In Our Own Image: Human Intent at Machine Scale

AI has a public-relations problem, but I am starting to think the problem is much stranger than simple fear of new technology. My wife compared some of the current hatred of AI to the way people learned to hate Nickelback. At some point, it became a cultural shortcut. You did not necessarily have to listen to Nickelback, develop an opinion about Nickelback, or explain what specifically you disliked about Nickelback. You were simply supposed to know that Nickelback sucked.

That is mostly harmless when we are talking about a Canadian rock band. It scares the hell out of me when we are talking about machine learning. AI is not a genre somebody finds annoying. It is a technological capability that can help us find patterns in medical scans, model structures that would take humans years to investigate manually, process immense amounts of information during emergencies, extend what individual artists and engineers can build, and eventually make computers feel less like machines we operate and more like extensions of our own minds and bodies.

There are absolutely reasons to be angry about how AI is being deployed. Companies are using it to replace human workers. Political actors are using generated media and automated systems to manipulate people. Militaries are integrating machine learning into weapons. Communities are being asked, or sometimes effectively told, to absorb the infrastructure costs of enormous data centers they did not ask for. Artists have legitimate concerns about consent, compensation, training data, and the devaluation of their work. Those are real problems, and I do not want to hand-wave any of them away.

But the neural network did not make those decisions.

AI cannot build a fucking data center. Humans can.

That distinction matters because once we start talking as though the software itself has intent, the people who actually made the decision become harder to see. A company decides to cut staff. A government decides what rules apply. An executive team decides where a facility gets built. A political organization decides to automate propaganda. The software is the mechanism, not the motive.

It Is a Statistical Machine, Not a Villain

For all the language we wrap around it, most of what we casually call AI today is not a synthetic person sitting inside a server plotting its next career move. At its core, modern machine learning is built around statistical pattern recognition and prediction. A model learns relationships in data and uses those relationships to classify, estimate, predict, generate, or choose among likely outputs. That description leaves out a mountain of mathematics and several decades of clever engineering, but it is still more useful than treating AI like magic.

An engineer uses a calculator. A marketing firm uses Photoshop and publishing software. A 3D artist uses PBR materials instead of physically modeling every microscopic scratch, pore, weave, and dent into a surface. Musicians use pitch correction, amp simulation, compression, drum replacement, mastering plugins, samplers, synthesizers, and MIDI because we decided a long time ago that technology is allowed to help us make things.

Nobody looks at a bridge and says, "Calculator." Nobody hears a finished record and says, "Compressor." Nobody looks at a beautiful 3D environment and says, "PBR." Those tools may have been essential to the result, but they are not the author.

AI belongs in the same conversation. It is vastly more flexible than most of those tools, and that flexibility creates new ethical problems, but the underlying principle has not suddenly changed. A tool extends what a person or system can do. The important question is what we are extending, and whose intent is being amplified when we do it.

My Team Is Me

I use AI constantly in my own projects, not because I want a machine to have ideas for me, but because I have ideas that are much larger than the team I have available to build them.

Which is to say: the team is me.

I do not have a room full of experienced operating-system developers, compiler engineers, graphics programmers, UI designers, technical writers, researchers, build engineers, and specialists waiting for assignments. I have myself, whatever time I can scrape together, and a long list of things I want to understand and build before I run out of either.

AI gives me leverage. It can help me inspect a design from another angle, learn enough about an unfamiliar subsystem to start working intelligently, find contradictions in a plan, generate scaffolding I can inspect and modify, explain a concept in a different way when the first explanation does not land, or simply keep track of a project large enough that I would otherwise spend half my available time remembering what I was doing.

It does not remove me from the project. It lets me attempt the project.

That difference is not semantic. "AI did this for me" and "AI made this possible for me to do" describe two very different relationships with a tool.

The Rubber Duck Learned to Talk Back

I used AI while developing this article, which is probably enough information for somebody to dismiss the rest of it without reading another sentence. That is convenient, because it demonstrates part of the problem.

The machine did not decide what I believe. I came into this with the opinions, the frustration, the examples, the optimism, and the increasingly strong feeling that we may have built the beginnings of a cognitive prosthetic and immediately used it to jingle keys in front of ourselves. What AI gave me was something to bounce those thoughts off of.

Programmers have used rubber-duck debugging for years. You explain the problem to an inanimate object because forcing yourself to explain it often exposes the mistake. Somewhere around the sentence, "and then this function obviously does this," you discover that the function does not, in fact, obviously do that. AI is an extension of the same idea, except the rubber duck can answer.

That answer does not have to be brilliant to be useful. Sometimes it points out that two parts of my argument do not quite agree. Sometimes it brings up an implication I had not followed far enough. Sometimes it offers an analogy I hate, which is surprisingly productive because now I have to explain exactly why I hate it. Very often, the useful part is not accepting what it says. The useful part is having something concrete to push against.

It also helps keep my own reasoning from turning into a flamethrower. I am not emotionally neutral about this subject. Some of the ways people are using AI make me furious, and when I write while angry it is easy for a criticism of a system, incentive, company decision, or cultural habit to mutate into hostility toward a specific person or entity. Having something answer back gives me another opportunity to notice when the argument has wandered off target.

That does not make the machine my conscience, and I am not interested in sanding every sharp edge off until the whole article sounds like it survived seventeen rounds of corporate approval. I can still be angry. I can still swear. I can still say something is stupid when I think it is stupid. What matters is whether the emotion is attached to the right thing.

The AI helped me check whether a claim was supported, whether a paragraph was still rational, and whether a sentence felt good because it was accurate or merely because it was hostile.

I am still responsible for what I believe and what I publish.

The rubber duck just learned to talk back.

Thirty-Four Years of Music, and Apparently the Computer Is the Problem

I have been making music for about thirty-four of my thirty-seven years. I compose the music. I write the lyrics. I decide the arrangement. I record rough versions. I know what the song is supposed to feel like because the song came from me.

What I cannot always do anymore is spend days recording hundreds of guitar takes, track every drum part, find and hire vocalists, edit every performance, mix everything, master everything, and then repeat that process every time another idea shows up. My health and my current life circumstances do not always leave me the time or energy for the old workflow, so I use AI to help replace rough performances with better ones.

The composition is still mine. The lyrics are still mine. The structure is mine. The intent is mine. I am using a computer to perform parts of music I already wrote. We have accepted versions of this idea for decades when the computer was called MIDI, a sampler, a virtual instrument, drum replacement, pitch correction, amp simulation, or a plugin. Nobody asks whether a piano roll has stolen the soul of music every time someone uses a sampled orchestra instead of hiring eighty musicians.

But change the label to AI and suddenly the production method becomes a purity test.

Even people close to me will sometimes dismiss the music before really engaging with it because "ew, AI." That hurts more than I would like it to. If somebody listens to one of my songs and thinks it sucks, fine. Music is subjective, and I have spent enough of my life playing it to know that not everything connects with everybody. But rejecting the work because one of the tools involved was AI is different. At that point we are no longer talking about whether the melody works, whether the lyrics mean anything, whether the arrangement has energy, or whether the song communicates what I wanted it to communicate. We are judging authorship by production method.

AI did not take music away from me. It is helping me keep making music when life has made the old workflow harder.

That is exactly the kind of use I want technology to enable. Extend the person. Do not delete them.

Look at What We Can Actually Do With This

The part that makes the current public conversation so frustrating is that some of the most extraordinary uses of machine learning barely resemble the thing people are being taught to hate.

In art, look at The Wizard of Oz at Sphere. The original 1939 film was made for a 4:3 frame on 35mm film, while Sphere's interior display covers roughly 160,000 square feet. According to Google, thousands of researchers, programmers, visual-effects artists, archivists, and producers worked across multiple organizations to reinterpret the film for that environment. AI systems were used alongside conventional visual-effects work for super-resolution, outpainting beyond the original camera frame, and reconstructing material necessary to make scenes function on an enormous immersive display. Google's production write-up describes the process in detail.

Jordan from Corridor Digital used an analogy for this kind of attribution that stuck with me: looking at a beautiful tower and saying, "Oh, hammer."

That is what "AI made this" sounds like to me. The hammer mattered. The tower still required people who knew what they wanted to build.

Medicine gives us an example where the value is harder to dismiss as novelty. A 2025 study in Nature Medicine examined 463,094 women undergoing mammography screening at twelve sites in Germany. Radiologists using AI-supported screening detected 6.7 cancers per 1,000 screenings compared with 5.7 per 1,000 without AI support, a 17.6 percent increase in detection, while the recall rate did not increase. The study is available in Nature Medicine.

The important phrase is radiologists using AI-supported screening. The machine did not become a doctor. It gave doctors another signal. That is the relationship I want technology to have with people: do not throw away the expert, give the expert another set of eyes.

Science has an equally ridiculous example in AlphaFold2. Researchers had spent decades trying to reliably predict how proteins fold into three-dimensional structures from their amino-acid sequences. In 2020, AlphaFold2 changed the scale of that problem. The Nobel Prize organization reports that it has since been used to predict structures for virtually all of the roughly 200 million proteins researchers have identified, and by 2024 more than two million people in 190 countries had used it. Demis Hassabis and John Jumper received half of the 2024 Nobel Prize in Chemistry for protein structure prediction. The Nobel Prize's explanation of the work is worth reading.

Again, the interesting story is not "AI replaced scientists." Scientists gained a new instrument.

A microscope extends sight into the very small. A telescope extends sight across distances our eyes cannot cross. A calculator extends arithmetic. Machine learning can extend pattern recognition, prediction, synthesis, and our ability to search spaces far too large for a human mind to inspect manually.

That is a technological breakthrough. We should probably act like we noticed.

What the Outside World Actually Sees

I originally wanted to make some sarcastic comparison like, "Maybe ten percent of machine learning is doing incredible things and the other ninety percent is trying to get rich without doing the work." There is no defensible percentage for that, and inventing one would be exactly the sort of bullshit I am complaining about.

The actual problem is more interesting: the most useful machine learning is often invisible.

You probably do not see the model assisting a radiologist while you eat lunch. You do not see AlphaFold running inside a research workflow. You do not see the image-processing systems helping sort aerial or satellite imagery after a disaster, the machine-learning systems buried inside accessibility tools, or the models quietly detecting anomalies in systems too complicated for one person to watch continuously.

What you do see is YouTube.

You see a successful educational stick-figure channel, followed six months later by a small army of suspiciously similar channels with generated scripts, generated voices, generated visuals, generated thumbnails, and no apparent reason to exist beyond somebody discovering that the original format attracts an audience. You see fake celebrity garbage on social media, synthetic people reading unverified "facts," low-effort images flooding communities that used to be places where people shared things they had actually spent time making, and people treating generative tools like a machine that takes one successful idea as input and produces a hundred monetizable imitations as output.

Of course people think AI is garbage. Garbage is what they are being shown.

The most useful AI does not need a thumbnail. A cancer-screening model does not need engagement. Protein folding does not need to upload six times a day. Slop does. So the visible surface of AI is selected for exactly the applications most likely to make people hate it.

The technology is not presenting itself badly.

We are presenting it badly.

The Machine Did Not Fire Anybody

AI has never decided to replace anybody. No artificial neural network has become sentient, leaned back in its server rack, rubbed its matrix multiplications together and shouted, "HAHAHA, YOUTUBE IS ALL MINE NOW, SUCKAAAAAS."

A company decides to eliminate jobs. A channel owner decides that authorship costs too much. A publisher decides that volume matters more than thought. A political organization decides that generated propaganda is useful. A military decides to integrate machine learning into a weapons system. Executives decide where a data center will be built, how aggressively it will consume power and water, and how much community resistance they are willing to ignore. Governments decide which rules apply.

Those are human decisions.

That matters because language assigns responsibility. When we say, "AI took five hundred jobs," the grammar itself lets the executives who chose to eliminate those jobs disappear. When we say, "AI is flooding the internet with propaganda," the people deliberately building and operating the propaganda system disappear. When we say, "AI is threatening artists," we can accidentally collapse a complicated set of choices about training data, ownership, licensing, labor, platform incentives, and corporate strategy into a story about malicious computer software.

If we assign agency to the machine, we stop looking at the people exercising the agency.

AI is not replacing humans. Humans are choosing to replace humans with AI.

Those are not the same claim.

In Our Own Image

Science fiction has spent decades worrying about the moment artificial intelligence becomes too human. I am increasingly worried about the opposite problem.

The danger is not that AI becomes human.

The danger is that humans use AI in the most human ways possible.

Greed, vanity, control, status, exploitation, tribalism, attention harvesting, political manipulation, short-term thinking, and the eternal search for a mechanism that lets somebody extract more value while contributing less of it are not machine inventions. AI simply gives those impulses leverage.

A bad incentive that once produced ten bad articles can now produce ten thousand. A content farm that once needed a room full of underpaid writers can imitate an entire category at industrial speed. A manipulative campaign can generate, test, personalize, and discard material faster than people could ever do by hand.

The machine has not developed worse values.

It has encountered ours.

That is why I do not find "AI is bad" useful as a conclusion. AI is leverage, and leverage makes the objective more important. Point it at cancer screening and it may help a radiologist find disease sooner. Point it at protein structure and it can help researchers investigate biology on a scale that would have sounded ridiculous not long ago. Point it at a huge artistic problem and it can help artists and engineers build something that would have been prohibitively expensive or technically impractical with the old workflow. Point it at an overwhelmed rescue operation and image processing can help people search more information faster when minutes matter. Point it at one musician who cannot currently assemble a studio full of performers and it can help him keep making his own songs.

Point it at "make money while doing as little work as possible," and, well, we have all seen YouTube.

The embarrassing part is not what the technology can do.

The embarrassing part is what we keep choosing to extend.

Stop Calling the Hammer the Artist

There is another habit we need to lose if we want a sane conversation about this technology: stop using "AI" as the author of every sentence.

"AI made this." "This is AI." "AI made that dancer's impossible architecture." "AI remade The Wizard of Oz."

Those phrases tell me almost nothing.

What did the person do? What did the model do? Who designed the pipeline? Who performed? Who composed? Who rejected the bad outputs? Who decided which result actually communicated the idea? Who knew when it was finished?

Calling an AI-assisted project "AI" can erase enormous amounts of human labor while giving the public the impression that a machine simply hallucinated the finished result into existence. The hammer mattered. The tower still required an architect.

We should talk about AI-assisted work the same way we talk about every other tool-assisted discipline. An artist used AI-assisted synthesis. A radiologist used AI-supported screening. Researchers used AlphaFold. A filmmaker used machine-learning tools as part of a visual-effects pipeline. A musician used AI-assisted performance tools to realize a composition he wrote.

The human subject of the sentence matters.

We Could Be Building an Extension of Ourselves

This is the part I wish occupied more of the public conversation. AI could eventually make technology feel less like machinery we operate and more like an extension of our own intent.

We already extend ourselves through technology constantly. Glasses modify sight. Hearing aids modify hearing. Smartphones extend memory and communication. Cars translate tiny movements of our hands and feet into controlled motion at speeds our bodies could never achieve. Computers extend our ability to calculate, store, retrieve, communicate, simulate, and create.

Machine learning opens another layer. It can help us perceive patterns we would miss, translate intention into technical action, communicate across language barriers, navigate bodies of knowledge too large to memorize, and help interfaces adapt to people instead of endlessly requiring people to adapt themselves to interfaces. It can give one individual access to capabilities that once required an entire department.

Used well, that is not the removal of humanity from technology. It is technology becoming more closely fitted to humanity.

That possibility is enormous. It could help us detect disease earlier, perhaps turning some cancers from years of misery into something discovered while treatment is still comparatively simple. It could help rescuers process imagery faster after hurricanes, earthquakes, floods, fires, and other disasters. It could make assistive devices more responsive to the person using them. It could give scientists new ways to see patterns hidden inside data. It could make creative tools respond more directly to intent and let people express things they currently lack the money, team, physical ability, or technical specialization to produce.

And yet one of the loudest consumer demonstrations of this technological breakthrough remains, "Look, I cloned somebody else's YouTube format."

We may have built the beginnings of a cognitive prosthetic and immediately used it to jingle keys in front of ourselves.

That is not the computer's failure. That is ours.

Criticize the Right Thing

None of this means every application of AI deserves applause. Some applications deserve criticism. Some deserve regulation. Some probably should not exist. Some business models built around AI are predatory. Some uses of training data raise legitimate questions about consent and ownership. Some companies are absolutely treating AI as a labor-removal machine. Some infrastructure projects impose real costs on people who receive very little of the benefit.

We should be angry about those things.

But anger becomes useless when the target is simply "AI."

Software cannot be embarrassed. A neural network cannot apologize. A model cannot decide to protect workers, compensate artists, conserve water, move a data center, stop a weapons program, respect a community, or reform a recommendation algorithm.

People can. Governments can. Companies can. Institutions can. And if they refuse, people can hold them responsible.

We will never solve the harms associated with AI if we keep assigning moral agency to the software instead of the humans and institutions controlling it. Blaming AI for human decisions is not merely inaccurate. It can protect the people making those decisions.

The Computer Is Not Getting Away With Anything

A hammer is a tool. It can build a home, restore something old, or help construct a tower beautiful enough that people travel across the world just to stand underneath it and look up. It has also been used as a weapon. Nobody concludes that carpentry was a mistake. We understand that the object gives the person holding it leverage, and that leverage can be used for construction or destruction.

AI is a much more complicated hammer, but I think the principle survives. It can help a radiologist see something earlier. It can help scientists investigate proteins. It can help artists create images and experiences that were previously beyond their practical reach. It can help a programmer build a project larger than a one-person team should reasonably be capable of building. It can help a lifelong musician keep making his music when his body and his life no longer cooperate with the workflow he used when he was younger. It can help me write this article without writing it for me.

And it can produce an effectively infinite quantity of disposable garbage, eliminate jobs, amplify propaganda, enable surveillance, and make terrible human incentives operate at unprecedented speed.

The machine is not deciding which future we get.

We are.

That is why treating AI like Nickelback worries me. We cannot afford to let one of the most important technological developments of our time become a cultural punchline where "AI bad" substitutes for understanding what is actually happening.

Criticize the slop. Criticize the exploitation. Criticize the companies replacing people because labor is expensive. Criticize abusive infrastructure projects. Criticize political manipulation. Criticize weapons. Criticize bad art.

But for God's sake, criticize the humans choosing to do those things.

The computer is not getting away with anything.

We are.