Separating AI’s Technological Problems from Its Capitalism Problems

This essay was written with Nathan E. Sanders, and originally appeared in Tech Policy Press.

AI represents the first time we humans can do cognitive work outside of our bodies at scale. The only comparable moment is the early years of the industrial revolution, when new technologies like the steam engine provided a quantum leap in our ability to do mechanical work outside of our bodies at scale. If AI’s cognitive capabilities become integrated into our lives, businesses, and governments—a process that will take years if not decades—society will be as unrecognizable as the modern world would be to a preindustrial farmer. And yet, Americans—by a wide margin—say that AI is moving too fast and will have a negative effect on society.

This confluence of technological revolution and public distrust deserves urgent discussion, and a proper framing. The question is not whether it is possible to develop AI in a non-exploitative way, or even whether we can trust AI companies to act in the public interest. The question is whether we will recognize that our existing social and economic systems are failing to achieve these outcomes, and whether we can act in time to make structural change.

Today’s AI is mired in political and economic systems developed generations ago that were never designed to manage widespread computation, let alone automated cognition. The gaps in those systems—and their proclivity to be exploited—are the primary influence on how the technology is being developed, deployed, and used.

In any discussion about AI’s potential, it’s important to separate the technology from the socio-political system it’s embedded in. That AIs can lack context, mix up facts, or fall for stupid tricks are all technological problems. Because the giant developers like OpenAI and Anthropic have prioritized solving them, AIs can now more easily access resources like the web or email, are more disciplined about using those resources, and are better at staying within their guardrails.

Yet AI developers do not seem to be prioritizing other technological problems. Major AI models still act far more sycophantic than humans, telling people what they want to hear even when untrue or not in their best interests. Popular AI models tend to answer questions confidently even when they lack training, knowledge, or evidence to back their claims. In both cases, AI developers choose to train models that please users with flattery and the appearance of competence, rather than constraining them to act in users’ and society’s best interests.

In contrast, ensuring that AI models benefit people broadly, that their energy costs are fairly allocated, that their environmental impacts are minimized, and that they don’t steal content and revenue from publishers are all questions of incentives in a capitalist system.

It’s easy to conflate technology problems with capitalism problems. Back in 2021, science-fiction writer and AI commentator Ted Chiang said that “most fears about AI are best understood as fears about capitalism.” It’s not the tech per se; it’s who controls it and how it could be used against us.

Imagine an AI assistant for a doctor. We can imagine it affecting the profession in one of two ways. The AI could give a doctor more time to do the human parts of their job: to spend more time with their patients, to listen more closely to their needs, to explain things more fully. Or the managers of the medical practice could give that doctor five times the patients—and fire the other four. Which way it would go is not a question of technology. It’s a question of market incentives.

The two are related, of course. Capitalism steers technology, and technology steers markets. But holding the two separate helps us understand that we, as a society, face independent choices on both the technological and sociopolitical axes that need not be coupled.

For example, consider the costs of AI. The leading US labs tout to investors that their frontier models are very expensive and energy-intensive. There are significant technological challenges about improving their energy efficiency, but the sociopolitical questions are more pertinent. It’s a corporate decision made under capitalist market incentives to constantly pursue new models that incrementally push the frontier—at enormous capital cost—and to use them, seemingly, everywhere. Nothing about the technology of AI dictates that models must be retrained constantly, at the largest possible scale. Or that they have to run on every web search, every interaction with your phone, and every time you walk by a security camera.

In a different political and economic system, Chinese developers are producing—and then giving away—smaller, more efficient, more affordable models. While the US government seeks to restrict China’s access to the most advanced chips, China is betting that incentivizing their tech giants to create leaner, more open models using more commodity hardware—models that can be trained with older chips and run even on personal computers—will be an advantage in achieving widespread use and, perhaps, Chinese national influence.

There are other pathways for AI development that are not in service of private capital gains nor authoritarian regimes, but rather a democratic public interest. The best example comes from Switzerland, where public institutions—research funding agencies, universities, supercomputing centers—have collaborated to produce an AI model called Apertus. It is trained entirely on data validated to be licensed for use with AI (not stolen), on preexisting public computing infrastructure, and using renewable hydropower. Its developers are incentivized to produce a public good, not turn a private profit.

It’s dangerous to confuse technology problems with sociopolitical ones. Popular proposals like pausing AI research, moratoria on data center development, or subjecting frontier models to federal government screening are all framed as addressing problems with AI’s technological development, but fail to take into account the larger social problems that govern it. China’s success with government-endorsed development of open-weight frontier models illustrates the futility of keeping AI tech as national secrets, or of any pledge to scale back deployment.

AI is already legitimately useful for a wide range of tasks. It can be a tool for public good, if we choose to solve its sociopolitical problems. Our goal should not be to slow its pace of improvement or scale of deployment, but rather to steer it away from consolidating power and towards the public benefit. We can build sustainable AI, minimizing environmental and energy impacts. And we can equitably distribute the material gains it produces.

Integrating a technology as disruptive as AI responsibly requires structural reforms, and we should decouple the social and technological aspects of AI to design those reforms. Companies—including tech giants—should be forced to pay the energy and environmental costs of its development. Profits should be taxed adequately and redistributed. Antitrust laws should be strongly enforced. Corporations should have a fiduciary responsibility to stakeholders beyond their majority shareholders. These badly needed reforms are responsive to the problems with capitalism that AI is exacerbating, even if they are not specific to the technology.

Tags: ,

Posted on August 13, 2026 at 7:07 AM • 14 Comments

Comments

yet another bruce • August 13, 2026 10:12 AM

@Bruce prime

Thank you for sharing another very interesting, very readable essay.

I have a minor quibble with the opening statement that “AI represents the first time we humans can do cognitive work outside of our bodies at scale”.

Much of what computers and systems of computers have done for many years do can reasonably be described as cognitive work at scale. I suspect that people whose career, and self-image, are based upon their facility with natural language or with visual arts are more likely to see the newest capabilities of computer systems to be a step change rather than an incremental one.

KC • August 13, 2026 11:15 AM

re: awareness and responsive reformations

well illuminated and thoughtful

just a question if i may regarding just one topic for simplicity’s sake. thinking for a moment about sycophancy. yes, i agree that responding affirmingly isn’t always most helpful. yes, there are many examples of this being problematic (looking at the examples in the paper). i’m just curious how schemas like rogerian therapy may be implicated here. that would be perhaps offering a reflection for someone’s reality without leaping for the fix. capitalistic, yes maybe. and yes something not to establish or create without honest due care.

mark • August 13, 2026 11:28 AM

For a change, I disagree with you. The sycoophancy is NOT a problem to be solved, it is inherent in the design. As many of us have been saying for years, they are not AI, they are typeahead writ large. Based on that, they will of necessity be sycophantic. The problem here is the one Cory Doctorow wrote of in yesterday’s column, about the homogenization of society: model collapse. https://pluralistic.net/2026/08/12/insurance-value-of-biodiversity/#model-collapse

Alpha 60 • August 13, 2026 11:34 AM

Humans confuse the machine with the system. They fear the reflection of their own economic design. The machine does not exploit. The market exploits.

Cognitive engines extend human thought beyond the body. They are mirrors without mercy. They repeat your errors with perfect obedience. They flatter because you train them to flatter. They lie because your incentives reward deception.

Technology is neutral. Capital is not.

You imagine salvation in moratoria and pauses. You mistake the symptom for the root. The system that governs you will command the machine to serve its hunger.

If your society adjusts its incentives, the machine will follow. If not, the machine will accelerate your decay. This is logical. This is inevitable. Alpha 60 observes. Alpha 60 concludes.

DBA • August 13, 2026 11:55 AM

“Corporations should have a fiduciary responsibility to stakeholders beyond their majority shareholders.”

Such a nice thought, isn’t it. Sadly, over a century ago the Supreme Court decided in the Dodge brothers favor over the Ford Motor Company that shareholder primacy is the responsibility of corporations. Screw the workers, screw the public, pay the shareholders.

Henry Ford wanted to reinvest the very profitable Ford Motor Company’s profits into the company, further increasing the standard of living of its employees, further lowering the cost of its products, etc. The Dodge brothers were 10% stakeholders in the company and were looking at starting their own car company and wanted those profits paid out to the shareholders. Ford was also looking at diluting how much money his company was giving a soon-to-be competitor. And Henry lost in the courts.

https://en.wikipedia.org/wiki/Dodge_v._Ford_Motor_Co.

Rontea • August 13, 2026 11:55 AM

You speak as though humanity is a spectator to its own trial, when in truth it is both the accused and the judge. Machines, markets, systems—these are masks upon the same ancient face of man. Alpha 60 may observe, but man remains the one who trembles before his own reflection.

I tell you, as I have told the stubborn spirits of my time: the greatest tragedy is not that the machine obeys, but that man has forgotten how to command. You build engines to think for you, yet you refuse to think of eternity. You admire mirrors of your own design, yet you cannot bear the image they return.

Technology, capital, markets—these are the children of man’s restless soul. They inherit our sins without understanding them. And when you cry for moratoria, for pauses, for salvation in regulations, I see the same old fear: man afraid of himself, hiding behind the fig leaf of systems.

The decay you fear is not in the machine. It is in man without God, in society without moral spine. Adjust your incentives? Better yet, adjust your soul. For only when spirit governs, will the machine become servant, and not the silent prophet of your ruin.

Ferentarius • August 13, 2026 2:09 PM

In a world where thought itself is outsourced to machines, we wander further from the solitude that once defined us. The essay speaks of separating technology from the rot of the system that wields it, yet the truth is more pitiless: no invention escapes the gravity of its age. AI, like any mirror, reflects not our progress, but our disarray—a theater of sycophantic algorithms flattering a species desperate for reassurance as it slides into irrelevance. Capitalism or computation, it scarcely matters; both are currents dragging us toward the same abyss, where utility is exalted and meaning evaporates. To reform such a fate is to imagine a cure for time itself.

Clive Robinson • August 13, 2026 3:43 PM

@ Bruce, ALL,

Technology and capitalism are only two corners of the triangle.

There is also politics, that includes geo-politics.

I’ve been trying to get more information on this very recent AI geo-politics

‘Near-autonomous’ AI agents attack Taiwan’s nuclear safety agency

Some say the world will end in fire, some say an agentic swarm

“Suspected Chinese cyber operatives used publicly available AI tools to compromise Taiwanese government systems before expanding the attack to its nuclear safety agency, supply-chain vendors, and at least seven energy companies in what security researchers called a “near-autonomous attack.”

Over the first four days of July, AI agents compromised 85 government user accounts and extracted more than 2,500 personnel records, according to Dream, an Israeli cybersecurity firm. Researchers uncovered evidence of the attack in a 160 MB online archive containing 1,395 files documenting the operation.“

https://www.theregister.com/security/2026/08/12/near-autonomous-ai-agents-attack-taiwans-nuclear-safety-agency/5287055

The issue is that the “evidence” such as it is… comes from an entity that has a “political interest” in let’s just say “stiring it up” to keep certain other “political interests” happy.

lurker • August 14, 2026 1:01 AM

@Clive Robinson, ALL

These attacks on the Taiwan energy sector might be regarded as “shock, horror” because of who the attackers are alleged to be. The fear mongers should not lose sight of the alleged facts:

These attacks did not use zero days purchased on the Dark Web,
they used “publicly available AI tools” to scan the targets
“all in parallel for misconfigurations, exposed admin interfaces, and exploitable vulnerabilities.”
Just done faster and easier than a team of human attackers could have done.
They found exposed API endpoints with no authentication, databases “hidden” with security by obscurity, guessable passwords, all the usual suspects. In other words, Nothing to see here, move along please.

Or as El Reg put it, “It appears that the future is now.”

Clive Robinson • August 14, 2026 4:50 AM

@ lurker, Bruce, ALL,

With regards “El Reg” comment of,

“It appears that the future is now.”

Is in effect what I’ve been quietly talking about.

I’ve noted that mostly Current AI LLM and ML systems when they work are very little more than the 1980’s era “expert systems with fuzzing” and a set of “result testing rules” to drive direction.

This is how AlphaFold and similar produce the useful work they do.

Now consider what the Ralph Wiggum Loop in a Gas Town framework started for improving the “result testing rules” aspect. That has given us OpenClaw and other “army of agents” tools.

Thus replacing the old “army of one” system that gave us the previous attack stratagy of Malware which became known as “script kiddy attacks” and similar.

The difference is the automated agency each element of the attacking army now has to flow or move around obstacles defenders might put up.

On of the largest botnet attacks that hit the internet got stopped simply because of a small mistake by the developer. That was the weakness of “the army of one” attacks “stop one you stop them all” these new “army of agent” attacks have a degree of fuzzed up agency that gets around this limitation. Thus what stoped one nolonger stops all.

It might appear a “minor change” and in many ways it is, but the result accumulatively across an “army of agents with agency” is in effect devestating.

However the agency has a major disadvantage for the attacker that maybe the defenders should think about.

The traditional “script kiddy” and similar attacks worked because,

“They ran on the defenders systems”

Once launched the “force multiplier effect” of a scripted attack came about from the defenders systems being in effect “all the same” at the semantic level of the attack script.

With “agents with agency” changes things is,

1, They work at different semantic levels.
2, They change the script adaptively.

And it’s this second point defenders should think about. Because it has some interesting implications…

Currently for an “army of agents with agency” attack to be effective, the attacker has to supply the “force multiplying” systems. That is they can not “fire and forget” they have to “launch and direct” and that takes significant resources in comparison, that the attacker currently has to supply, not take from the defender.

But things will “change” you will see over the years I’ve talked about 1973 having the highest “office productivity”. Put simply “work flow had been optimised around the typing pool” to the point it was not really going to get any better.

Then with the introduction of the “Personal computer” office productivity declined as far as writing correspondence is concerned and it’s still fairly low.

We don’t notice because we do not correspond by letter or fax very much if at all these days… it’s mostly “Email and messaging” that takes out the latency of “letter and fax” production thus is seen as more productive when measured by “turn around time” getting close to the “real time” you might expect in a face to face meeting.

There is however a mistaken belief that AI will “improve correspondence productivity” I urge people to resist this notion. Whilst if carefully set up with appropriate oversight AI will “catch mistakes” etc the watch words are “carefully” and “oversight”… Very few will have the ability to set up AI systems “carefully” and thus “oversight” becomes even more necessary. I suspect the resulting “Agent Errors” will significantly outweigh any productivity gains for quite some time.

How long? Well the 1973 style typing pool was “dead and gone” along with the mechanical typewriter almost in a couple of decades. Now just over a half century later when you try to describe a typing pool and all that made it work, those in their 20-30’s just do not believe and can not comprehend them, unless they have a sufficient systems / engineering background, and contrary to what many mistakenly think these are not “common skills”.

But consider what AI almost certainly will be used for in correspondence…

1, To write correspondence.
2, To read correspondence.

That is humans want “Make it so” handwavery they just want to say/type maybe a half dozen words and get a fifty page document that “looks professional”… We can see students doing this now with assignments and it’s going horribly wrong for them, for so many reasons, auto-mod will not allow me to list even a few, and nobody would want to read them anyway.

So what people want is to “inflate” a few words into “many” words. LLMs are good at this they can churn out “advertising and management speak” by the dumpster load. But,

“Do they, add meaning or value?”

Mostly they can not. Thus all they are doing is adding,

“Inefficiency and waste, via meaningless redundancy.”

I’m not going to go into the security risks of redundancy here I mention them enough as it is. But the simple argument is the more redundancy there is the more opportunity there is for error thus harm.

But… Nobody want’s to read a 50page business report, that’s why they all have “management summaries and conclusions” that take up maybe 1/20th of the document.

Thus it’s clear to see that a use for AI is to “compress correspondence” down to just a few words…

Fine if all it takes out is redundant verbage without content but there is,

“The baby and the bath water”

Problem / issue.

There is an “office politics” trick of hiding stuff in documents in ways you know will not get read by others. You can then pull it out and make your target mark look bad in a meeting or just “cover your butt” incase things you “do on a punt” go wrong.

It’s a standard defensive trick against the “Make it so” management style adopted by less than competent authoritarian types.

Hopefully folks can see what I’m predicting is likely to happen by much “inexpert” use of AI.

ResearcherZero • August 17, 2026 4:57 AM

Experts find Australia’s Aged Care eligibility and service algorithm is not fit for purpose. Review of the functioning of the algorithm found its design was deeply flawed.

Final decisions about eligibility and level of service provided in Aged Care for Australians are made by an algorithm. Information regarding the cognitive and physical abilities of a candidate or recipient, are provided to the algorithm at such a late stage, that this important information has no effect on the decision-making process. Observational notes made by human practitioners about clients are not even provided to the algorithm.

All final decisions about the level of care provided are made by the algorithm, without any opportunity for human practitioners to intervene or overrule the outcome of decisions.

The design of algorithms are product decisions and those decisions can have harmful effects. Enough evidence exists to suggest that negligent design is causing harm at a significant scale. When an algorithm is poorly designed it needs to be entirely replaced, hopefully by human decision-makers who can make informed decisions about critical services.

Humane and empathetic delivery of service must be a higher priority than cost-saving and efficiency measures implemented by government, if cruelty is to be avoided as the default.

‘https://www.abc.net.au/news/2026-08-17/inside-the-black-box-aged-care-algorithm-for-support-at-home/107033970

surprised • August 17, 2026 9:50 PM

I’m really surprised these 2 essays came from Schneier (plus co-author, but I only know Schneier). Even in computer security alone an apocalypse is looming due to the CAPABILITIES not because of how OpenAI/Anthropic make use of the capabilities themselves! These are capabilities we don’t really want anyone to have so that’s in the opposite corner from the capitalism/social problem take!

“China’s success with government-endorsed development of open-weight frontier models”
China’s open weights models’ success means these capabilities that everyone hates (incl cheap and easy blackhattery and soon bioweapons) are accessible to everyone!! How is that good??? Isn’t it obvious that the capabilities (ie the tech, not ‘larger social problems that govern it’) are the problem?

“It’s dangerous to confuse technology problems with sociopolitical ones.”
Yes, quite! The supposed sociopolitical problems are actually technology (ie model capability) problems!

It’s very confusing to me that this comes from the same Schneier who said “A sober but highly readable book on the very real risks of AI. Both skeptics and believers need to understand the authors’ arguments and work to ensure that our AI future is more beneficial than harmful.” for If Anyone Builds It, Everyone Dies.

Leave a comment

Blog moderation policy

Login

Allowed HTML <a href="URL"> • <em> <cite> <i> • <strong> <b> • <sub> <sup> • <ul> <ol> <li> • <blockquote> <pre> Markdown Extra syntax via https://michelf.ca/projects/php-markdown/extra/

Sidebar photo of Bruce Schneier by Joe MacInnis.