Who Gets to Shape the Age of AI?

I had this post half-finished when Pope Leo XIV went and published his 42,000+ word encyclical on artificial intelligence. There is something humbling about getting scooped on your own blog topic and core argument on AI by the Bishop of Rome. He has a billion or so readers and he still beat me to hitting “Publish.” And his encyclical asks good questions, and it is worth your time no matter your beliefs.

I am not going to recap all of what the Pope thinks, but I think our outlook on AI lands in a similar place: this moment deserves more than a shrug, and the people living through it deserve more than being told “Get used to it.”

This is a strange and often unsettling moment to be alive, and AI isn’t exactly helping fill the uncertainty gap. In conversations with my friends and colleagues about the future of AI, I’ve heard excitement and dread in equal measure. Many are amazed by what these tools can do while worrying about what they might do to them. I feel some of that myself.

This widespread sense of uneasiness has been hardening into something sharper. We can see the anger breaking into the open. We see the anger in graduates booing commencement speakers telling them AI is the future they should be grateful for. We see the anger in communities protesting local governments and data center construction projects.

This frustration and fear wouldn’t exist if this was all a gimmick. People are unsettled because what is being built is genuinely powerful and is rapidly integrating into our daily lives. The only fair way to reckon with what this technology might take from us is to be just as clear-eyed on what it can give. So, before explaining my worries - and I carry plenty of them - let’s look at the promises of an AI future.

There can be real good here

Some of what AI has already made possible is genuinely remarkable. Take the advancements around protein folding. For decades, determining the three-dimensional structure of a single protein required years of painstaking and expensive lab work. Then, an AI system called AlphaFold learned to predict those structures directly from amino acid sequences, turning what once took years into something that can often be done in minutes.

That matters because proteins are the tiny machines that power nearly every process in the human body, and their function depends entirely on the shape they fold into. When that process goes wrong, the consequences can be devastating, contributing to diseases like Alzheimer’s and Parkinson’s. AlphaFold has now mapped roughly 200 million protein structures and made the results freely available to researchers around the world. More than three million scientists across 190 countries have begun using the database to accelerate drug discovery and deepen our understanding of human disease. The significance of that breakthrough was recognized with the 2024 Nobel Prize in Chemistry, and the pace of progress has only continued since.

The same story is beginning to play out beyond the laboratory as well. Self-driving cars have sounded like a sci-fi concept for years, but now Waymo cars have driven over 170 million miles with no one behind the wheel. There is a version of this future where a blind person or an aging parent who can no longer drive gets their independence back, and where the road is no longer shared with someone driving drunk, texting, or nodding off at the wheel.

But much of AI’s promise is quieter than a Nobel Prize or a car that drives itself. It could be the patient tutor helping a student at eleven at night before a test, the kind of support that typically only belonged to families who could afford it. It could also help the government work better for the people it serves, where the IRS prepares your taxes using information it already has or where applying for one benefit automatically connects you to the others you qualify for instead of burying you in repetitive paperwork.

The pieces for this future already exist, but pieces existing is not the same as building them. And this is where my optimism begins to meet a hard wall. None of these good outcomes are guaranteed nor did they emerge from a broad public conversation about what AI should be used for. What gets built first is whatever pays the fastest for the few companies large enough to develop these tools, and what pays the fastest is not always what helps people the most.

The same speed that gave us AlphaFold is also pushing AI into our lives faster than we can absorb it and faster than we can protect the people it puts at risk. I am not asking us to halt this progress or abandon its promises, but we should insist that its benefits extend beyond the companies profiting from it. Because for many Americans, the version of AI that has shown up in their homes and workplaces has not looked like a medical breakthrough or a safer drive home. It has looked like a threat.

We are witnessing a future being arranged without us

Nobody knocked on our doors and asked whether we wanted our jobs rewritten, our kids’ classroom turned upside down, or our work fed into a machine that now competes with us. A handful of companies made this decision, and the rest of us were told to keep up.

A year ago, the people running these companies were remarkably candid about their ambitions. Anthropic CEO Dario Amodei warned that AI could eliminate half of all entry-level white-collar jobs within a few years. Now that public anxiety and backlash have grown, the rhetoric has softened. The language is less about replacing workers and more about “optimizing” work, “boosting productivity,” and freeing people to focus on higher-value tasks.

But the layoffs have not waited for the messaging to catch up. Companies ranging from Amazon and Microsoft to UPS and Citigroup have all announced major job cuts, pointing in part to AI-driven efficiencies and the need for a different mix of skills. Whatever AI executives choose to call it now, many workers have lived through the original version of their promise: jobs disappearing first and explanations arriving afterward.

But even among the companies pushing AI adoption the hardest, there is remarkably little agreement on what good use actually looks like. Workers are being told to embrace AI, often with little guidance beyond “use it more,” while many executives are still trying to determine whether the promised productivity gains justify the cost.

The uncertainty is showing up in the numbers. KPMG now tracks employee AI usage and folds it into performance reviews, setting targets for how frequently workers use the tools. Yet some of the industry’s own leaders have begun pulling back. Microsoft reportedly moved engineers away from Claude Code after costs ballooned. Uber’s Chief Technology Officer said the company exhausted its annual AI coding budget within four months. Even an executive at Nvidia acknowledged that, for some workloads, compute now costs more than labor. A 2025 industry survey found that 85 percent of companies missed their AI cost forecasts by more than 10 percent.

The same executives who told workers this future was inevitable are now running the numbers and reconsidering their own assumptions. Caught in the middle is the ordinary employee, who never got a say in the matter and is just trying to hold onto a job while the people building the technology and the bosses mandating its use continue arguing about what it is actually for.

It is also worth asking a simple question: what were these systems built from? They did not learn to write and paint and compose out of thin air. They learned from us - from books, articles, music, artwork, and countless other pieces of human creativity - almost none of whose creators were asked for permission and almost none of whom were paid.

That question has now become one of the defining legal battles of the AI era. Authors, artists, musicians, publishers, and news organizations including the New York Times have all sued, arguing that AI companies built enormously valuable products from copyrighted work without permission. The courts have begun drawing some lines and overseeing some mindboggling settlement amounts, but the central question remains unresolved: does training an AI model on someone else’s work constitute fair use, or does it require compensation?

So far, the lines emerging from the courts have been narrow. Learning from copyrighted work may be permissible in some circumstances. Learning from pirated copies appears far less defensible. But for many creators, these distinctions feel secondary. The underlying reality remains that their work helped build products worth billions of dollars, and only afterward did anyone begin debating what they were owed.

And the costs of this do not stop with the people whose work got replaced or whose creations trained these systems. They increasingly reach communities that may have never cared one way or another about ChatGPT or Claude. All this computing has to live somewhere, in data centers that must be built somewhere.

I will be honest: the carbon and water demands of these facilities are not nothing, but I do not believe they are the strongest reason to be concerned. The strongest reason arrives in our monthly bills.

Electricity costs across the country are rising faster than inflation, and the explosion of data centers is becoming part of that story. A single hyperscale data center can consume as much electricity as a small city, and projections suggest data centers could account for roughly 12 percent of U.S. electricity demand by 2028. In places like Virginia, where data centers already consume more than a quarter of the state’s electricity, residents are beginning to feel the consequences in their monthly bills.

Put all of this together and the anger I described at the start makes sense. We have been handed a tool and told to use it, a tool that might eventually replace us, and the only reason it works so well is because it was trained on our expertise and the work of people like us, then taught to hand it all back faster than we ever could. And even if we make peace with AI in our work lives, many of us cannot escape it at home. It is in the electric bill that creeps up every month to keep that new data center running. Everywhere we turn, something enormous is being built on top of our work and our lives, and nobody ever stopped to ask if we wanted it. That is what so many people are reacting to: not just the technology itself, but the feeling that the future is being arranged around them without their consent.

So what do we do?

Every part of what I just described was a choice. Somebody chose what these models would be trained on. Somebody chose to lay people off and blame the software. Somebody chose to run the power lines to the data center and send the bill to you down the road. And the thing about choices is that they can be made differently.

I am not going to pretend there is one clean fix for all of this or that we get to do these in sequence and catch our breath between them. AI has moved faster than even the people working to see it benefit all of humanity could keep up with, so we must take action on several fronts at once, before the technology hardens into a shape we no longer get a say in.

Draw the red lines we will not cross.

We have already witnessed uses of this technology that are simply wrong, and we should say so plainly and ban them outright.

A machine should never be allowed to decide on its own to kill a human being. A synthetic video of a real person saying something they never said should not be allowed to swing an election or destroy a reputation. Nothing built to imitate a child for the purpose of sexual exploitation should be permitted to exist, full stop.

And a person should never be denied the things that hold a life together - medical care, housing, unemployment benefits - by an algorithm with no human being accountable for the decision and no meaningful way to appeal it. This is not hypothetical. A major health insurer is currently being sued over an AI system that allegedly denied care to elderly patients at such a high error rate that most challenged decisions were eventually overturned.

Here is what bothers me: most of these are not clearly illegal yet. We need to start here not because these bans solve the whole problem but because they are where all sides can agree and where Congress could act without having to win the bigger directional fight first. It would prove that we can still look at a powerful technology and decide together that there are places we will not let it go.

Stop letting the AI companies be the only ones to grade their own homework.

The reason this all feels so out of control comes down to how the competition is scored. The prize goes to whoever ships the most powerful system first, and the only people deciding whether that system is safe enough to go live are the same people sprinting to release it. They write the tests, read the results, and they alone decide whether to put it out into the world while the rest of us live with whatever they decide.

Before an AI company turns its most powerful models loose on the public, someone who does not work for that company and is not racing alongside it should get to test the thing and report what they find.

We treat this as ordinary common sense in plenty of other contexts where the stakes run high. An independent health inspector certifies a restaurant’s kitchen. An outside firm audits the books of a public company, a requirement we wrote only after watching what happened when we trusted companies to police themselves. The most powerful technology ever built should not be the one product we let vouch for itself.

Now, an inspection is only as good as the standard it measures against, and right now that standard does not exist. We need a clear, federally established definition of what these systems must be checked against before release and the specific dangers they have to be tested for, written by people who understand the technology and answerable to the rest of us rather than to the companies.

From there, we can handle it the way we already handle other high-stakes industries. Accredited private firms do most of the hands-on testing, measuring each model against that public standard, the same way the firms that audit corporate books are private companies and not government clerks. The government’s job should be to set the bar and keep the power to examine the most powerful models itself, the way the SEC stands behind the auditors without combing through every company’s ledger on its own.

Make the data centers pay their own way.

That electric bill quietly climbing to power a server farm we never agreed to is a choice too, and a fixable one. Right now, the deal runs backwards: AI companies decide they need an enormous amount of power, utilities expand the grid to deliver it, and the costs of those upgrades often get spread across everyone else’s bills.

The rule should be simple: if a company wants to build a massive AI data center, it should pay for the infrastructure required to support it. New substations, transmission lines, and grid upgrades should be borne by the companies creating the demand, not by households who happen to live nearby.

But there is a larger opportunity hiding inside this problem. These companies should not just have to pay for the grid upgrades their demand requires. They should bring new power with them, and it should be clean.

The single largest surge in electricity demand in a generation is coming whether we like it or not. We can meet it by building more fossil fuel infrastructure and locking in decades of pollution and fuel-price volatility, or we can use this moment, and this flood of private investment, to accelerate the transition we have spent years talking about but too often failed to build. Solar, wind, battery storage, and next-generation nuclear power all need investment at a scale that governments and utilities have struggled to deliver. The AI industry is uniquely positioned to provide it because it urgently needs the electricity.

For decades, we have heard that modernizing the grid would be too expensive or too slow. Now, one of the richest industries in history is actively searching for places to invest hundreds of billions of dollars in energy infrastructure. We should make that urgency work for us. A company willing to fund both the grid upgrades and the clean generation its demand requires should move to the front of the line. One that refuses should wait.

Requiring companies to bear those costs would also discourage speculative projects, where firms seek grid capacity they never seriously intend to use. Today, a company can blanket a dozen towns with applications, for more data centers than it ever plans to build, just to see who offers the sweetest deal, and everyone’s bills go up to fund grid expansions for projects that never break ground. Make the company pay its own way, and it stops asking for power it was never serious about using. Handled correctly, this boom could leave us with a better grid than we started with instead of just a bigger bill.

And some of this effort is already underway in red states and blue ones. As of this spring, more than twenty states have set up special rates that require large users such as data centers to cover more of the grid costs they create rather than passing them on to households. The White House has also secured commitments from several major technology companies to build, bring, or purchase their own electricity and cover the associated costs.

This is also where we deal with the argument that appears the moment anyone asks the AI industry to slow down or pay up: what about China?

I take that competition seriously. But the biggest brake on American AI growth is the power supply itself. There is not enough electricity, and there is not enough infrastructure to deliver it quickly. Power, not policy, is increasingly becoming the bottleneck.

So the question runs the other way around. Can we really afford to keep expanding generation the slow, publicly subsidized way and call that winning? If AI companies can help finance a major buildout of clean energy and grid infrastructure, they bring more power online faster, which is exactly what this competition requires. And they avoid the surest way to lose public support: forcing Americans to watch their own bills climb to subsidize someone else’s server farm.

Put AI to work on what matters.

All the actions so far have focused on limiting harm or making the people building these systems bear more responsibility for the costs they create. But this idea is the opposite, and it is the part that we talk about the least. The era of AI is upon us, and if we are going to live with all of this, it should deliver something more than a faster way to write emails and trim our payrolls. It should make good on the things that were promised when this all began: cures, cleaner energy, a government that works efficiently for the people paying for it.

Left alone, the money will flow to whatever pays the fastest, which is usually replacing a worker, not curing a disease. So, the public has to put its thumb on the scale, the way it always has for things that matter and don’t immediately turn a profit. We did not get to the moon by waiting for a company to find it worthwhile. We funded it, and we pointed it. The same principle applies here: public money and public direction aimed at the problems worth solving. Fund the science the market is too impatient for: not just the blockbuster drug, but the cure for the disease too rare to turn a profit, the cleaner battery, the crop that survives a drought. Put serious computing power into the hands of researchers, who today share a sliver of it while the companies hold millions of chips to themselves. Build the tools that will never make anyone rich but could help provide the diagnosis in the rural clinic that has no specialist for two hundred miles or could serve as the patient tutor for the kid whose school could never afford one.

I will be honest: this is the hardest and most aspirational thing on the list, and the one I am least certain of our ability to define and pursue. The others you can write into law and enforce, but this one asks for patience to fund ventures without knowing which bets will pay off. But if we skip past this patience, I fear we will see the expansion of a technology that will make a few companies unimaginably rich while continuing to serve as an excuse for diminishing wages and job opportunities. We will have paid the whole price of this thing and pocketed almost none of the promise.

And if that promise does arrive, the people whose work and knowledge helped make it possible deserve to share in it too.

Share the AI upside with the people who made it possible.

These AI models learned from millions of books, articles, songs, images, and other forms of human creativity, much of it made by people who were never asked and never paid, and that “borrowed” work is a big part of what made these systems as capable as they are now. Right now, the only meaningful action on this issue is happening in courtrooms, one lawsuit at a time. Even when creators win, the payments often arrive years later and amount to a fraction of the value their work helped generate.

We should establish a clearer principle than the courts have yet managed to define: the people whose work made these systems possible have a real claim on the wealth they now produce.

The challenge, of course, is scale. No company can negotiate individually with every writer, musician, artist, photographer, or journalist whose work contributed to training these models, and no single creator can realistically take on a trillion-dollar firm alone. But this is not an entirely new problem.

When a song plays on the radio or streams online, the artist does not send an invoice to every station or platform. A licensing organization sits in the middle, allowing broad use of the catalog, collecting fees, and distributing payments back to the people who made the music. The system is imperfect, but it recognizes an important truth: creative work has value, and the people who create it deserve to share in the proceeds when others profit from it.

Applying that model to AI is harder. We cannot perfectly trace which sentence in which novel shaped a particular response the way we track a song being played. So, the realistic solution will necessarily be rougher. A modest fee on commercially successful AI products could flow into collective licensing systems that compensate creators based on agreed-upon formulas. It would be imperfect and approximate, but approximate-and-paid still beats precise-and-uncompensated, which is the arrangement many creators face today.

Yet the deeper we examine this issue, the harder it becomes to answer who exactly counts as a “creator.” The novelist and musician are obvious cases. But these systems learned from more than books and albums. They learned from product reviews, blog posts, photographs, forum discussions, and the ordinary digital contributions of hundreds of millions of people who never considered themselves artists at all. The closer we look for a clean dividing line between “creators” and everyone else, the more that line begins to disappear.

Which points to a larger idea: the AI dividend.

The Alaska Permanent Fund offers an example of this principle in practice. Since 1976, the oil beneath Alaska has been treated as a shared public resource, and every resident receives a dividend from the wealth it generates. Nobody treats it as charity or a handout. It is their fair share of a common resource.

The knowledge, creativity, and data that made modern AI possible are, in many ways, our common inheritance too. The exact mechanism is open for debate, and I will not pretend there is a simple answer. But the underlying principle seems difficult to deny: if these systems were built from all of us, then some meaningful share of the prosperity they create should flow back to all of us as well. The people whose contributions made this technology possible should not be the last in line to benefit from it.

Wrap it up, Connor

I was only partially kidding that the Pope beat me to it on getting my core argument on AI out there. I keep coming back to his encyclical. He never offered any policy in it, but he grounded his writing in a single question, and it turned out to be the one sitting underneath everything I have written here:

What kind of future are we pointing AI toward, one that serves people, or one that runs them over?

Every proposal in this post is really an attempt to answer that question. Not because any of them are perfect, and not because we can stop technological change, but because technological change is not something that simply happens to us. It is shaped by laws, incentives, investments, and choices made by people.

Right now, too many of these choices are being made by a handful of executives, investors, and engineers while everyone else is told to adapt. The people whose jobs, communities, bills, and children will be affected by these decisions deserve a voice in them.

AI may become one of the most important technologies ever developed. If it does, then its future cannot belong solely to the companies building it. It must also belong to the people who will live with it.

That is the choice in front of us. Not whether AI arrives, but whether it serves the public that made it possible.