Speaking of ethics...
Jürgen, Cory, and LLMs.
Before I say anything, I must confess: I may be in over my head on this.
I've never studied ethics formally, haven't read much Kant or Locke or anyone of the like. But even without rigorous academic training, I'm still on the hook for evaluating situations and conducting myself with some amount of ethical praxis because cogito ergo sum and I have a responsibility to other humans to treat them with dignity, even those who are a world away.
So anyway here's a Mastodon thread starring:
- Jürgen Geuter (@tante), writer of Smashing Frames, expert in computer science and philosophy.
- Cory Doctorow (@pluralistic), technologist, activist, writer, speaker with the Electronic Frontier Foundation, hardly in need of introduction.
- An ensemble cast from across federated social media.
To set the stage a bit, the root of the thread starts:
Yesterday Cory Doctorow argued that refusal to use LLMs was mere "neoliberal purity culture". I think his argument is a strawman, doesn't align with his own actions and delegitimizes important political actions we need to make in order to build a better cyberphysical world.
EDIT: Diskussions under this are fine, but I do not want this to turn into an ad hominem attack to Cory. Be fucking respectful
The particular argument Geuter is referencing comes from Doctorow's Pluralistic: Six Years of Pluralistic, the excerpt reading:
Doubtless some of you are affronted by my modest use of an LLM. You think that LLMs are “fruits of the poisoned tree” and must be eschewed because they are saturated with the sin of their origins. I think this is a very bad take, the kind of rathole that purity culture always ends up in.
The rest of the thread went about as expected.
Despite all of the miscommunication and accusations of rhetorical impropriety, there are some thoughts in there that have caused me to have a few of my own. Many of my own conclusions are much more deft in Geuter's Acting ethically in an imperfect world, which directly addresses Doctorow's perspective, but I'd like to relate how all of this laid waste to the ethical question of LLMs and perhaps reach the other side of this mess with actionables in hand.
Also that thread gives me heartburn. Hence writing here instead of there.
"Purity culture"
Perhaps the best way to start is on so-called "purity culture" and ethical consumption.
To quickly poke a couple dead horses:
- The people who espouse "purity culture" are garden variety jerks. Their finger-wagging does not necessarily compromise the values system they are finger-wagging about.
- Moral perfection is impossible in practice, but that does not mean we as people should give up on trying to be better given the circumstances. Too much Reddit karma has been spilled on the altar of "you claim to hate capitalism yet you have iPhone, curious" for it to be a serious line of dialogue.
Geuter has already pointed out, as have many much more intelligent people before any of us, that there is almost no way to live, much less consume, in a 100% ethical way.
The issue is that those who tend to employ verbiage like "purity culture" are almost always applying an impractical ethical standard to the person they are accusing. Put another way, an argument of the form such that Doctorow used to defend his usage of LLMs relies on a rhetorical sleight-of-hand that ultimately supplants his interlocutor's "ethical consumption" with his own "ethically-pure consumption", then continues a line of reasoning predicated on his interlocutor's newfound moral weakness, assuming they don't check him the moment he attempts the deception.
This reasoning is best exemplified in Doctorow's own words:
Let’s start with some context. If you don’t want to use technology that was created under immoral circumstances or that sprang from an immoral mind, then you are totally fucked. I mean, all the way down to the silicon chips in your device, which can never be fully disentangled from the odious, paranoid racist William Shockley, who won the Nobel Prize for co-inventing the silicon transistor.
Geuter remarks:
The strawman is his claim that people who criticize LLM usage are doing that for some form of absolutist reasons. That they have a fully binary view of the world as separated into “acceptable, pure things” and “garbage”. Which is of course false. Because they are using a computer, using warm water that’s probably heated through the use of fossil fuels etc.
He attacks a ridiculous made-up figure to deflect from specific criticism of LLM use (that many probably wouldn’t even apply that strongly to his use case).
Doctorow's particular angle, the usage of terms like "purity culture", is a bit unusual given his many years spent advocating for a better world. He addresses a potential criticism of LLM usage to wholesale shield himself from all criticism. It's a tactic employed by supermarket rags and is sorely out of place coming from Doctorow.
There are plenty of cogent arguments to not use LLMs, self-hosted or otherwise, crafted for daily use that don't require moral absolutism to parse or practice. On the flipside, it's unbecoming to construct a poor argument to maintain a feeling of ethical unassailability, and downright irresponsible when espoused on a platform as massive as Doctorow's.
His usage of LLMs is not all that important. He doesn't need to be perfect. He's human just like anyone else, capable of wack takes like the rest of us. If he uses LLMs and posts this admission explicitly for public consumption, then I'm entitled to an opinion of disappointment and he certainly shouldn't be written off, at least not for this. But it's his perceived need to preemptively strike at his supposed detractors that pushes this beyond a simple wack take. Perhaps he feels that this admission of LLM usage soils his credibility in some way and so he went elbows-up; the worry isn't unfounded given the general sentiment of his average reader. He just has so much further to fall.
Whatever his reason, the result is that he created a fake critic in his blog post, was responded to by markedly real person Geuter, then continued to argue without acknowledging the difference between real critics and the small, manageable one fashioned with his own hands.
It's embarrassing to say the least, and certainly not anything he can't fix or be forgiven of. But that grace must be earned given his conduct.
On LLMs
I'd like to establish the baseline perspective on LLMs before moving forward, there is quite a bit of signal noise when it comes to dialogue on this topic including and especially from software engineers.
LLMs are what they do
Later on, when we dive into the thread on Mastodon, we'll see a very common categorical error that tends to be the linchpin for a majority of misunderstandings. There is a difference between LLMs the technology and LLMs the product.
Critics of LLMs very rarely take umbrage with the technology; the foundational principles have been around for a long while and been applicable to a whole host of fuzzy problems. The specific flavor of artificial intelligence that is the LLM doesn't in itself have an ethical value, it's a mathematical construct, an abstracted method for generating a text output.
No, critics of LLMs and of artificial intelligence — as it pertains to the modern colloquial usage of the term — are resistant to how these technologies are implemented as of late and the effects of that implementation on people and the environment.
- The training datasets for any broadly-applicable LLM or generative system contain massive amounts of copyrighted material created by artists, musicians, and writers who remain to be paid fairly by the corporations who benefit.
- Part of training involves manually screening for explicit and harmful material, typically outsourced to impoverished populations for a pittance relative to the mental harm caused.
- The resources required to train LLMs are a stress on our public utilities and the environment, with taxpayers largely footing the bill for corporations' indiscriminate and wasteful usage of resources.
- The resources required to build the datacenters that allow LLMs to be trained are wresting control of general purpose computation away from the consumer and redirecting the globalized production of computer hardware into the hands of a few corporations.
- "Open-source" models that can be run on self-hosted hardware require all of the above to be true. More open-source models are being made all the time using the same exploitative and extractive methods as their closed-source counterparts.
These points are independent of any other arguments regarding the erosion of human creativity and learning, invasion of privacy due to nonconsensual integration of agentic systems, dire economic conditions caused by overhype, lack of value return, semantic ablation, lack of oversight on AI corporations, weakening of a shared objective reality, deepfake porn, and the beat goes on.
The key here is that none of this has anything to do with the math that makes all this possible. These problems aren't theoretical, these aren't in some fantasy future where a superintelligence exists, these are happening right now in the real world. It is all of this material harm that must be weighted against the potential benefits of LLMs, of which there is only one: convenience.
It's not radical nor "purity culture" to say that convenience is an insult of an excuse for their continued usage.
What do you do about them?
All that being said, there must exist a practical ethic to weigh our actions in walking life; this is not something philosophical quibbling over definitions can provide. Real life happens in context, not abstraction. We need to talk praxis, and so what is within the set of actions that can be taken given the circumstances?
Any action should be tempered by the fact that nothing I individually can do will change much. No matter how much I loathe LLMs (the product), how much I post about them, I can get my word count into the tens of thousands over them, none of that can stop them from proliferating and causing problems. Frankly, I don't have the kind of egotism to believe that my personal resistance will change the world.
At the same time, that doesn't mean that my personal resistance is in itself meaningless. Nobody serious scoffs when someone chooses a lab-grown diamond, a fair trade bag of coffee, or saves their money to support artists on Bandcamp Friday. If we are to be consumers — and we are all consumers — these choices are part of doing what we can given the current reality. While we should demand and fight for better options, we still need to make judgement calls on a set of less-than-ideal choices in our day-to-day.
As Geuter said, we have an obligation to try to make ethical choices in an imperfect world. Those with privilege may have the ability to choose while others will have to make moral compromises to put food on the table. I'm fortunate to work in a highly-paid industry, so I can choose to buy local, I can choose to buy used, I can choose to repair and mend, and I can choose to support causes that I believe are doing the good work.
In the same way I can choose, I can also choose not. If given a set of less-than-ideal choices and I'm not situationally forced to choose, then I can opt out of the decision altogether. LLMs (the product) are a convenience, not a necessity. I know what they are, what they do, the systems required to support them, and because I know I now must act. Me not using agentic systems will not topple OpenAI no more than choosing not to smoke will bring about Newport's demise. But that doesn't change the fact that smoking is bad for me and bad for the people around me.
With all that in mind, I keep creating anyway, putting in the work and proving that 1 line of human-intentioned code is worth 100 from the machine. I keep honing my skills, I stay curious, I keep reading, I keep writing and rewriting, and encourage the people around me to do the same. The convenience of LLMs (the product) is tempting, but I'd rather hold myself to a higher standard.
If I must, I make my stance clear in conversation, but I'm not going to act so righteous as to browbeat anyone else (too hard) for using LLMs (the product). They choose to engage me on this topic at their own peril. They know my mind; and if they don't, they'll learn.
The individual prerogative is simple: choose not.
But there is also a more important imperative: get involved. Go to town halls, vote, protest, give time and, if you can, money to organizations keeping up the fight. In any way you can, be part of the political process that determines if these systems show up in your backyard. And if none of these actions are within reach, then encourage others to take up the mantle. These are the kinds of actions that bind humans together in common belief and build communities. If there were any merit to the accusation of "purity culture" it would exist insofar as words are devoid of action, individual and collective.
Assorted takes
Now for the lightning round.
All of humanity has access to it
Original post for posterity.
LLMs are based on extraction, exploitation and subjugation
So is torrenting. This is a very capitalist argument, coming from someone that self-identifies as a communist, that one deserves to reap the rewards of them adding value to humanity through some form of gatekeeping and is entitled to a reward from such gatekeeping. You're literally arguing on the side of Elsevier and JSTOR against Aaron Swartz
What does it matter if human knowledge is available as a book or an LLM? The important part is that all of humanity has access to it.
This appears to be a variation of an anti-copyright argument, that people being paid for their work is unethical because copyright and capitalism broadly are unethical systems. The issues, at least in this context, are threefold:
- People who make creative work for a living need to eat today and a potential system in the future cannot provide them food in time. Copyright and capitalism are the best any individual can do at the moment given the circumstances; yes, these need to be mended into or outright replaced with more just systems, but that's beyond the scope of this particular conversation about individual LLM usage.
- By saying "so is torrenting", there is no defense of LLMs here. The point is ceded that LLMs are based on extraction, exploitation, and subjugation. Everything else said can be true, but that doesn't address the extraction, exploitation, and subjugation. The plot is lost from the first sentence.
- All of humanity does not have access to LLMs. Only the privileged with computers and sufficient Internet access have access to LLMs. Everyone else foots the bill either way.
Someone who just understands the arguments
Original post for posterity and its parent post posing the questions which this post answers. Rather than a straight quote of each, I'll interpolate them Q&A-style for readability.
@tante@tldr.nettime.org questions for the leftists and liberals from a confused anarchist:
Do you think you can put the cat back in the bag with LLMs? How?
@komali_2 answering as a leftist AI-moderate who just understands the arguments better than most:
It varies, they either haven't thought that far or they have such a flawed view of AI that they think it's possible (ie. if it can only exist on massive datacenters, then shutting down those datacenters would stop it, and if it was only a novelty then a shaming campaign would shut it down)
This is the categorical error of technology-vs-product. The technology is abstract, the product is material. Nobody actually cares that the LLM technology exists, the ethics lie solely in its implementation. The framing of the question and of the answer are rooted in the idea that detractors just want to "win" against AI; both the question and the answer are deeply unserious.
For those that believe that LLMs were trained on stolen data, what does it mean for data to be private, scarce property, that can be "stolen?"
That's a a very hot button topic. People have taken copyright for granted to the point that the discussion about the ethics of copyright is just as offensive to people as the topic of AI. Even though this case isn't even covered by copyright law, they feel entitled to "ownership" of their art. The argument of "theft" didn't even kick off until AI got good enough that people started using it in lieu of paying artists, and is mostly inbetween a demand for payment (not realizing that any payment would be laughably tiny) or a way to try and pressure these into not existing (see question 1).
For the question, "stolen" is not a philosophical concept, but a legal one with a clear definition in most jurisdictions as it pertains to creative work. In the United States at least, this comes with copyright and fair use implications, and there's ongoing litigation to determine if companies like Meta are in breach of one or more laws. Anthropic, as linked earlier, actually got the smackdown. The only way the question is some kind of gotcha is if "stolen" and "property" are relegated to philosophy rather than having specific, contextual, real-world meaning and have the capacity to really hurt people.
As for the answer, see the previous section, we've been there and done that.
What about models that just steal from the big boys, like the PRC ones? Theft from capitalists, surely ethical?
Most of these people know nothing about variations in training data, and they'll either dismiss this question (because they're siding with said capitalists) or they'll divert to arguing about the ecological impact of training the big models.
The ethics of theft are not within a hundred miles of the topic at hand. The bullying-the-bullies angle has nothing to do with LLMs, and "variations in training data" has nothing to do with the bullying-the-bullies angle. Of course, diverting to the ecological impact of training risks bringing the conversation back down to earth and addressing material harms that are really happening right now rather than theorycrafting on the ethics of theft.
Will you not using any LLMs cause Sam Altman and friends to lose control of your country?
They're just trying to social pressure everyone into no longer using them, because again see question 1 where they think they can put the cat back in the bag.
The answer is no, and no one who takes this topic any degree seriously claims otherwise. There is no direct line of causality between one person not using LLMs and Sam Altman on his knees wretched from all of his apes being gone. But that line does exist when we start from activism, protest, legislation, and coordinated rejection of his machine and others of its stripe. These are all social activities, not individual ones.
More than anything, this line of thinking is indicative of a certain antisocial nihilism, that the only actions that exist are individual ones and individual actions can't change the world. Therefore nothing can be done. Except that things can and are being done, not at the individual level, but at the collective level. If anyone is to move a mountain, it'll take a lot of people pushing together, and the antisocialites as always will be pushed along for the ride all the while lamenting that — from their perspective and against observable reality — the mountain isn't moving. It's pretty irritating, but that's life I guess. The good work continues with or without their help.
There is a certain myopia in the supposed ends for LLM detractors, to "social pressure everyone", as if there exists no legitimate criticism for these systems, that the only reason people are against their proliferation is to simply have the moral high ground.
For the most part, this line of questioning is at best indifferent to the real world harms of agentic systems and at worst approving.
Cory's LLM usage is not that interesting
Original post, maybe this one will go to posterity.
@tante@tldr.nettime.org Frankly, I don't think there are any ethical concerns with how he's using it.
The reason AI is a violation when it trains on openly available data and then outputs similar stuff is that it's creating derivative works. Something that reads everything produced by man and then uses that information to score similar output does NOT. It's completely fair use and it's a GOOD application of AI.
IFF all the evil crap that the people who made it are up to wasn't a concern there'd be none.
AI is a violation for a plethora of other reasons, some would even say there are reasons more important than how the training data is sourced. It has nothing to do with who is at the helm of these corporations and everything to do with what is being done in service of these systems, the human and environmental cost of it all. Again, open-source models are created using the same flawed, exploitative processes as their closed-source counterparts often by the same owners. There is no truly "open-source" model.
But ignoring the problem of origin for a moment, Doctorow's usage of LLMs was supposed to be a simple and largely unremarkable disappointment. Humans are imperfect, there are no perfect heroes. It was his kneejerk response to a nonexistent critic that was the issue, dismissing any criticism of LLMs — his usage of LLMs — as "purity culture". He could not and cannot conceive that someone criticizing his usage could have any degree of merit.
It was an abuse of his notoriety to undermine genuine political discussion around these systems by bravely inserting himself into a conversation that was always much bigger than him. Confined to his blog, it's wack. Spilling over into multiple hours of replying on Mastodon, defending his own credibility with shoddy high-school debate club tactics is utterly embarrassing and further confuses a public conversation rife with misdirection and intentional obfuscation.
From the man himself
Original post. Posterity.
@tante That doesn't seem to be the best idea @pluralistic
AI and LLM output is 90% bullshit, and most people don't have the time nor the patience to work out which 10% might actually be useful.
That's completely ignoring the environmental and human impacts of the AI bubble.
Try buying DDR memory, a GPU or an SSD / HDD at the moment.
@simonzerafa @tante What is the incremental environmental damage created by running an existing LLM locally on your own laptop?
"Incremental" was not among the words used by his interlocutor, and "human impact" was nowhere to be seen in his response. No acknowledgement that these models are the product of people in the global south combing through mountains of abusive and disturbing text and image data so that Cory can have a nondeterministic spell checker.
In closing
Ultimately, there are real ethical questions around LLMs and much more discussion to be had regarding what's to be done about them, where personal responsibility starts and stops, and what we expect out of a political solution.
There is also a much greater responsibility on those with a platform like Doctorow's to wield it with dignity. It was his decision to clap back at his nonexistent detractors with his specifically-crafted criticism on his own blog. Geuter responded to that point with a cogent rebuttal on his blog, and then Doctorow proceeded to spend the rest of the day farming gotchas in the federated trenches, LLM-truthers bravely by his side.
All that really happened in the end is that the conversation about the ethics of LLMs and modern AI systems was further obfuscated, confused by one of the most influential voices on the federated Internet.
As Geuter wrote, not everyone wants to settle for the utilitarian answer: that these systems are already here, might as well use them. Some people strive for better, speak of better, show up in person, build community, and advocate for a higher standard even if they themselves are imperfect. None of these constitute "purity culture".
In my estimation, there is no ethical usage of LLMs, so I don't use LLMs. I'd encourage everyone not use LLMs. I don't care if you use LLMs, but if you want to talk about LLMs, you've been warned.
Doctorow is entitled to a wack take just as anyone who gazed upon this spectacle is entitled to disappointment, his work still moves mountains and that counts for so much more. But I also remain sober to the fact that he's a person, not a mythology; perhaps this lapse will recalibrate his readers back into healthy skepticism as we all should when it comes to figureheads. For the moment, his general activism might be a step in the right direction, even if not in the perfect direction, depends on who you ask. If his current objectives find success, then it'll remain to be seen if people's hunger for an even better world passes him by. We'll see.
I personally don't care that Doctorow uses LLMs. I only care that others would find him upon this small hill and be discouraged from acting on a higher moral standard.
