This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.
Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.
We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:
-
Shaming.
-
Attempting to 'build consensus' or enforce ideological conformity.
-
Making sweeping generalizations to vilify a group you dislike.
-
Recruiting for a cause.
-
Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.
In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:
-
Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.
-
Be as precise and charitable as you can. Don't paraphrase unflatteringly.
-
Don't imply that someone said something they did not say, even if you think it follows from what they said.
-
Write like everyone is reading and you want them to be included in the discussion.
On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.
Jump in the discussion.
No email address required.
Notes -
There is a great new Substack article about the populist backlash to AI data centers by Jasmine Sun. She also did an interview with Ezra Klein for those of you who like audio content.
I like the article because Sun speaks fluent Silicon Valleyish while also taking the time and effort to visit and talk to the people on the ground in the rural Midwest instead of condecending at them from the coasts. She asks an important question: why precisely do people hate data centers so much?
Looking only at the statistics, AI land use and water use issues are fake. Nowhere is running out of land, and nowhere (West of San Antonio at least) is running out of water. Are the people in these communities stupid? Do they not know how to shut up and multiply?
It turns out that a lot of the fear and anger stems from a combination of distrust and PTSD from deindustrialization. An interesting example is Janesville, Wisconsin, a town with a giant concrete slab in the middle of it where the GM plant used to be.
Fascinating. The collapse of the auto industry in 2008 was apparently so bad that it is better to prevent industrial development alltogether in order to avoid the risk of another collapse. It almost makes sense until you think about it for 30 seconds.
Some of the objections Sun hears in other towns fare a bit better.
This is the steelman of the anti-data center position. People are worried that the big companies proposing data centers are going to Foxconn them and not deliver the supposed benefits.
One of the things that comes across especially clear in the Ezra Klein interview is just how unprepared the AI companies were for the backlash. They didn't realize how sketchy it comes across to insist on NDAs with local officials when negotiating deals. They thought, as a straightforward application of Coase's Theorem would tell you, that any bumps in the road could be ironed out by handing over a big bag of money. It turns out that people don't work that way. It feels like a bribe, and Americans do not like being bribed.
I haven't read the Substack article yet, but I did read the Klein interview, and what stood out to me was this:
A big part of the dissonance, it seems to me, is around the perception of what 'AI' is and whether it's valuable. Clearly a lot of people in Silicon Valley believe that this is a massively transformative technology that will be vital to the future of the country if not the world. They're frequently either outright singularitans, or simply believe that it willl turbo-charge the economy, massively increase wealth, change everything about employment, and so on.
But these people, and I admit I think they sound a lot more sensible than the tech people online, actually use the technology and go, "meh". They see it as a mildly amusing toy. Maybe it has some uses, in some niche fields, but for the most part it's just not very interesting. If they didn't get access to it, they would not particularly care.
An auto factory - you at least understand why cars are useful, and why the factories need to be built. A data centre is a huge investment for a thing that... isn't useful. All of this is in aid of a technology that's just fundamentally pretty crap. No one wants to make sacrifices for the infrastructure to produce more LLMs because no one thinks that LLMs are worth sacrificing for.
There is the marketing issue as well, of course. Pro-'AI' people tend to sound very cult-y online, at least in my experience. Many of them sound like lunatics. A lot of them also, well, to be blunt, are un-charismatic nerds. Sam Altman, Dario Amodei, Elon Musk, Mark Zuckerberg, etc., are not convincing ambassadors for a technology. I can understand nobody believing their promises - none of them project credibility, or even really respectability. The 'tech billionaire' is a cultural archetype at this point, and it is an archetype that one associates with immaturity, frivolity, narcissism, and callous disregard for other people and for human value in general. Last week I was talking about the book destruction story and it fits with the image again, as if there's this small cult-like group of anti-human tech-heads eager to sell the whole inheritance of the human community for digital pottage. I understand, in my gut, the feeling of disgust.
Except the people spending all this money on them? Dave wants his book proofread more cheaply, Pete wants to ERP, Melanie wants dashboards to show key business insights, John wants some spreadsheets analyzed, Dean wants to fix bugs in his software. Even Johnny Jihadi wants to know how to make use of unexploded munitions left by govt forces in the Sahel and he goes to AI for that.
Why would the opinions of people who don't pay for the product because they don't find it useful outweigh those who do?
How does that address the point? People advocating against data centres do so on the basis that constructing data centres affects the communities they live in. That's why they get a say. People care about things that affect them. People who pay for chatbots don't get some sort of veto over people who don't. That would be lunatic.
That was the point of Sun's comparison to cars. You may grumble at an auto factory, but you do, at least, understand why cars are necessary. Even if you personally don't own a car and rely on public transport to get everywhere, it's obvious that there's real public demand for cars, on the basis of their obvious usefulness. You might not want a factory near you, but you understand why there need to be factories somewhere. LLMs have not cleared that hurdle. It seems that large numbers of people do not see why these things are so necessary that all these data centres should be built. We could just, you know, not have LLMs. That's an option.
The examples you give - proofreading a book, cybersex, reading spreadsheets - are generally all in the 'toy' category. Even the serious one, the hypothetical Jihadist, could just as easily google the munitions. The use cases for LLMs are so marginal that there is no publicly-felt sense of why they are needed, certainly to the extent of justifying the demands that data centre construction make on the public.
I don't think you're completely wrong here, but in this case the comparison to make if you think AI will be extremely useful would be if the anti-car urbanists started showing up to protest your car factory. Sure, you might not want it in your neighborhood, but you buy a new (to you, at least) car every so often and don't want them to become scarce resources, and people show up to protest every single car factory on the basis of "induced demand" traffic claims, "just walk, bruh", accident statistics, and a demand that you commute by never-to-be-finished passenger rail instead. They are convinced cars have no real use, and it'd be rather frustrating to normies if they doubled car prices without delivering any of the promised better alternatives first.
How to tell which framing is more relevant seems difficult from inside the situation.
I suppose the point of the comparison is to disambiguate two scenarios.
In the first scenario, people generally concede that the unwanted facility needs to be built somewhere. You don't want an auto factory or solar farm near you, but you concede that it needs to go somewhere, because we need cars and we need energy generation.
In the second scenario, people don't want the unwanted facility built at all. You do not want it anywhere, because you do not think the product or service that it enables needs to be made full stop.
The people that Sun interviewed, at least, are in the second scenario, and that changes the kind of conversation you would need to have with them, at least if you're pro-'AI'.
More options
Context Copy link
More options
Context Copy link
No offense, but this is an incredible statement of ignorance. I use them at work and they are amazing. Not in any way a toy or marginal in value.
You get it. Same thing for LLMs. Most Americans regularly use LLMs. A vast supermajority of younger adults do. There's such real and obvious demand.
What do you do, out of curiosity?
Are you using it mostly for coding, then?
To the latter point - I hope you can understand why many people are skeptical of arguments that go, "oh, of course the samples of our product that you can try are crap, you have to pay us, then you'll realise why it's worthwhile". Well, no. It's the company's job to sell the product to me, not vice versa. There's also a bit of a psychological trick there as well, in that once you pay money for something, you have a bias toward convincing yourself that it was worthwhile. Lastly, I am skeptical because people who seem to have good judgement who do pay for it also seem to report that it's crap. All the patterns of conversation look the same. I've read transcripts of LLM conversations on advanced models and... I'm sorry, it's all the same old beige. There is no evidence available to me that they're worthwhile, and "oh, you have to pay for it in order to understand why it's worth paying for" is not a reasonable argument.
For what it's worth, I work in a very different area to you. I work in aged care, and I cannot think of a use for LLMs here. The main way they have changed my work is just by requiring me to sort through and filter out a lot more slop when I try to find a picture, piece of music, or video to show to residents, so the main way they have featured in my workplace is as a mild annoyance. What on Earth could I use generative 'AI' for?
As another data point, I currently use a $100/month Claude subscription, duck.ai free from the browser, and midjourney.
Claude was paid for and set up from work, and it is very, very good. I am an artist, and it is much better at coding than me, and is able to implement code solutions for the technical problems I run into integrating my art into my company's project. It is tremendously efficient and effective so far with all the tasks I've thrown at it.
Midjourney, I paid a hundred bucks for a year's subscription. I used it very heavily on a project, and got some excellent results out of it, but the project petered out and trying to apply it to other projects has not worked well so far. Mostly I've dropped it for the time being to work on other things.
Duck.AI I've found is pretty good at solving code problems, but more than that is very good at assisting in finding solutions to technical problems with complex art software like blender. It is much, much better than the old solution of googling questions, even before google search enshittified, because it generates topical summaries aimed specifically at the question presented rather than websites which contain keyword matches somewhere within them. I've been making really good progress in areas that were too frustrating to pursue previously, and it's frankly pretty great.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
LLMs are obviously useful in the same way that businesses obviously need to analyse spreadsheets and obviously need to write code, amongst a myriad of other tasks where LLMs are useful.
The point you were putting forth is that because LLMs aren't very useful (except for some niche fields) and aren't interesting, there is no need to build datacentres. But that's not the case. We can discern that's not the case because people are prepared to pay for these LLMs. The general public. Coders. Companies. Militaries. Governments.
It is abundantly clear that LLM technology is in fact useful and there is an abundance of examples to back this up. Use of LLMs for targeting/ISR in the Iran war, use of LLMs in hacking, use of LLMs in writing code, use of LLMs in mathematics research, chemistry and biology, use of LLMs in entertainment, making images, translation, proofreading, examining legal or financial data. These are not niche fields.
So people are prepared to sacrifice money for these things, which is why these companies are building datacentres. They're not being built without reason. Companies are paying for all this compute and electricity. Sure, the electricity use is an issue for local communities. Industry has been known to require power, make noise and pollution, which should all ideally be minimized.
Again, this is not the case. If googling was just as easy then he'd do that and not waste time jailbreaking AIs instead.
It does seem that many people don't see a use for AI. But they're wrong about that, maybe they haven't looked into it or misunderstand what is going on.
For most people current AI is a little bit useful and a lot ominous, whereas the car, say, is a lot useful and only a little bit ominous. For coders this calculus may be different. You point out that it's also useful in law, finance, proofreading and entertainment, and these are indeed not niche fields, but I don't think there are all that many genuinely advantageous use cases of AI in these fields. In most cases one just gets increased convenience with evident downsides as well.
People may well have misunderstood how well AI will serve them personally in the future, when it's even smarter, but that's something reasonable people can disagree about.
Just speaking from experience, the increased convenience is a really big thing. I'm having trouble with my employer in a way that's very time-sensitive, and being able to check their work while my lawyer was away on summer holiday was a life saver. I cannot stress enough how much it helped me.
More options
Context Copy link
More options
Context Copy link
I'm not actually sure they are widely useful. So far the primary use case for them is coding, right? But not very many people are coders, it's a niche field, and as Freddie has argued (disclaimer: I do not wholly agree with him), very little has actually come of this supposed improvement to coding. We haven't seen a flowering of new programs, and neither have existing programs gotten better.
But even if we grant that LLMs are very useful in the handful of niche fields where they are currently used, the problems remain that 1) most members of the public do not understand or care about those uses, and 2) most members of the public do not think those uses are worth the costs they are being asked to bear.
Also, don't forget that there are plenty of programmers who do not find LLMs useful for programming either. To be fair I will say that there are those who do, but anyone who tells you that LLMs have revolutionized the industry is not telling you the truth. The truth is that the level of benefit they bring is controversial and not settled, even among programmers.
More options
Context Copy link
So besides coding, mathematics, physics, chemistry, biology, images, translation, proofreading, analysing contracts, tech support, playing Minecraft with, ERP, military targeting analysis, cybersecurity (defensive/offensive), drafting emails, summarizing documents, explaining history/science, doing everything that a search engine can do and more, writing and analysing spreadsheet functions, making charts and dashboards, sales research, writing fanfiction, earnings call analysis, document dump analysis, home repair/DIY fixes, recipes for cooking, car diagnostics from symptoms, real-time captioning for the deaf and literally lifesaving medical diagnoses...
Besides this handful of niche fields, what have LLMs ever done for us?
Doesn't it surprise you that Google is planning to spend $200 billion on AI this year? Mysteriously they don't seem to think it's a niche use-case but vital to their whole company's future since their core business of 'search' is another word for AI. And LLMs are at the heart of AI.
When I have issues with a piece of software not working and spitting out some incomprehensible error message I can now ask an AI and get a much better answer compared to trawling through forum posts from 8 years ago. Or get it to directly analyse and fix the code. Most programs have gotten much better in useability.
There are strong arguments against AI but 'it's mostly useless' is extremely silly. If it were useless nobody would care much about it, mostly useless is Siri-tier AI in the 2010s. Nobody cared much about Siri.
Furthermore, the general public's opinions don't really matter much. Concentrated, wealthy lobby groups like big tech and the military-industrial base have much more of a say.
That's a long list you've got there, but it includes things that LLMs have not made any useful contributions to (e.g. teaching, physics, chemistry), things that are obviously frivolous (e.g. cybersex, playing video games), things that LLMs are fundamentally untrustworthy with (e.g. legal contracts, most analyses), and things that LLMs are just plain terrible at (e.g. fiction, image generation), so... yes, besides those things. I think that was a long list of nonsense intended to overawe, rather than a serious discussion of what these bots are able to do.
I think there are some use cases for LLMs, but overall much less than their boosters think. When the general public rejects this technology as not useful for them, the general public is mostly correct.
This is exactly the problem. What we are talking about is public opposition to data centres, which, as Sun discusses, is predicated at least in part on the judgement that bots are not particularly useful. If 'AI' companies want to earn the public trust, they need to demonstrate both their trustworthiness and the usefulness of their products, and thus far they are failing at both those tasks. Sneering at the peasants is certainly not going to help them in these tasks.
Fiction is middling at best but LLMs (or adjacent technologies) are now pretty great at image generation. perhaps a bit narrow but the six-fingered hand is a thing of the past, and text is more or less solved on top models.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
Partly it's happening, just slower than you think and concentrated on business-to-business because these days building software for users has very high expectations for ease of use and frankly it's not worth it.
But the main issue is regulation and incumbents sitting on top of the bottlenecks. Want an app on your phone to point out you've missed an email that's urgent? Tough, only Apple gets to see your emails. Want it to be able to point out when you're doomscrolling? Tough, Apple doesn't let apps see what you're doing. Want your phone to take care of some trivial tasks without you? Tough, Apple definitely doesn't allow apps to use other apps. There's good security reasons for this but it means massive swathes of AI use cases can literally only be looked at by Apple and Google, who are slow-moving and mostly uninterested because they already own the market.
And this extends all over the place. Want to help people with something medical? I hope you've got a twenty person compliance team and you're friends with the regulators, and that's before you have a clue if your putative product will be popular or profitable. Similar for legal, similar for finance.
In short, the number of people who can do interesting things that you @OliveTapenade might notice in your personal life are limited to about 1-10 companies, who mostly have other priorities.
I also think this is too narrow. I've had a dispute with my employer and having something that can not only go through thirty pages of legal documents immediately and spot traps but also provide a legal fix so I don't have to rely purely on their drafting has been incredibly useful.
This is always the case with technology: it can take a surprising amount of time to replace existing infrastructure even when new technology is clearly better. Existing equipment was expensive and still works, teaching staff new workflows has overhead, and the impetus to actually change things is often low. OPM was still processing paper records for employee retirement until very recently. How many industrial machines are still running DOS out there?
I can imagine AI being both transformative and seemingly-ubiquitous, but still not reaching all the dark corners of the workforce for a generation or two.
More options
Context Copy link
More options
Context Copy link
This may still be overly optimistic. I think many people's main impression of modern software is negative experiences like ubiquitous data harvesting, addictive social media algorithms, and frustrating experiences with confusing or buggy new technology.
Promising to deliver more of these faster won't be considered a positive effect of AI.
This is kind of where I'm at. Sure, my phone has some cool new toys that let me edit my photos and identify bird calls, but by and large my experience with tech is significantly worse now than it was 10 or 20 years ago. It seems like most recent advances have mostly just ennabled increasing enshittification by making it easier to track me and engage in variable pricing. About 10 years ago, things stopped getting better and just started getting more expensive.
There is real value in the ~10% of my time it saves in researching and drafting emails. But I'd trade that 10% efficiency gain for no more AI ads/scams, no flock cameras, less bot spam, no more people using screenshots of AI hallucinations as evidence for stupid arguments, etc. in a heartbeat. I'm not even convinced AI is worth the price of free, let alone using my tax dollars to build a windowless concrete box in the middle of town.
More options
Context Copy link
As far as I can tell, most Americans think that AI is making society worse. More and more Americans have encountered and used the technology, but it is the most widely-despised technological advance in a generation.
For this story I think the key part there is Pew's point 10. Experts are vastly more optimistic about LLMs than the population in general. That divergence is what drives this conflict. If many members of the public believe that the situation is that a small, wealthy elite are forcing a technology on the country that the country does not want... well, those members of the public are correct.
I mean, good? I do not want members of the public to have veto power over what smart engineers are permitted to develop (as long as they're paying for it, of course). Imagine if we'd held a plebiscite on whether universities should continue spending money on this "Internet" thing in friggin' 1973, 4 years after ARPANET began. Imagine if journalists were publishing articles in 1907 asking for the government to shut down research on these new-fangled planes because they're flimsy and dangerous and nobody they know has seen any benefits from them.
Experts have, well, expertise in the subject. They know the technology has a ridiculous number of use cases. But it takes time for it to percolate through society.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
If software is a niche field, then the US is a niche economy. And, although the public might think math and software is pointless stuff for nerds, they do care about cratering the economy.
Do the majority of software developers do math? I don’t do software so I don’t know, but my guess is software developers doing math stuff was a niche.
Obviously, 50 years ago every software engineer did math.
I don't know how you can do software development without doing math. Maybe they aren't doing the kind of math that people usually recognize as math, because we never bother to teach things like algorithms or automata as "math", but they are very much doing math.
More options
Context Copy link
No, not typically. But it's one of "the handful of niche fields" where AI has proven very powerful and successful.
More options
Context Copy link
More options
Context Copy link
There are 1.84 million software developers in the U.S., so roughly half a percent of the population. I suspect most of these are heavily concentrated on the coasts, rather than scattered throughout the areas of the country most of these data centers are going.
Joe Six-Pack probably knows umpteen different welders, heavy equipment operators, machinists, and he certainly knows plenty of people who drive cars. He's probably never met a software developer, nor does he see any reason to. In his mind (and many others), AI might as well be fictitious.
Joe Sixpack may not know any software developers, but he uses the Internet, including some kind of search engine... which probably includes AI prominently now. If he doesn't live near Detroit or one of the few other centers of car manufacture, he probably doesn't know any automotive engineers either, but he damn sure knows about automobiles.
More options
Context Copy link
There are only 350k machinists, I propose a moratorium on building tooling for this niche field of which I personally know no practitioners.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
I think more than that there is also the case that people understand the value that a factory provides to the local community while data centers are just a blight locally and producing value elsewhere, perhaps not even in your state, nevermind your town.
So you might then say that the companies should compensate the locals for this. Well, for one, they're generally not proposing to do that (remotely commensurate with the costs they impose) and secondly people don't believe them even when they are proposing that. Companies a well known to lie outrageously about things like data center employment even before the AI boom.
Sun did talk about the companies trying to figure out how much they need to 'bribe' people. I'm not sure the deficit is just money. It seems like, in her argument at least, it's a combination of both trust and public utility.
I agree, but I wanted to expand on why the deficit involves both trust and public utility. Data centres disrupt the usual tacit bargain between a community and a major industrial project: the community accepts some local costs because the project also creates visible and durable local benefits.
With data centres, the costs are highly concentrated, land, water and electricity use, infrastructure demands, noise and visual blight, while most of the benefits are diffuse and accrue to users, employees and shareholders somewhere else. They generally provide very little permanent local employment, and the companies involved have a history of exaggerating those benefits and stiffing the locals. That makes promises of compensation harder to trust, not just in some general sense but for data centers specifically.
A factory can also impose serious costs, of course, but once operating it normally has an obvious continuing relationship with its community through employment, suppliers and associated economic activity. A data centre is less a reciprocal relationship and more like a company extracting locally scarce resources to produce value elsewhere. So even if people accepted that AI had some public utility, there would still be a separate question of why particular communities should bear a disproportionate share of its costs and why they should believe the promises of the relevant companies that have a history of lies.
I don't think the people involved quite understand scale of the costs they impose and the lack of benefits they give relative to other regular businesses or even doing nothing at all. The issues with data centers aren't something new, recognition of them and opposition too them have been growing for a long time, this rapid expansion just highlights it and makes it a news story.
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link
More options
Context Copy link