Theft, legally and conceptually, involves the unauthorized taking of property such that the original owner is deprived of it. Copyright infringement, by contrast, involves unauthorized reproduction or use of a protected work—without depriving the original owner of possession.
This is not a semantic quibble. It defines how law, economics, and enforcement mechanisms respond:
Theft is zero-sum: one party gains, the other loses.
Copyright infringement is non-rivalrous: the original still exists, but the right to exclusive use is violated.
U.S. law (Title 17, U.S. Code) treats infringement as a civil or, in some cases, criminal offense—but not as theft. The Supreme Court and multiple circuit courts affirm this distinction. Institutions use “theft” language not for accuracy, but to evoke moral panic—to justify criminalization through analogy.
That rhetorical sleight of hand undermines public understanding of intellectual property, distorts legal discourse, and disguises protectionism as victimhood.
Fair use doctrines (U.S.) and similar exceptions (UK) exist precisely because viewing content—like visiting a website—is not theft. It is access under public terms. No possession is taken, no deprivation occurs. It’s reading, not stealing.
Yes, court cases are pending. But that is not unique to AI. Copyright ambiguity has accompanied every major technology: photocopiers, cassette tapes, VCRs, search engines, and cloud storage. This is not crisis—it’s jurisprudential lag, and it’s normal.
Singling out generative AI as inherently illegitimate because of unresolved copyright disputes is a category error. It’s an argument that has been made—and failed—at every prior inflection point in technological history.
The myth of random internet scraping
No, AI companies do not “randomly suck up the internet.” Foundational models are trained on highly curated, structured datasets, often:
Annotated,
Legally reviewed,
Filtered for quality and bias,
Or outright licensed.
Why? Because models trained on noise perform like noise. Random input yields unusable output. These systems require signal-rich, semantically coherent corpora to function at scale.
In practice, this includes licensed books, codebases, scientific repositories, and public data under permissive terms. The idea of “pirate models” trawling the web is a fiction—it’s structurally incoherent and technologically infeasible.
So yes, there are governance questions around dataset provenance—but this is not chaos. It’s a design question, not a scandal.
AI does not inherently displace people or drain public resources.
Land use follows ownership and planning—not algorithmic intent. Unless land is expropriated or seized, data centers are built with legal transfer, market negotiation, and zoning permission. Displacement results from housing markets, inequality, and policy failure—not GPUs.
AI didn’t invent gentrification, homelessness, or urban land scarcity.
And the ecological argument? It misses the basics:
AI compute is consolidated in hyperscale cloud environments—more efficient than fragmented, firm-level data centers.
AI chips (TPUs, GPUs) are radically more efficient per task than traditional CPUs.
Major cloud providers operate at higher renewable ratios than the global average.
The energy footprint is real—but lower per unit of productivity than traditional compute.
“Every dollar spent on AI is a dollar not spent on [insert favorite cause]”
That is not budgeting—it’s authoritarian framing. Budgeting is not subtraction. It’s prioritization under competing constraints. Unless a dollar is explicitly diverted from a specific program, this argument becomes a moral imposition masquerading as logic.
You’re not analyzing—you’re dictating values.
And for the record, I prefer pluralism and open systems. If someone wants to prioritize AI, climate tech, or education, that’s their right—not yours to preempt.
Misusing Porter’s value chain
Finally, claiming Michael Porter’s name while advocating for AI abolition is intellectually dishonest.
Porter’s value chain was designed to help firms analyze how they create competitive value. Your chart flips this logic: it assumes every activity is harmful and no value is created.
If that were true, no AI firm would survive market forces. But they do. Because users find value.
Your diagram is not a value chain. It is a moral ledger with no credits, only debits. That’s not Porter. It’s not strategy. And it’s not analysis.
It’s activism pretending to be theory—and that distinction matters.
Thanks Swen, this is helpful feedback -- the construction "copyright theft" is clearly problematic here. I'll take this on board. IP theft is different and this is an important distinction.
Our positions are clearly at direct odds, and that's fine, but I do disagree about the accusation of "intellectual dishonesty". I've written and researched extensively on the complexities of value concepts, and this is a fair bit more than a simple borrowing of Porter's work. The notion of "consumer value" has been developed significantly over time since the 1980s, and this broad upstream/downstream mapping is now a key part of carbon accounting models. I believe it's right to acknowledge the origins of the value chain concept, but this is many generations removed from Porter's work.
That makes sense. Thank you. For what it’s worth, I hope you keep an open mind about AI — it will serve a vital function, but it can’t replace all coders and such. That’s a self-serving spin from AI companies, but the structure of AI makes that impossible. Still, it allows us something that is difficult for humans — and that is bias-free perception.
I hope I haven't been presumptuous, I'm assuming not, but I've organically found reasons to share this article and handy dandy important informative resources twice today. Got into an interesting conversation with an engineer offering a free talk on AI for us Gippslanders. Seemed to think talking about the ethical considerations is too time consuming and he's not planning to do a Ted Talk...
Not presumptuous at all!!! Please share away, only acknowledging that of course this is super limited and only meant as an introduction to some of the most public harms. There are SO many more, it breaks my heart.
Thank you for helping make space for these conversations <3
I love this article for going deeply into something that is mostly talked about vaguely and conceptually. I do wonder if many of the issues you bring up about AI could also be solved by AI. Do we think that the inability to power data centers sustainably could be improved by the research power that AI could bring to science? Could AI help to identify negative discriminatory social trends and root them out in a way that is no longer considered biased? Reading stuff like this, I just worry that we're throwing out the tool baby with the bathwater because of the tools that use it like assholes.
Great start! Here is a source you may find valuable: https://journals.sagepub.com/doi/10.1177/20539517251340603
Can you believe that paper has been open in a browser tab on my computer all month?!?!
Amazing! I have been meaning to dig into it and yours is a great way in
This article is incredibly helpful in describing issues and solutions.
Copyright infringement is not theft.
Theft, legally and conceptually, involves the unauthorized taking of property such that the original owner is deprived of it. Copyright infringement, by contrast, involves unauthorized reproduction or use of a protected work—without depriving the original owner of possession.
This is not a semantic quibble. It defines how law, economics, and enforcement mechanisms respond:
Theft is zero-sum: one party gains, the other loses.
Copyright infringement is non-rivalrous: the original still exists, but the right to exclusive use is violated.
U.S. law (Title 17, U.S. Code) treats infringement as a civil or, in some cases, criminal offense—but not as theft. The Supreme Court and multiple circuit courts affirm this distinction. Institutions use “theft” language not for accuracy, but to evoke moral panic—to justify criminalization through analogy.
That rhetorical sleight of hand undermines public understanding of intellectual property, distorts legal discourse, and disguises protectionism as victimhood.
Fair use doctrines (U.S.) and similar exceptions (UK) exist precisely because viewing content—like visiting a website—is not theft. It is access under public terms. No possession is taken, no deprivation occurs. It’s reading, not stealing.
Yes, court cases are pending. But that is not unique to AI. Copyright ambiguity has accompanied every major technology: photocopiers, cassette tapes, VCRs, search engines, and cloud storage. This is not crisis—it’s jurisprudential lag, and it’s normal.
Singling out generative AI as inherently illegitimate because of unresolved copyright disputes is a category error. It’s an argument that has been made—and failed—at every prior inflection point in technological history.
The myth of random internet scraping
No, AI companies do not “randomly suck up the internet.” Foundational models are trained on highly curated, structured datasets, often:
Annotated,
Legally reviewed,
Filtered for quality and bias,
Or outright licensed.
Why? Because models trained on noise perform like noise. Random input yields unusable output. These systems require signal-rich, semantically coherent corpora to function at scale.
In practice, this includes licensed books, codebases, scientific repositories, and public data under permissive terms. The idea of “pirate models” trawling the web is a fiction—it’s structurally incoherent and technologically infeasible.
So yes, there are governance questions around dataset provenance—but this is not chaos. It’s a design question, not a scandal.
AI does not inherently displace people or drain public resources.
Land use follows ownership and planning—not algorithmic intent. Unless land is expropriated or seized, data centers are built with legal transfer, market negotiation, and zoning permission. Displacement results from housing markets, inequality, and policy failure—not GPUs.
AI didn’t invent gentrification, homelessness, or urban land scarcity.
And the ecological argument? It misses the basics:
AI compute is consolidated in hyperscale cloud environments—more efficient than fragmented, firm-level data centers.
AI chips (TPUs, GPUs) are radically more efficient per task than traditional CPUs.
Major cloud providers operate at higher renewable ratios than the global average.
The energy footprint is real—but lower per unit of productivity than traditional compute.
“Every dollar spent on AI is a dollar not spent on [insert favorite cause]”
That is not budgeting—it’s authoritarian framing. Budgeting is not subtraction. It’s prioritization under competing constraints. Unless a dollar is explicitly diverted from a specific program, this argument becomes a moral imposition masquerading as logic.
You’re not analyzing—you’re dictating values.
And for the record, I prefer pluralism and open systems. If someone wants to prioritize AI, climate tech, or education, that’s their right—not yours to preempt.
Misusing Porter’s value chain
Finally, claiming Michael Porter’s name while advocating for AI abolition is intellectually dishonest.
Porter’s value chain was designed to help firms analyze how they create competitive value. Your chart flips this logic: it assumes every activity is harmful and no value is created.
If that were true, no AI firm would survive market forces. But they do. Because users find value.
Your diagram is not a value chain. It is a moral ledger with no credits, only debits. That’s not Porter. It’s not strategy. And it’s not analysis.
It’s activism pretending to be theory—and that distinction matters.
Thanks Swen, this is helpful feedback -- the construction "copyright theft" is clearly problematic here. I'll take this on board. IP theft is different and this is an important distinction.
Our positions are clearly at direct odds, and that's fine, but I do disagree about the accusation of "intellectual dishonesty". I've written and researched extensively on the complexities of value concepts, and this is a fair bit more than a simple borrowing of Porter's work. The notion of "consumer value" has been developed significantly over time since the 1980s, and this broad upstream/downstream mapping is now a key part of carbon accounting models. I believe it's right to acknowledge the origins of the value chain concept, but this is many generations removed from Porter's work.
Thank you for sharing and restacking!
That makes sense. Thank you. For what it’s worth, I hope you keep an open mind about AI — it will serve a vital function, but it can’t replace all coders and such. That’s a self-serving spin from AI companies, but the structure of AI makes that impossible. Still, it allows us something that is difficult for humans — and that is bias-free perception.
I hope I haven't been presumptuous, I'm assuming not, but I've organically found reasons to share this article and handy dandy important informative resources twice today. Got into an interesting conversation with an engineer offering a free talk on AI for us Gippslanders. Seemed to think talking about the ethical considerations is too time consuming and he's not planning to do a Ted Talk...
Not presumptuous at all!!! Please share away, only acknowledging that of course this is super limited and only meant as an introduction to some of the most public harms. There are SO many more, it breaks my heart.
Thank you for helping make space for these conversations <3
I love this article for going deeply into something that is mostly talked about vaguely and conceptually. I do wonder if many of the issues you bring up about AI could also be solved by AI. Do we think that the inability to power data centers sustainably could be improved by the research power that AI could bring to science? Could AI help to identify negative discriminatory social trends and root them out in a way that is no longer considered biased? Reading stuff like this, I just worry that we're throwing out the tool baby with the bathwater because of the tools that use it like assholes.