Writing

I recently returned to an essay Dario Amodei wrote way back in the fall of 2024 called “Machines of Loving Grace: How AI Could Transform the World for the Better.” Dare I say: you should spend some time with it.

Dario’s aim is to describe what “powerful AI” could mean for society in the five to ten years after it arrives, not in vague sci-fi terms or “fairy dust,” but in concrete domains: health, neuroscience, economic development, governance, work, and meaning.

His point of view is very ambitious and optimistic.

Rereading it now, I’m struck by how far AI has come since he wrote it: in raw capability, remarkable accomplishments, and the extent to which it has spread into ordinary life.

Beyond the substance of the essay, what surprised me most was how it felt to read about a future that was both grounded and inspiring at the same time. That combination is rarer than it should be, in contrast to much of the current discourse, which risks dwelling on the limitations of today’s AI capabilities and fears about a degraded tomorrow. Sometimes I even hear appeals to outright reject a future that includes artificial intelligence, or a wish for things to simply remain the same.

Dario’s piece tacks the other way, toward the most desirable part of the equation, and leaves aside several of the compromises and negative possibilities that I know weigh on so many people. This is not to say that those concerns don’t matter, but perhaps that we’ve let them crowd out a question that deserves at least equal weighting: what would it mean if these predictions are right?

More pointedly: what if the projections coming from those inside the frontier labs — from the people with the most information about where this technology is headed — aren’t doom marketing, aren’t IPO hype, but a genuine description of the near-term horizon?

Admittedly, those inside frontier labs aren’t neutral observers — they, understandably, carry the most ambitious, fastest-moving version of the future. But conviction is not, by itself, a reason to dismiss their forecast.

If that vision came to be, what would it actually mean for us? For our work, our lives, and our collective ambitions?

The adoption gap

I’ve spent much of the last few months thinking about the adoption gap in public-serving institutions: AI capability is scaling far faster than institutions can metabolize change. This isn’t a new phenomenon; we already know a lot about the pace and manner with which institutions absorb technology as they’ve struggled to keep pace with the rate of change for decades.

Why does the adoption gap matter? Confidence in our institutions now sits at historic lows, the result of years of interactions that have too often felt challenging to navigate, unfair, or out-of-touch. Residents deposit a check from a phone and get same-day delivery, but when they try to access benefits or resolve an unemployment claim, it’s like they enter a different world.

Of course public institutions carry obligations that private companies don’t, but when people feel that distance every day, it chips away at their trust.

New technological capabilities could help address many of these problems, but only if institutions can successfully absorb them. Closing the gap, in other words, isn’t just IT modernization; it may be the most direct path back to faith in institutions.

If public-serving institutions aren’t ready, they risk missing a window — possibly a once-in-a-generation one — to leapfrog over challenges that have ailed them for decades.

The essay points to the same opening. There is, Dario writes, “a clear opportunity for AI to be used to help provision government services” that are “in principle available to everyone but in practice often severely lacking.” An AI “whose job is to give you everything you’re legally entitled to by the government in a way you can understand… would be a big deal.” Building that kind of state capacity, he argues, “both helps to deliver on the promise of equality under the law, and strengthens respect for democratic governance.”

The underlying cause of the adoption gap is a mismatch: The technology grows exponentially. Institutional coordination and governance change slowly, and against great friction — some of it warranted (security, privacy, and legal precedent, etc.), and some of it simply the sediment of how things have always been done.

Dario describes artificial intelligence as a new factor of economic production, entering a world full of older ones — capital, regulation, physical constraints, human institutions. “We should imagine a picture where intelligence is initially heavily bottlenecked by the other factors of production,” he writes, “but over time intelligence itself increasingly routes around the other factors… The key question is how fast it all happens and in what order.” Institutions are central among those other factors. For those of us who champion them, the question is one of fitness: do they adapt to what’s coming or get routed around as the rest of the world moves forward?

So the mission in front of us, as I see it, becomes this: if the exponential pace of the technology is a given, how ought institutions adapt? How do they anticipate the waves of opportunity (and threat) headed toward them — and keep the gap from widening beyond reach?

Two things can be true at the same time

Thinking about this can be dizzying, because, for the most part, the world we walk around in still feels like business as usual. And yet the technology news — even this week, with the release of Anthropic’s Fable model — and the conversations with friends and colleagues at the heart of the technology industry tell a wildly different story. Two things both feel true: that much of today’s world has not been outright transformed by AI, and that we’re conceivably on the verge of a wide-scale technological revolution.

I imagine this is exactly what the early years of every technological revolution feels like. The contradiction isn’t evidence of hot air, but a signature of a gap between what’s being invented and what’s been disseminated. Cue the quintessential William Gibson quote that “the future is already here — it’s just not evenly distributed.” This is why we ought not disregard the view of the frontier from those closest to it.

Granted, nobody can claim to know the exact timing of these changes, but what does it cost to be wrong in each direction?

Be wrong in one direction and you’ve braced for a social transformation whose timing or scale was misjudged. You’re early to a party that won’t happen for years. Embarrassing, maybe. Some wasted resources, for sure.

Be wrong in the other direction and the consequences are harder to stomach. We didn’t heed the warning, which was trumpeted loudly and clearly. We didn’t prepare ourselves, our colleagues, our communities. And when the capability arrived, we were caught flat-footed — surprised by something we’d had advance notice about.

We’re all plotting ourselves along these two axes:

• Belief: from convinced this future is plausible to deeply doubtful
• Action: from experimenting proactively to looking the other way

I’d argue the wisest position (for individuals and institutions) is a dynamic one. Taking in new information, recalibrating as the story unfolds, and experimenting your way forward. Not jumping the gun but agile and at the ready.

Choosing the positive vision on purpose

I work to encourage more people into that experimental, ready-to-engage posture. But I’ll go one step further and tell you where I’ve landed personally: I’m reaching out to those who treat a positive future as the rational target to set our sights on. In my experience, a bold and promising vision has been the only thing that mobilizes individuals and communities to take powerful steps toward change.

This moment asks us to balance hubris and humility — nobody can predict the future, and the labs have been wrong about timing before. Dario concedes as much about his own predictions: they “may be unrealistically utopian. But the important thing is to have an ambitious vision, to be willing to dream big and try things out.” Even if the vision turns out to be only roughly right, it’s still worth steering toward (“shoot for the moon and you’ll land amongst the stars”). And it takes a certain bravery, too — a willingness to look early for a while, or even wrong.

A charge to look inward

There’s something easy to miss about “Machines of Loving Grace.” The categories Dario walks through describe first- and second-order effects. But it’s not a roadmap, it’s an experiment in world building that aims to stretch our temporal imagination. We’re all invited to step inside a changed state to reevaluate our own ambitions from within.

Take one relevant example. In his section on poverty, he imagines advances in crop technology and agricultural supply chains producing “an AI-driven second Green Revolution” — a sequel to the twentieth-century leap in crop yields that saved millions from hunger — “helping close the gap between the developing and developed world.” If anything like that is on the near-term horizon, then institutions dedicated to food access today ought to be first in line to understand it — and to support the pathway toward it — because it sketches a more durable and abundant vision than today’s paradigm of means-tested eligibility programs. A surplus on that scale would drive down the cost of food to the point where the distribution of nutrition is a fundamentally different game. The same logic extends to housing, health, benefits, education, and so on.

The risk of mission-driven individuals sitting this moment out is addressed directly. He says that the positive outcome of powerful AI is not structurally guaranteed: “I see no strong reason to believe AI will preferentially or structurally advance democracy and peace, in the same way that I think it will structurally advance human health and alleviate poverty… If anything, some structural factors seem worrying: AI seems likely to enable much better propaganda and surveillance, both major tools in the autocrat’s toolkit. It’s therefore up to us as individual actors to tilt things in the right direction: if we want AI to favor democracy and individual rights, we are going to have to fight for that outcome.”

In other words, this is a charge to ask ourselves what qualities of public-serving institutions do we want to champion in the new world order? Fairness? Dignity? Humanity? Democracy itself? These need not only present-day advocates, but early adopters activating on tomorrow’s world.

The courage to imagine the curve

If I have a critique of the essay, it’s that the section closest to our work is its thinnest. Dario gives the role of public-serving institutions just a few paragraphs. I suspect that’s because few of us — inside or outside the labs — have a crystalline, detailed vision for how our institutions ought to evolve.

Faster? More efficient? Less bureaucratic? Sure.

But wholly novel ways to serve the public interest? Institutions that anticipate needs and stave off crisis? Systems that do all of this while preserving sovereignty, equity, and privacy?

Those visions are much harder to come by. We haven’t matured our collective imagination for how that future should look and feel. Contributing to it is an important part of the work ahead.

This starts with understanding the implications of exponential growth.

I recently learned that our bodies and minds have been genetically programmed over many generations to expect only linear growth in the world, such as how trees grow a few inches each year. The effect of compounding is hard to hold in the mind’s eye, because for most of human history, we literally never saw it. So linear change is usually all we have the mental capacity to imagine. Researchers call this exponential growth bias, and it comes with a twist: we grossly underestimate exponential growth while remaining confident in our poor judgment.

Try this. Think about how far AI has come in the last twelve or eighteen months. Now, do your best to imagine that the rate of progress doesn’t merely hold steady but catalyzes and multiplies upon itself several fold. It’s tough, right?

Even if progress stopped today, it would take institutions several years to absorb all the capabilities that already exist. The gap is real in the current technology; everything beyond it only widens.

This moment demands cultivating a fundamentally different mindset in which we assume tomorrow’s capabilities will not mirror today’s.

That’s the posture I’m inviting others into: experimental, clear-eyed about the risks, but deliberately pointed toward the future we actually want. And maybe the technology takes longer to arrive than expected. There are worse things than arriving too early to the party — at least you’re ready to party!