Atom feed Latest essays

Why training AI can't be IP theft

  • Posted in cyber

AI is a huge subject, so it’s hard to boil my thoughts down into any single digestible take. That’s probably a good thing. As a rule, if you can fit your understanding of something complex into a tweet, you’re usually wrong. So I’m continuing to divide and conquer here, eat the elephant one bite at a time, etc.

Right now I want to address one specific question: whether people have the right to train AI in the first place. The argument that they do not1 goes like this:

When a corporation trains generative AI they have unfairly used other people’s work without consent or compensation to create a new product they own. Worse, the new product directly competes with the original workers. Since the corporations didn’t own the original material and weren’t granted any specific rights to use it for training, they did not have the right to train with it. When the work was published, there was no expectation it would be used like this, as the technology didn’t exist and people did not even consider “training” as a possibility. Ultimately, the material is copyrighted, and this action violates the authors’ copyright.

I have spent a lot of time thinking about this argument and its implications. Unfortunately, even though I think that while this identifies a legitimate complaint, the argument is dangerously wrong, and the consequences of acting on it (especially enforcing a new IP right) would be disastrous. Let me work through why:

The complaint is real

Artists wanting to use copyright to limit the “right to train” isn’t the right approach, but not because their complaint isn’t valid. Sometimes a course of action is bad because the goal is bad, but in this case I think people making this complaint are trying to address a real problem.

I agree that the dynamic of corporations making for-profit tools using previously published material to directly compete with the original authors, especially when that work was published freely, is “bad.” This is also a real thing companies want to do. Replacing labor that has to be paid wages with capital that can be owned outright increases profits, which is every company’s purpose. And there’s certainly a push right now to do this. For owners and executives production without workers has always been the dream. But even though it’s economically incentivized for corporations, the wholesale replacement of human work in creative industries would be disastrous for art, artists, and society as a whole.

So there’s a fine line to walk here, because I don’t want to dismiss the fear. The problem is real and the emotions are valid, but that doesn’t mean none of the reactions are reactionary and dangerous. And the idea that corporations training on material is copyright infringement is just that.

The learning rights approach

So let me focus in on the idea that one needs to license a “right to train”, especially for training that uses copyrighted work. Although I’m ultimately going to argue against it, I think this is a reasonable first thought. It’s also a very serious proposal that’s actively being argued for in significant forums.

Copyright isn’t a stupid first thought. Copyright (or creative rights in general) intuitively seems like the relevant mechanism for protecting work from unauthorized uses and plagiarism, since the AI models are trained using copyrighted work that is licensed for public viewing but not for commercial use. Fundamentally, the thing copyright is “for” is making sure artists are paid for their work.

This was one of my first thoughts too. Looking at the inputs and outputs, as well as the overall dynamic of unfair exploitation of creative work, “copyright violation” is a good place to start. I even have a draft article where I was going to argue for this same point myself. But as I’ve thought through the problem further, that logic breaks down. And the more I work through it, every IP-based argument I’ve seen to try to support artists has massively harmful implications that make the cure worse than the disease.

Definition, proposals, assertions

The idea of a learning right is this: in addition to the traditional reproduction right copyright reserves to the author, authors should be able to prevent people from training AI on their work by withholding the right.

This learning right would be parallel to other reservable rights, like reproduction: it could be denied outright, or licensed separately from both viewing and reproduction rights at the discretion of the rightsholder. Material could be published such that people were freely able to view it but not able to use it as part of a process that would eventually create new work, including training AI. The mechanical ability to train data is not severable from the ability to view it, but the legal right would be.

This is already being widely discussed in various forms, usually as a theory of legal interpretation or a proposal for new policy.

Asserting this right already exists

Typically, when the learning rights theory is seen in the wild it’s being pushed by copyright rightsholders who are asserting that the right to restrict others from training on their works already exists.

A prime example of this is the book publishing company Penguin Random House, which asserts that the right to train an AI from a work is already a right that they can reserve:

Penguin Random House Copyright Statement (Oct 2024) No part of this book may be used or reproduced in any manner for the purpose of training artificial intelligence technologies or systems. In accordance with Article 4(3) of the Digital Single Market Directive 2019/790, Penguin Random House expressly reserves this work from the text and data mining exception.

In the same story, the Society of Authors explicitly affirms the idea that AI training cannot be done without a license, especially if that right is explicitly claimed:

The imperfections of Murder Drones

  • Posted in fandom

I love murder drones. I think they’re such great little guys. Bring me a robot maid and I am yours forever, etc. But watching through the series itself actually took me a few stabs, and I think it’s due to a few design decisions that make following the plot unintuitive and add some friction to what’s otherwise a very fun show. So I want to talk a little bit about that friction, even though the entire thing is still a good time overall.

Indie Animation

First, the obviously relevant context is that Murder Drones is made by Glitch, which is a small independent animation studio. And independent animation necessarily comes with constraints. It’s incredibly exciting that we have the technology for small teams to make work with this quality and scale, and I don’t at all want to take that for granted. But I think a lot of the friction I have to talk about comes from fundamental trade-offs that come from that setup.

Since their resources are very limited and good animation is expensive work, there’s a pressure for everything to be compressed. Short episodes with short shots in an eight-episode miniseries mean the project is feasible, but it’s hard to get all your fun ideas in while still sufficiently paving the way for them to land properly.

get tunnel visioned on spooky corpse robot reveal, work backwards from there

Structurally, a small indie team also carries the risk of skill gaps. I don’t mean to make any criticisms of anyone in particular on the project here, but this kind of team might not necessarily have experienced television writers or producers. And, with a small independent team, there might not be enough of a test audience to catch things that could be improved, or not enough budget to re-iterate for minor improvements. So those are all categories of things that can easily run into trouble.

Independent serialized animation like this is a relatively new phenomenon, but these are going to be the same sorts of challenges projects like RWBY and Helluva Boss have. (Although I think Murder Drones is significantly better than both of those.) So while there are common environmental factors that can make this kind of project a little extra rough, the way that roughness actually manifests is interesting.

It’s not glaringly bad

The reason I’m interested in talking about this at all is that I noticed the friction as part of my own experience, but it wasn’t linked to any obvious problems. In fact, the whole reason I’m writing this is Murder Drones felt like it should be great, and I was surprised there were things that still weren’t quite clicking. In re-watching the series to write this, slowing down and zooming in to catch every piece made the effect much harder to see. It’s hard to put my finger on exactly what caused the effect. Which is why I want to! The dynamics you can barely see are always the most interesting to understand.

FSE sprite compression

  • Posted in dev

This was originally published 2020-07-07 as a reward for sponsors of Befriendus

A Domain-Specific Compression Algorithm — as I later found out this is called — is a compression algorithm that uses the specific nature of the target data as a way to efficiently compress it. The more you know about the structure of the data you’re compressing and what tools you have to reconstruct data, the more efficient the system can be.

I wrote a script for the Fansim Engine that does this with character sprites. It takes character poses, identifies the parts that have changed and the parts that stay the same, and creates identical Ren’py displayables that take up dramatically less room.

Making Thanos work

  • Posted in fandom

Did you know there are still people who think the MCU’s Thanos is a deep character with interesting motivations? For all the CinemaSins “why didn’t he use his powers to end scarcity, is he stupid” types, there are still “Thanos did nothing wrong” chuds.

This is stupid, of course. But after seeing people be wrong on the internet, it occurred to me recently that there are a couple of genuinely interesting ways to spin the character without changing his mechanical role in the story. In fact, with just a tiny bit of re-framing, you can turn Thanos from a stupid dumb-dumb into a genuinely great villain.

Why Thanos doesn’t work

First, a super-quick summary of what I’m reacting to.

Verification on Bluesky is already perfect

  • Posted in tech

Bluesky has very quickly become a serious social media platform. This means it’s having to deal with all the problems social media platforms have to deal with, including impersonation. A lot of people flocked to Bluesky from Twitter, and so recreating something like Twitter’s verification system seems like a natural step.

But you don’t need to do that! Bluesky’s current verification system is actually very good and does what verification is supposed to do.

In 2022 I wrote a retrospective essay about the “verified account” design pattern on Twitter, which tried to preempt this conversation a little bit, but unfortunately got bogged down a little with Elon breaking Twitter verification. This piece will talk about a lot of the same ideas, but applied more specifically to Bluesky’s ecosystem.

The ambiguous "use"

I keep seeing people make this error, especially in social media discourse. Somebody wants to “use” something. Except obviously, it’s not theirs, and so it’s absurd for them to make that demand, right?

Quick examples

I’m not trying to pick on this person at all: they’re not a twitter main character, they’re not expressing an unusual opinion here, they seem completely nice and cool. But I think this cartoon they drew does a good job of capturing this sort of argument-interaction, which I’ve seen a lot:

I’ve also seen the exact inverse of this: people getting upset at artists because once the work is “out there” anyone should be able to “use” it. (But I don’t have a cartoon of this.)

There is an extremely specific error being made in both cases here, and if you can learn to spot it, you can save yourself some grief. What misuse is being objected to? What are the rights to “certain things” being claimed?

The problem is that “use” is an extremely ambiguous word that can mean anything from “study” to “pirate” to “copy and resell”. It can also cover particularly sensitive cases, like creating pornography or editing it to make a political argument.

webcomicname: beliefs you do not agree with

But everything people do is “using” something. By itself, “use” is not a meaningful category or designation. Say you buy a song — listening to it, sampling it, sharing it, performing it, discussing it, and using it in a video are all “uses”, but the conversations about whether each is appropriate or not are extremely distinct. If you have an objection, it matters a lot what specific use you’re talking about.

But if you’re not specific, there are unlimited combinations of “uses” you could be talking about, and you could mean any of them. And when people respond, they could be responding to any of those interpretations. There’s no coherent argument in any sweeping statement about “use”; the only things being communicated are frustration and a team-sports-style siding with either “artists” or “consumers” (which is a terrible distinction to make!).

Formal logic

This is not a new problem. This is the Fallacy of Equivocation, which is a subcategory of Fallacies of Ambiguity. This is when a word (in this case, “use”) has more than one meaning, and an argument uses the word in such a way that the entire position and its validity hinge on which definition the reader assumes.

The example of this that always comes to my mind first is “respect”, because this one tumblr post from 2015 said it so well:

flyingpurplepizzaeater Sometimes people use “respect” to mean “treating someone like a person” and sometimes they use “respect” to mean “treating someone like an authority”

and sometimes people who are used to being treated like an authority say “if you won’t respect me I won’t respect you” and they mean “if you won’t treat me like an authority I won’t treat you like a person”

and they think they’re being fair but they aren’t, and it’s not okay.

See, here the “argument” relies on implying a false symmetry between two clauses that use the same word but with totally different meanings. And, in disambiguating the word, the problem becomes obvious.

Short-form social media really exacerbates the equivocation problem by encouraging people to be concise, which leads to accidental ambiguity. But social media also encourages people to take offense at someone else being wrong as the beginning of a “conversation”, which encourages people to use whatever definition of other people’s words makes them the wrongest.

Since I’m already aware that copyright is a special interest of mine, I try to avoid falling into the trap of modeling everything in terms of copyright by default, Boss Baby style. But this is literally the case of a debate over who has the “right” to various “uses” of things that are usually intangible ideas, so I think it’s unavoidably copyright time again.

Game patent grab bag

This was originally something I was going to talk about in Corporations have Rejected Copyright, back when that series was going to just be one long post (really!). But since I saw Nintendo apparently sued Palworld today, I wanted to put this up as background information.

You should definitely read You’ve Never Seen Copyright first, particularly the explanation of what patents are, because this conversation directly follows from that. The most important thing to pick up on is how the Doctrine of Equivalents lets companies use patents that are supposedly very specific to threaten other implementations that are similar, even if they aren’t using the patented design.

Game patents are revelatory, because game rules as a category explicitly do not fall within the realm of patent rights, but companies have managed to file and defend fraudulent patents anyway.

Is AI eating all the energy? Part 2/2

  • Posted in cyber

Part 2: Growth, Waste, and Externalities

The AI tools are efficient according to the numbers, but unfortunately that doesn’t mean there isn’t a power problem. If we look at the overall effects in terms of power usage (as most people do), there are some major problems. But if we’ve ruled out operational inefficiency as the reason, what’s left?

The energy problems aren’t coming from inefficient technology, they’re coming from inefficient economics. For the most part, the energy issues are caused by the AI “arms race” and how irresponsibly corporations are pushing their AI products on the market. Even with operational efficiency ruled out as a cause, AI is causing two killer energy problems: waste and externalities.

Is AI eating all the energy? Part 1/2

  • Posted in cyber

Recent tech trends have followed a pattern of being huge society-disrupting systems that people don’t actually want. Worse, it then turns out there’s some reason they’re not just useless, they’re actively harmful. While planned obsolescence means this applies to consumer products in general, the recent major tech fad hypes — cryptocurrency, “the metaverse”, artificial intelligence… — all seem to be comically expensive boondoggles that only really benefit the salesmen.

simpsons monorail screencap Monorail!

The most recent tech-fad-and-why-it’s-bad pairing seems to be AI and its energy use. This product-problem combo has hit the mainstream as an evocative illustration of waste, with headlines like Google AI Uses Enough Electricity In 1 Second To Charge 7 Electric Cars and ChatGPT requires 15 times more energy than a traditional web search.

It’s a narrative that’s very much in line with what a disillusioned tech consumer expects. There is a justified resentment boiling for big tech companies right now, and AI seems to slot in as another step in the wrong direction. The latest tech push isn’t just capital trying to control the world with a product people don’t want, it’s burning through the planet to do it.

But, when it comes to AI, is that actually the case?

What are the actual ramifications of the explosive growth of AI when it comes to power consumption? How much more expensive is it to run an AI model than to use the next-best method? Do we have the resources to switch to using AI on things we weren’t before, and is it responsible to use them for that? Is it worth it?

These are really worthwhile questions, and I don’t think the answers are as easy as “it’s enough like the last thing that we might as well hate it too.” There are proportional costs we have to weigh in order to make a well-grounded judgement, and after looking at them, I think the energy numbers are surprisingly good, compared to the discourse.

Copyright abusers lost their claim

or, the many people who said movies like Coyote v. Acme that were killed for a tax write-off should be forced into the public domain were right, and here’s why

A healthy system of creative rights, including a balanced form of copyright, is a reciprocal arrangement between creators, consumers, and the commons. Creators are granted some temporary exclusive rights by the government over qualifying intellectual work in order to incentivize creativity. These privileges are granted in exchange for creating valuable new information — the existence of which is a contribution to the public good — and for providing it in such a way that others will be able to build on it in the future. It’s an incentive for providing a specific social good, one which the market alone might not reward otherwise. Fortunately, this is actually how US copyright was designed; see You’ve Never Seen Copyright.

The takeaway from that, though, is not just that there’s a fair version of copyright, but that copyright must look like that fair model. The fact that such a thing as “good copyright” exists as a sound philosophy is not a broad defense of the word “copyright” itself, it’s an imperative requirement for the legitimacy of any system of power that claims to enforce copyright. The soundness of the philosophy doesn’t legitimate the system of power that shares its name, it damns it for failing its requirements.

When they invoke the philosophy of copyright to justify thuggery, it matters that they’re wrong.

The requirements for reciprocity intrinsic in copyright are how the system must work, but it’s not what actually happens today. In practice, corporations regularly violate the fundamental principles of creative rights — both in letter and in spirit — and use copyright protections to profit without showing the required reciprocity.

I can’t possibly list all the stories of what these violations look like. Seriously, just the thought of me having to give a representative sample of companies abusing IP law made me dread writing this series, it’s such a prolific problem. But I have shown a sample: Nintendo using copyright to kill new creative work, Apple using trademarks to keep competitors from conducting trade at all, book publishers trying to destroy the idea of buying and selling books… they’re all examples of how companies do everything they possibly can to get out of fulfilling their side of the bargain.

Case studies are fun, but just listing out a bunch of horrors isn’t what I set out to do; that’s just groundwork for thinking about the problem. What’s important is that they’re a representative sample of a kind of behavior. With all that established, you can read this with the knowledge that yes, they violate the purpose of the law as written and yes, violations are so regular they seem to define the practice.

So what does it all add up to?

Here’s what I say: If you want out of the deal, so be it. When someone won’t participate constructively — if they don’t work in good faith, or at least begrudgingly accept the limits the system of copyright puts on them — we stop respecting their claim to special privileges within it as legitimate, and understand it as the double-dealing overreach it is.

As self-evident as it sounds when I say it out loud, this argument is my nuclear option. This is what I would have to say if it ever got this bad; if, between the two of them, the courts and the corps ever broke the system beyond my last bit of tolerance. And I’ll be damned if they haven’t done just that.

Legitimacy

In You’ve Never Seen Copyright, I talked about how the word “copyright” can refer to two very different things: either a philosophical basis that justifies copyright as a legal doctrine, or the system of power that describes how copyright is actually enforced, what enforcement looks like, and who it benefits.

But the fact that the power structure has diverged from the original philosophical intent doesn’t just create a communication issue. Yes, it becomes increasingly unclear what people who say “copyright” are talking about, but the legitimacy of the power structure depends entirely on being an implementation of a sound legal doctrine.