new frontiers

Many undercurrents and overcurrents have pushed AI from niche research circles to the hot seat of global politics: within the labs, and increasingly so the government, it is well-understood that this is no longer just technology or even industry, but a question of national power — and one we may have no choice but to pursue quickly. And yet, outside of these rooms, people encounter the same “miracle” technology as unemployment or AI slop reels, all while being asked to accept the inevitability of it all. A big, ugly, windowless colossus of a building is coming to your backyard. It will ruin your view, raise your utility bills, and eventually take your job.

There is a distinct and noticeable gap between the promised grand AGI future that we are supposedly building and the state of the world today — generative AI pretty much only has PMF in search, code, and making images. Everything else is a joke.

This is not shocking information, and I think most techies would also agree that this is the case. What concerns me, then, is what to make of it? What does it mean when the most transformative technology of our time is charted to usher in an age of abundance and simultaneously struggles to find use cases beyond better Google or doomscrollable slop?

And if even the people at the epicenter are hazy at best about the elephant in the room, how do we expect the rest of the country to feel?

It’s no surprise, then, that anti-AI sentiment is becoming largely bipartisan, polling more negatively than ICE (but still above Iran!). This presents a huge issue when the public interest is in conflict with the national interest, and the government “by the people, for the people” is in a divergent reality from its people. And yet, from America’s AI Action Plan 2025, we gotta “Build, Baby, Build!”

There is some general awareness that public sentiment around AI is negative, and I often hear the following takes:

• Most people are Luddites, and their opinions don’t matter because they are uneducated chuds.
• Their opinions matter, but will change if we make the product really good.

Both of these require that time be on our side, and I’m not sure that it is. The sheer magnitude of just how much people hate AI seems to be grossly underestimated, and there is a pretty obvious reckoning approaching.

For illustrative purposes, suppose you are candidate Douggy McDougface and you’re running for president in the 2028 election. You don’t need to be technical to notice that being anti-AI polls well. People are angry and scared: about data centers in their backyard, about their brain-rotted borderline-illiterate kids, about the tech oligarchy. And, even better for you, this holds across party lines. So you do the publicly rational thing: No more data centers eating up our power and guzzling our water. Protect American jobs. Stop Silicon Valley from running the country.

Then you win.

Unfortunately, sometime between [insert your town] and the Oval Office, you have been made acutely aware that AI is becoming a question of national power; that China is not going to stop building because Americans find data centers ugly; that the intelligence sitting inside a handful of American companies may be one of the most strategically important assets of the century.

So now Douggy McDougface has a problem.

You can imagine where this starts to go: maybe Douggy slams Sam and Dario on Twitter and then slides into the DMs asking what they need to build another ten gigawatts. Maybe the public notices, and maybe they get angrier. Maybe the political response is then to regulate, restrict, or even nationalize the labs — which, from Douggy’s perspective, conveniently resolves both problems at once: the state gets control of an increasingly strategic resource, and he gets to say he finally put Silicon Valley in its place.

It’s unclear how any of this actually goes (and thanks for humoring me), but the point is that, the way things are going, Democratic incentives increasingly reward politicians for opposing AI at exactly the same time that geopolitical incentives reward governments for accelerating it. In such a way, the labs could win Washington and still lose the country.

The problem is that much of the negative effects of AI are measured in months/years, and the positive ones in decades. As such, public opinion is existential: I fear a future in which we absorb the immediate costs (job displacement, civil unrest) and prevent ourselves from reaching the future that might justify them (scientific discovery, abundance, robotics, etc.). The good future is won by turning bits into atoms: factories, labs, grids, hospitals, logistics, defense. But a buildout of that scale requires public consent measured in decades, which we have not put nearly enough effort into earning.

The most pressing bottleneck for AI is legitimacy.

It’s not that frontier labs are irrational or malicious; I feel extraordinarily lucky to work with some of the kindest, smartest, and most thoughtful people I have ever met. It is that the rational action for each actor on a macro scale leads to net worse outcomes. Labs receive the symbolic status of national instruments, but the incentives of privatized labs are inherently misaligned with “AI for good.” The billions in spending must eventually be justified by revenue, and the shortest path to revenue runs through enterprises. The easiest product to sell is one that reduces labor costs. By contrast, curing neglected diseases, improving public schools, modernizing infrastructure, and accelerating basic science produce benefits that are diffuse, difficult to price, and slow to arrive.

There is a sense of hopelessness and deep societal mistrust that seems to be the overtone of my generation, and it is both incredibly saddening and concerning to me that AI is shaping up to be the flagship issue at which we direct this sentiment. Too much nationalism isn’t a good thing, but too little is how nations fall, especially in such a critical era of new technology.

But what does this look like? To me, some form of nationalization seems inevitable. My hope is that we can kill two birds with one stone, in that not only would the incentives be shifted to actually useful research instead of enterprise sales, but it would also give America a reason to dream.

The federal government has the capital, time horizon, and authority to make abundance a primary objective. The model of the government setting a mission, funding research, and then becoming a large customer has been one historical way of promoting healthy competition, and helped produce the space program, the internet, modern semiconductors, and much of biotech.

AI needs a similar contract directed toward a clear set of unmistakably national goals, be it drug discovery, education, manufacturing, energy, etc.

As an aside, this would be more aligned with the long-term outcomes of frontier labs anyways. There comes a certain point where more intelligence yields diminishing returns: the difference in output from a 130 IQ person vs. a 150 IQ person when filing taxes is negligible. The difference between a very smart physicist and Feynman is historic. The primary focus of the labs is for frontier intelligence, which much more aligns in the long term with goals like discovering new drugs, understanding the human body, and decoding the universe much more than it does enterprise SaaS. And yet, the incentives that be have forced us into some backward race.

As an observation, the sentiment in China seems much more optimistic and accepting of AI. Automation is less of a concern when there is reason to believe that the state will secure your livelihood. It helps, too, to watch skylines transform from straw villages to skyscrapers in one lifetime. There is reason to believe.

The United States, on the other hand, provides no such assurance for its people. Whether or not massive labor displacement actually occurs (which I don’t think it will), the sentiment is real, and it is pressing. Our country needs a reason to believe in not only the government, but a collectively brighter future.

This is not a task solvable by branding exercises or traditional marketing. Aesthetically vague Codex billboards or influencer campaigns serve enterprise and consumer purposes, but this is categorically different from what it takes to build a compelling cultural vision of AI.

Like it or not, OpenAI/ChatGPT is far and wide synonymous to AI. With great power comes great responsibility, and we have both reason and duty to care — it is our responsibility to enable people to understand, imagine, and hope for the incredible technology and the future that we represent.

Though real, durable optimism will come with the tangible benefits of the technology, there is much to be done culturally in the meantime. This is what I had hoped would come out of Patrick Collison’s new aesthetics grant, or A24’s AI lab, both of which were exciting steps in the right direction, yet neither have been able to create a compelling vision. I have many more thoughts as to why (deterioration of centralized media, the delocalization of aesthetics, the spirit of art in general, etc.), but that is an entire essay in itself. Cultural change can be propagated through keystone social nodes: all of which are human, all of which we must inspire. The point remains that this is incredibly directionally important, and, maybe, the best position to figure it out is from within a lab. AI needs a public mandate, not better ads.

How do we earn back trust?

There is much work being done with public interest deployment groups (in hospitals, schools, labs, government): OpenAI tools have reached more than 500,000 students and employees across the CSU system, 90,000 workers across 3,500 government agencies, major hospital systems, and national laboratories that host roughly 15,000 scientists. Yet, these efforts remain faint-to-invisible in the public eye. It is difficult for people to draw a clean line from more intelligence to shorter hospital waits, better schools, cheaper energy, or longer lives.

In 2019, China’s Ministry of Science and Technology defined national AI experimental zones as places for “technology demonstrations, policy experiments, and social experiments.” Cities were instructed to test AI in education, healthcare, eldercare, government, transportation, environmental protection, and other public-facing domains; track effects on behavior, employment, and income; submit annual reports; and produce models that could be replicated nationally. By 2024, China reported operating 18 such zones.

While the sanctity of such experiments under real-world political incentives is fuzzy at best, the directional initiative of real, integrated research — taking real action and bringing real people into the loop — is crucial. We must show that we want to iterate on AI as a public good for everyone, rather than waiting for enterprise products to spill over into daily life.

What might this look like? We could start, for one, with the kids (who are not exactly alright). Sixty-four percent of American teens already use AI chatbots, yet 40% of American fourth graders currently read below NAEP Basic. Concretely, this could start with funding experimental opt-in pilot programs for AI-assisted education in districts that need it the most. We treat it like the serious research that it is: if it works, we scale, and if not, we say so and iterate. Trust follows proactive transparency.

The same idea could apply to other surfaces: hospitals, power grids, supply chains, etc., etc. We must treat AI diffusion with the same optimistic, open, and iterative mindset that we do model research: this is maybe the most important experiment of our time!

Done properly, AI has the potential to become not just another industry, but the quest of a generation — my generation. This is the future I believe in, and the future I want to work toward. American AI needs a public mandate, and America needs a new frontier.