In defense of the ordinary people using artificial intelligence to survive a world that isn’t waiting for their permission
There’s a particular kind of argument that appears whenever a new technology begins eating somebody’s livelihood.
One side points to the machine and says, “Look what it can do.”
The other side points to the people being displaced and says, “Look what it’s doing to us.”
Both are usually telling the truth. They are also talking past each other so energetically that, if we could attach a generator to the discussion, we might solve the electricity shortage.
Then somebody notices a third person. A writer using an AI to organize research. A graphic artist testing compositions with an image generator. A programmer asking a model to untangle some miserable legacy system. A musician trying out arrangements they can’t afford to record with an orchestra. A small-business owner who needs a brochure and has approximately fourteen dollars left after paying the power bill.
And this person becomes the villain.
Apparently, by touching the software, they’ve endorsed every business practice of the company that made it, every infringement someone might commit with it, every worker who may lose a contract because of it, and the eventual destruction of human civilization. Presumably they’re also personally responsible for the increased price of eggs.
I find this a peculiar assignment of guilt.
Not because there are no victims. There are. Not because the companies involved have behaved impeccably. Some haven’t. Not because the people who object to AI are ignorant, hysterical, or somehow afraid of progress. Many of them understand the consequences better than the enthusiasts do, because they’re watching the work disappear from their own desks.
But there’s something grotesque about watching two people standing in the path of an oncoming truck, then deciding that the one who tries to climb onto it must be conspiring with the driver.
Using a tool is not the same thing as approving of the circumstances under which the tool was built.
And wanting to survive an economic upheaval is not a declaration of war against the people being hurt by it.
Those shouldn’t be controversial statements.
Evidently, they are.
I want to begin with the people who are angry, because they’ve earned the right to be.
Imagine being an illustrator who spent fifteen years learning to draw convincing human figures, build an environment, arrange light, tell a story with a composition, and develop a style that clients recognize. You did what the world told you to do. You practiced. You assembled a portfolio. You met deadlines. You learned that “small revision” is a phrase capable of destroying an entire afternoon.
And then a client who previously hired you announces that they can now get “basically the same thing” from a generator. They’d be delighted to offer you a smaller fee to correct the hands.
That’s insulting. It’s also a real business threat.
The photographer who loses routine product shoots, the session musician whose parts are replaced by synthetic tracks, the copywriter watching a publication fill its pages with disposable prose, and the designer asked to produce ten times as many assets for the same salary have comparable reasons to worry.
Even where nobody has yet been fired, the bargaining position changes. The threat of replacement can do economic damage before the replacement works particularly well.
This is not imaginary. The International Labour Organization has documented substantial exposure to generative AI across media and cultural occupations and warned that the transition requires attention to job displacement, compensation, and creative control. A 2026 Stanford working paper examining payroll data found no widespread economy-wide displacement, but reported a pronounced relative employment gap among young workers entering AI-exposed occupations. The picture isn’t a simple apocalypse, and it isn’t harmless, either.[1][2]
So no, I’m not going to tell the illustrator to stop whining and reinvent herself as a prompt engineer. That would be contempt masquerading as career advice.
Nor will I tell the musician that the emotional value of an original performance will magically persuade the advertising agency to pay for one. The agency may genuinely prefer the performance. It may also prefer not to pay the invoice. These preferences have a way of resolving themselves when the budget is approved.
The economic problem deserves serious answers: collective bargaining, meaningful consent and licensing, contracts that protect creative work, transparency about the materials used to build these systems, enforceable remedies for impersonation and infringement, and honest consideration of how the benefits of increased productivity ought to be shared.
It may also require laws, regulations, and institutions that don’t yet exist or aren’t adequate to the job. The details matter, and reasonable people can disagree about what should be prohibited, licensed, compensated, or left alone.
Good. Have that argument. Organize. Sue when there’s a legitimate cause of action. Demand evidence. Negotiate. Vote. Refuse work on terms you find unacceptable.
But then tell me why you’re aiming the moral outrage at a broke freelancer using the same software to keep the lights on.
How, exactly, did the freelancer become the enemy?
I work with computers. If that has taught me anything, it’s that people are remarkably eager to confuse a system’s user with the person who designed the system, the person who owns the system, and the person who profits from the system.
They’re different people.
Sometimes they don’t even like one another.
Suppose a company installs a scheduling system that assigns delivery drivers unrealistic routes. The drivers are punished when they fail to meet the algorithm’s expectations. A driver learns how the system works and uses that knowledge to avoid being assigned a route that will get them fired.
Has the driver endorsed the scheduling system?
Or has the driver recognized that understanding the machinery might keep it from chewing off a finger?
Suppose a teacher uses an AI to adapt reading material for a student who needs a different presentation, then checks every sentence before handing it over. Suppose a small nonprofit uses one to prepare a first draft of a grant application because the alternative is not applying. Suppose an independent comic artist uses generated reference material to develop a pitch that would otherwise be financially impossible.
These situations are not identical. They involve different responsibilities and different risks. Some uses will be appropriate, others questionable, and some flatly unacceptable.
But declaring all of them morally equivalent to a corporation exploiting a dataset is not analysis.
It’s a refusal to analyze.
Programmers have a useful term for this sort of thing: wrong abstraction layer.
You’re blaming the user interface for the decisions made by the database administrator, the software vendor, and the board of directors. The interface may need fixing. It may even be dangerous. But yelling at the person entering a form doesn’t necessarily change the code behind the Submit button.
And if you’ve spent much time around enterprise software, you’ll know that the Submit button is seldom where the truly terrible decisions were made.
Those usually happened six meetings earlier, in a conference room, while somebody explained that the new process would improve efficiency.
They never specify whose efficiency.
There’s an idea buried in a lot of anti-AI rhetoric that I find both understandable and deeply unhelpful: if enough decent people refuse to touch the technology, perhaps it will go away.
In some circumstances, collective refusal can work. Boycotts can change incentives. Labor action can stop production. Public pressure can force companies to withdraw products or change their practices. There is nothing quaint about organized resistance, and nobody should confuse my argument with a suggestion that people surrender it.
But a boycott is a strategy, not a sacrament. It requires a target, a demand, a theory of leverage, and some understanding of what success would look like.
“Nobody respectable should use this” is not, by itself, a plan for dealing with a technology that has already been installed in search engines, office suites, design tools, coding environments, customer-service systems, and phones.
And the people most likely to be hurt by refusing every useful application are not necessarily the people who own those products.
A large corporation can afford a legal department, a research group, an AI procurement policy, and five different ways to automate its operations without calling any of them AI. A sole proprietor can barely afford a replacement laptop.
If the company automates, that is described as digital transformation.
If the sole proprietor uses a generator to put together a flyer, that is described as betraying the arts.
Forgive me if I detect a slight imbalance in the distribution of moral responsibility.
The technology may be bad for a particular labor market. Its implementation may be exploitative. Its outputs may flood marketplaces with cheap, repetitive crap. These are all reasons to ask what rules and economic arrangements ought to govern it.
They are not reasons to pretend that a person facing those conditions has an obligation to remain helpless until the argument is settled.
Reality does not submit an RFC and wait thirty days for comments.
The layoffs, cancellations, new requirements, and shrinking budgets arrive on their own schedule.
You can be entirely correct about the unfairness of a system and still have rent due on Friday.
Of course, there are AI enthusiasts who deserve ridicule.
You know the type. They’ve watched three product demonstrations, absorbed half a dozen venture-capital slogans, and decided the entire history of human creativity was merely an inefficient method for arriving at their subscription service.
They announce that artists are obsolete because they generated a dramatic picture of a woman holding a sword. They declare writers unnecessary after producing six paragraphs that sound like an insurance company attempting to describe love. They predict that doctors, teachers, composers, judges, and probably the household dog will all be replaced by Thursday.
Some have taken to discussing the Singularity with the reverence normally reserved for a religious event. One gets the impression that if the machines don’t become gods soon, these fellows will demand a refund.
Fine. Laugh at them. I do.
But an enthusiast is not necessarily a cultist. Sometimes enthusiasm means a person has found something useful.
The person excited about AI-assisted coding may have spent years trying to learn a language just well enough to automate a tedious part of their work. The person experimenting with a music generator may be a songwriter who can hear a full arrangement in their head but cannot hire a string section or perform their own vocals. The person using image generation may be an artist with limited mobility, or a photographer exploring a surreal composition that would otherwise require an elaborate physical set.
Others simply enjoy making things. They get curious. They test the machine. They discover that it can do something interesting, or something hilarious, or something surprisingly beautiful, and they want to see what happens next.
We used to encourage curiosity.
Now, in certain corners of the internet, an admission of curiosity is treated as a confession of ideological allegiance.
You tried it? Then you’re one of them.
Who the hell is them?
The venture capitalists? The people scraping copyrighted material? The executives planning layoffs? The researchers studying accessibility? The student trying to understand calculus? The woman who made a picture of her elderly dog wearing a crown because the dog is seventeen and probably deserves one?
If you cannot distinguish those people, you have not identified a coherent enemy. You have selected a noun and started throwing rocks at it.
And yes, users have responsibilities. A person who knowingly generates counterfeit commissions, impersonates an artist, lies to a client about how work was produced, or passes synthetic evidence off as a real photograph has made choices that deserve criticism. “The computer did it” is no better an excuse than “the spreadsheet calculated it.”
But that is an argument about conduct.
It is not a useful definition of an entire category of people.
We also need to talk about the flood of garbage.
There is a great deal of it.
The internet was not exactly a pristine gallery before generative AI arrived, but we’ve now handed anyone who wants to saturate a search result, storefront, or social feed an industrial-strength hose. A million nearly identical pictures. Books assembled faster than anyone can check their facts. Synthetic songs whose chief artistic ambition appears to be occupying a search result between two things somebody actually wanted to hear.
Some of the material is harmless fun. Some is sincere amateur work. Some is commercial pollution intended to capture attention or extract money. Some is deliberate fraud.
I have no obligation to defend the whole pile merely because I think the tool can be used well.
In fact, abundant mediocre content is one of the strongest reasons to value editors, curators, artists, critics, archivists, working musicians, photographers, and everyone else whose judgment helps us distinguish what deserves attention from what merely occupies space.
But here is the unpleasant truth: the slush pile is not evidence that every person using the tool has committed a moral offense. It’s evidence that the cost of producing and distributing material has fallen faster than our ability to select, authenticate, compensate, and recommend it.
That’s a distribution problem, a market-design problem, and sometimes a fraud problem. Some of it is a quality problem. And some of it absolutely is a labor problem.
Calling the individual user a villain doesn’t solve any of those problems.
It doesn’t tell the bookstore how to identify fabricated reviews, the streaming service how to handle synthetic impersonation, the photographer how to prove provenance, or the musician how to collect payment when their recording is used without permission.
It does, however, supply the satisfying illusion of having taken a stand.
I would prefer a stand with a little more engineering behind it.
What is the failure? Where does it occur? Who can change it? What incentive keeps reproducing it? What rule would prevent it, and how would we know the rule worked?
We are perfectly capable of being furious and specific.
I recommend it.
There’s a related mistake that comes from the other direction.
Some proponents of AI insist that using the technology is not merely sensible but inevitable, and that anyone who objects is wasting time fighting history.
I disagree.
The direction of technology is not a weather forecast handed down from the heavens. It is shaped by investment, law, institutions, infrastructure, social pressure, and the people who decide what to build. Companies can be made to license material. Products can be regulated. Contracts can limit use. Workers can organize. Audiences can demand human-made work and pay for it.
Of course, this isn’t only about what comes out of the computers. There’s the matter of what goes into them.
A data center isn’t a fluffy white cloud drifting harmlessly across the countryside. It’s an industrial installation. The electricity has to come from somewhere, possibly through new transmission lines and substations. Some cooling systems consume considerable amounts of water, which might otherwise be needed by residents, farmers, or other businesses. These facilities vary enormously by location, design, and cooling technology; it would be dishonest to pretend they all use the same amount. But the costs are real, and so is the question of who pays when demand outruns the infrastructure already in place.[4]
And if you’re living next door to one, you may have a more immediate objection: the goddamned noise. Cooling fans hum around the clock. Backup generators have to be tested. Construction crews aren’t famous for their delicate touch. Maybe the developer can mitigate all that. Good! Make them do it. The neighbor trying to sleep through the racket isn’t an anti-technology zealot just because she would like to hear herself think.[5]
Then there are the deals governments make to lure these facilities into town. Tax abatements, exemptions, infrastructure assistance — and glowing promises of jobs. To be fair, construction creates jobs. Suppliers and other local businesses may benefit, too. A new facility can add taxable property value even when it receives breaks elsewhere. But don’t confuse a few hundred people building a data center with a few hundred people employed inside it ten years later. Those are different numbers, and we should insist that the officials handing out public money know the difference. Ideally before the ceremonial shovels are put away.[6]
I’m not especially reassured by the race to build ever more capable AI, either. There are people doing serious safety research, including evaluations of dangerous capabilities, and some developers have published frameworks for deciding when a system is too risky to release. That’s worth recognizing. It is not a substitute for independent scrutiny, enforceable standards, or an honest discussion about whether the scramble for market share and the promise of AGI are outrunning the safeguards. You can be concerned about an increasingly autonomous system making consequential decisions, being misused, or escaping the competence of its operators without asserting that the machine has developed a murderous personality.[7]
In other words, this is not the same argument as saying we’re building SkyNet. Real systems can do enormous harm while remaining thoroughly stupid in ways their marketing departments would prefer not to advertise. Catastrophic future risks are worth investigating, too, but a risk to study is not an apocalypse that has already happened. We can demand brakes without insisting the truck is secretly a Terminator.
And speaking of things that might become less intelligent, there’s a pretty good argument that careless AI use can make people worse at thinking. A student can turn in a beautiful essay without learning enough to explain the first paragraph. A worker can let a model write the report, summarize the meeting, make the recommendation, and eventually forget how to challenge any of it. Researchers are finding warning signs of cognitive offloading and reduced critical engagement, although the results depend heavily on how the tools are used. The evidence does not justify announcing that everyone who touches a chatbot is getting dumber.[8][9]
That’s the distinction. Having an AI test your reasoning, show you an alternative approach, or help you practice a difficult skill isn’t the same thing as letting it do the thinking you’re supposed to learn. If the tool is doing all your homework, congratulations — you’ve automated the learning out of the lesson.
So yes: ask hard questions about water rights, electric rates, noise ordinances, tax deals, safety evaluations, and what we’re teaching children to delegate. Hold the companies, utilities, school systems, and public officials accountable for the choices they actually control. These are real problems — not a moral invoice that automatically belongs to the freelance writer who used an assistant to meet a deadline.
A person can refuse generative AI for artistic, legal, ethical, or practical reasons and be making a completely reasonable decision.
You do not have to put a generator in your workflow to deserve a livelihood.
But the possibility of changing the rules is not the same thing as the rules having already changed.
And nobody owes the world a demonstration of personal purity at the expense of their ability to function under the rules that exist today.
That is the distinction I want people to grasp.
Opposing exploitative technology and learning how to use technology are not mutually exclusive activities.
You can demand better training-data practices and still learn what the models are capable of. You can support working artists and still use AI to generate a rough composition for your own project. You can insist on disclosures, permissions, and fair compensation while employing a tool whose manufacturer has not yet met every standard you would like to impose.
Sometimes you’ll decide the ethical compromises are too great and use something else. Good. Make that decision consciously. Sometimes you’ll find a model with clearer licensing or a workflow that keeps sensitive material under your control. Sometimes you’ll hire the human professional because you need their knowledge, their judgment, their relationship to the subject, or simply because you value their work and can afford to pay for it.
The point is not to excuse every choice. It’s to preserve the ability to make choices, instead of declaring that all decent people must pretend the choice doesn’t exist.
In technology, I have met people who were very proud of their principles and rather less interested in learning how the systems they opposed actually functioned.
This is a dangerous combination.
If you want to regulate a machine, understanding its capabilities is useful. If you want to negotiate about its effects on your profession, knowing what it can and cannot reliably replace is useful. If you want to catch deception, knowing how the deception is produced is useful.
The people who understand a tool best are not always the ones who approve of it.
Sometimes they’re the ones best equipped to stop its abuses.
A firefighter learns how buildings burn. Nobody accuses the firefighter of endorsing arson.
Here’s the part that makes me angry.
Working people are being asked to solve a structural economic problem through a series of personal loyalty tests.
An editor who has watched freelance budgets collapse is invited to denounce a struggling writer for using an AI-assisted draft. A photographer whose images may have entered a training dataset without consent is encouraged to despise an artist using generated backgrounds. A junior designer is told that learning the tools being adopted by potential employers is an act of treachery against the profession.
Meanwhile the publisher, platform, agency, and software vendor get on with their business.
The contracts are signed. The subscription renews. Someone prepares a slide deck about efficiency. The board asks whether the workforce reduction can be reflected in next quarter’s guidance.
The people at the bottom can spend the afternoon accusing one another of moral failure.
Excellent. Very convenient.
I’m not claiming that criticism of individual AI use is always misplaced. A creator who knowingly violates another person’s rights should be criticized, and sometimes held accountable. A company presenting mass-generated material as the work of named human artists deserves scrutiny. A person who substitutes a machine for a promised professional service without telling the customer is not conducting a courageous experiment in the future of art.
But those are identifiable wrongs.
If your complaint is that AI is depressing compensation and concentrating economic power, direct some of your anger toward compensation and economic power.
If your complaint is that training datasets exploit creators, demand rules for datasets, contracts, and remedies.[3]
If your complaint is that synthetic material is overwhelming the places where art is discovered, demand better provenance, ranking, labeling, and marketplace governance.
And if your complaint is that people are losing the ability to earn a living from work they spent decades learning to do, then for God’s sake, don’t make their situation worse by declaring that learning the new tools disqualifies them from your sympathy.
They’re not betraying their profession.
They’re trying to remain employed in it.
There is no prize for being the last person in the office who knows nothing about the system that replaced the old workflow.
There is also no shame in refusing to use it when your work or convictions require otherwise.
Solidarity ought to be large enough to accommodate both people.
I don’t think artificial intelligence is going to deliver paradise.
I don’t think it will inevitably destroy the world, either. Its consequences will be shaped by human decisions, institutional incentives, technical limitations, and the usual breathtaking talent of organizations to misunderstand their own objectives.
I expect remarkable work to be made with these tools. I expect an extraordinary amount of trash. I expect people to be helped, people to be cheated, careers to change, jobs to be lost, and new jobs to appear that don’t necessarily compensate the people who lost the old ones.
None of those outcomes entitles me to dismiss the real people making decisions in the middle of it.
The artists are not fools for worrying about their livelihoods. The programmers are not fools for wanting to understand the new software. The writers are not villains for discovering that an AI can help them structure an argument. The photographers are not obsolete because a generator can produce a convincing sunset. The musicians are not sentimental relics because a machine can assemble a passable melody.
And the enthusiast who sees an opportunity in this mess isn’t necessarily celebrating the suffering of everyone else.
Maybe they just recognize the machine is coming down the road, they’ve read the warning signs, and they’ve decided that standing in the middle of the lane yelling at the traffic isn’t an especially promising survival strategy.
Don’t mistake that for approval of the driver.
Don’t mistake it for ignorance of the people already hurt.
And don’t mistake the refusal to remain helpless for a willingness to throw somebody else under the bus.
If the road is dangerous, by all means demand a guardrail. Organize to get a speed limit. Investigate the manufacturer. Sue whoever disabled the brakes. Put up warning signs. Build a safer crossing.
Do something that might actually save the next person — or even the entire world.
But while you’re doing all that, get the fuck out of the road.
Learn how the machine works. Decide what you will and won’t use it for. Help other people understand the risks. Defend the people whose labor made the old world work. Fight for fair rules in the new one.
And stop assuming that everybody trying to keep their footing is secretly driving the truck.
The argument about which direction the world ought to turn matters.
So does the fact that it’s already moving.
I intend to keep my feet.
Research notes (context, not an excuse to mistake projections for observed events):
[1] International Labour Organization, Generative AI and the media and culture industry (February 27, 2025): https://www.ilo.org/publications/generative-ai-and-media-and-culture-industry
[2] Erik Brynjolfsson, Bharat Chandar, and Ruyu Chen, Canaries in the Coal Mine? Six Facts about the Recent Employment Effects of Artificial Intelligence, Stanford Digital Economy Lab, revised August 12, 2026: https://digitaleconomy.stanford.edu/publication/canaries-in-the-coal-mine-six-facts-about-the-recent-employment-effects-of-artificial-intelligence/
[3] U.S. Copyright Office, Copyright and Artificial Intelligence, including its report on copyrightability and ongoing analysis of training practices: https://www.copyright.gov/ai/
[4] Lawrence Berkeley National Laboratory, United States Data Center Energy Usage Report: 2025 Update (June 2026), https://bies.lbl.gov/publications/united-states-data-center-energy-2025; Nuoa Lei et al., The water use of data center workloads: A review and assessment of key determinants (June 2025), https://eta.lbl.gov/publications/water-use-data-center-workloads. These sources document rising electricity demand and strong variation in water use by facility and workload; they do not establish that every site’s costs fall on household ratepayers.
[5] Semafor, NIMBYism hits US AI infrastructure buildout (May 7, 2025), https://www.semafor.com/article/05/07/2025/us-ai-infrastructure-buildout-neighbors-complain-noise-virginia. Reports residents’ complaints about data center cooling noise and mitigation efforts.
[6] Washington State Joint Legislative Audit and Review Committee, 2026 tax preference review: Data Centers in Urban Counties (July 2026), https://leg.wa.gov/studies-audits-and-reports/performance-audits/published-audits/2026trdata/ . An example of tax expenditure, reported temporary construction employment, a smaller permanent workforce, and possible property-tax benefits; not a national jobs-to-tax-incentive ratio.
[7] METR, Frontier AI Safety Policies (accessed October 2026), https://evals.alignment.org/fsp . An index of published frontier-developer safety policies, provided for comparison rather than endorsement; the existence of policies is not itself evidence of their sufficiency.
[8] Hao-Ping Lee et al., The Impact of Generative AI on Critical Thinking: Self-Reported Reductions in Cognitive Effort and Confidence Effects From a Survey of Knowledge Workers (CHI 2025), https://www.microsoft.com/en-us/research/publication/the-impact-of-generative-ai-on-critical-thinking-self-reported-reductions-in-cognitive-effort-and-confidence-effects-from-a-survey-of-knowledge-workers/. Self-reported practices and associations, not proof of permanent cognitive decline.
[9] The cognitive impact of ChatGPT in higher education: A systematic review of critical and creative thinking outcomes (2026), https://doi.org/10.1016/j.caeai.2026.100571 . Reports both gains under structured use and risks of superficial engagement and cognitive offloading under unstructured use.
