions, but the real problem is navigating a transition where the endpoint remains fundamentally unknowable.
One might argue that this isn’t new. For instance, when the internet was being developed people didn’t envision what all came about as a result of it. The same could be said for the industrial revolution.
Nevertheless, the speed at which AI is evolving is unprecedented, thereby amplifying the potential impact because we’ll be forced to adapt even more quickly to a future for which we aren’t prepared.
The Consensus Paradox / Boundary Condition
If there isn’t stable consensus about what comes next, we’re confronted with a boundary condition for our species: the limit within which our system’s logic or organizing principles apply.
Beyond this boundary condition (AGI, Superintelligence, 2027-2035, etc) new dynamics may emerge, and we may be faced with a new logic – new organizing principles – subject to or governed by different rules or inputs. Changes of that magnitude can be frightening, no matter the circumstances (or even the outcomes), hence the “Too Fast Threshold” described above.
Part of what it means to invest, however, particularly at early stages, is to examine “the future” as a set of possibilities. Investors both imagine what the future might look like and – by allocating capital – participate in shaping it.
The excitement related to AI stems precisely from an explosion in that set of possibilities: the number of plausible futures has multiplied dramatically, and the frontier feels almost infinite. In fact, grappling with the notion of “infinity” and what it might be worth is why we’ve seen valuations explode beyond reason and people invoke “God” when referring to machine intelligence.
This is the paradox.
Because the set of possibilities that lie beyond the boundary condition is so vast – in short, because consensus doesn’t exist – investors have piled into the only consensus that does exist, which is scale, scale, and even more scale (chips, compute, data centers, etc.).
In other words, because they are risk averse enough to avoid thinking about and investing in what comes after scaling frontier models (thinking beyond the boundary condition), they have brought forward the risk that should be distributed to the future into the present by over-investing in consensus.
Scaling artificial intelligence has created immense value, but most of that value has already been captured. Investing in frontier AI labs was profitable when it was non-consensus. The infrastructure buildout will continue, but the marginal returns are already compressing. Everyone knows this is a bubble, and fortunes can be made during bubbles (!), the only question is when it pops.
What’s fascinating to explore now is what lies beyond the boundary condition. Considering what comes next. We at Nazaré believe it’s algorithms, and we will explore as much in our next essay.
Algorithms for training, inference, and anything that manages to materially improve cost, efficiency, and performance per unit of any constrained resouce. In many current cases the constrained resource is compute, but it is already becoming power.
Algorithms offer superior returns per marginal dollar precisely because they’re non‑consensus. They are not uncorrelated to the only consensus that currently exists, meaning that algorithms are to AI today what frontier models, scaled compute, and data centers were to AI in 2022.