Clojurians Log v2

I think the users ARE in fact paying for it. Higher electricity prices, and ultimately the non price sensitive bag holder that is their pensions and 401ks.

Not sure what the environmental impact is of a depression/financial crisis.

I was listening to a VC podcast recently, one of the VCs was pointing out that the government can't let the bubble pop precisely because it will tank everyone's 401ks and that will upset people more than all the AI slop they are getting.

A library around here put down signs during the heat wave that their escalator was inoperable. With the rising demand of energy (and the increased use of not-always-available renewables) the energy grid gets overloaded here. So energy companies are more strict on over consumption and the library had to choose between airco and escalators.

Thats a pretty non-consequential thing right; but increased energy demand is beginning to show in regular society

"Upset people boomers"

last time i questioned a CEO about this their response was "AI will solve climate change", that honestly sent me to despair as the estimated impact was pretty considerable for the 6mo operation at the time. I cant help but to feel that its like graphene and fusion, just a excuse to keep polluting in the meanwhile because "X will solve the climate issue"

I'm curious how people square this with the massive use of cloud services that preceded AI and already led to massive numbers of data centers everywhere? I was looking at a map of data centers in the USA and the number of existing pre-AI data centers currently dwarfs the number built and planned for AI specifically...

The scale is increasing: Microsoft recently announced that one of its new datacenters in the Netherlands consumes 1% of the total energy consumption of the Netherlands

I'm curious how people square this with the massive use of cloud services that preceded AI and already led to massive numbers of data centers everywhere?

@lennart.buit good point about increasing scale. Thank you.

I think the same arguments against AI apply to the cloud services industry.

Are "AI data centers" and "vanilla data centers" the same? I'm a layman in this regard, but I can imagine how the former is much worse due to a much greater demand of pretty much everything. The "bad per square meter" could be much higher. Too lazy to look it up though, because it's always a rabbit hole where 5 out of 10 articles are off by an order of magnitude.

AFAIK their consumption and emissions are not really comparable 1:1, but one thing i know is that AI data centers are using in-loco oil/gas generator, so their emissions are very significant for the local communities https://www.youtube.com/watch?v=5p426fSlYH4

A quick search seems to indicate the heavy reliance on GPUs raises the power consumption substantially and the cooling systems are different (air for regular data centers, liquid for AI).

The problem is that whether a datacenter is bad really depends. A small datacenter can be a problem, a big datacenter can not be a problem. It depends on where it is at, and what strain the DS puts on the local resources

Light and noise pollution are also a real concern.

and most of that for trivial uses or for accelerating already fast processes...

I don't think they are the same. AI requires much higher power consumption per rack. I think like 50 times more.

So I think it's the scale of things are different, and the projection is that by 2030 data-centers will consume 4 times the amount of power they do today, all due to AI growth.

And so it's often about ROI, I'd say most people find the current cloud provided services provide substantial value and good that it warrants their environmental costs. I suspect a lot of people would not agree to the same for AI.

I know a lot of people feel like they're spending all this energy and water on stupid queries or rewrites that they could easily have done themselves, or it's all wasted on slop that they also are annoyed with having to read through or look at, or that it'll even come at the cost of human workers so you lose your worth and the environment suffers at the same time, etc.

> about ROI, I believe this ROI is also distorted because the externalities aren't currently being included in the costs/investment.

How much water does it take to produce a new smartphone? Or a cup of coffee (growing, shipping, etc)? My non-expert understanding is that it's way more than you'd use from AI. > People are scrolling Instagram on a phone that took 3,000 gallons to make, wearing jeans that took 2,000 gallons, drinking coffee that took 37 gallons per cup, watering their lawn at 17,000 gallons a month — and they’re upset about AI using a bottle of water per query. From https://joshuaheathscott.substack.com/p/you-dont-actually-care-about-the

@thomas_cothran Are you arguing that we should just accept the increase in impact due to AI because we already have impacts due to smartphones?

(Personally, I reject the idea that "people don't care about the water". I think it's a combination of not knowing; acknowledging that there is currently an impact, wanting it to be better; and not caring. There's a difference between "not caring" and "wishing it were better but not knowing what to do about it/feeling helpless to do anything about it")

I'd argue that it's quite a good question how people who buy new smartphones or new jeans (when they could buy used) reconcile their actions with the concern they say they have.

but what about the wasted water AI use? does it not matter because of jeans? i dont get it

Arguing that people should be free of hypocrisy before asking or working for things to be better is a losing battle.

It's more than that. Water usage is a real environmental issue. Limiting your concern to one of the least effective ways to deal with it is counterproductive

one of the least effective ways

Did you RTA?

Given some of his counterexamples: • corn production: I'm personally against the subsidies that drive so much of our corn production (and its impacts on the Standard American Diet, to say the least) • lawn irrigation: I don't have a lawn • golf: "golf is a good walk spoiled" • bitcoin is bad, too • steel: I'd argue we get good ROI on steel production

FWIW, I'm generally a fan of JHS (and have more than a few of his pedals). I'm disappointed in this article with dismissive phrases like "breathless headlines". He talks about marketing and agendas, and he's doing the same, I'd argue. (And again, water is just part of it) That said, I'm just as worried about what he's worried about: the impact of using the tech on our minds. But this thread is about the environmental impacts.

one can even make a case that most examples from the article are industries related to old and well documented human necessities aka real needs (food, clothing, socialising, etc), but is not needed cause at the end of the day AI is using all the same (or more) water and meeting no necessities it didn't fabricate

For me the problem is that it encourages a kind of analysis which is least likely to actually help with environmental issues. Real solutions (and I am not an expert so take it fwiw) tend to be much broader and have measurably larger impacts. E.g., internalizing externalities in the price of water, at least for industry

Where do you (personally) start? Where are you likely going to have any impact at all? Reducing corn production in the US? The Steel industry? Wait until you have the perfect solution before starting anything?

In my view, water is too cheaply priced for industrial usage. It's a complicated policy issue, because pricing to discourage wasteful usage is inflationary. There's not much I can do at a grand scale other than vote

What can you do on a personal scale (other than vote)?

(And again, water is just part of it, IMO)

> water is too cheaply priced for industrial usage. Agreed > complicated policy issue, because pricing to discourage wasteful usage is inflationary. Sure. Just because it's complicated doesn't mean we throw up our hands or dismiss it because we don't like the conclusions

@thomas_cothran If you really look into, it uses a lot of water in problematic ways. For example, a single GPU used more water to make than your phone 😛 There are like 800 GPUs per rack, and then you have the RAM, CPUs, the rack itself, the cables, the controllers, and so on. And the energy consumption is one of the biggest source of water use, plus the cooling.

The biggest issue for water being that data-centers can often be located in places where water is scarce, so while the world may have enough water for it, some of the localities where the data-centers are may not.

> What can you do on a personal scale (other than vote)? May I offer a marxist point of view? A data center is a means of production, so its raison d'être is to realise capital - and realise capital it will. Faced with competition from other data centers, profit margins will be squeezed, and so will its idealistic aspirations (such as being carbon-free). This means that if realising more capital requires using a more polluting energy source or consuming a city's drinking water, at least one company will do it, and the others will follow. This is how we got here: there is an enormous amount of capital seeking realisation in this industry, and the AI bubble has intensified competition over data centers. From a marxist perspective, among the few forces capable of slowing this competition are the state (through local regulations and state or federal laws) and the metabolic rift between the rate at which the planet replenishes natural resources and the rate at which capital consumes them. As for the state, this seems like a plausible avenue in the EU and parts of Latin America, but much less so in the United States, where the state and the AI industry have become closely aligned. As for the metabolic rift, that's outcome we want to avoid, so accelerationism would only make matters worse. This is also applicable for other cases in the examples from @thomas_cothrans article with varying degrees. From this perspective, the most effective solution is to organise politically to contest state power and regulate the industry. No thought voting, but by building durable organisations capable of advancing political demands and shifting the balance of power between capital, the state, and civil society. The good news is that, for environmental campaigns, many of these organisations already exist. Joining or reaching out to a local group is often a good first step. (also they don't need to be marxists, just genuinely interested)

It is (pleasantly) surprising how many local groups are springing up against data centers and how much they are pushing local government to adopt stricter rules or complete moratoriums. There was a massive petition-signing campaign for Lordstown in the surrounding towns and cities, and a lot of "noise" made at council meetings etc. Our state governor's race is between Dr Amy Action and Vivek Ramaswamy—and VR is very pro-AI and pro-datacenter, so a lot of the grass roots campaigning on behalf of Acton is taking aim at her opponent's permissive stance on big tech, which is something even a lot of deep red Ohioans do not trust/want. The local and state level elections in November are going to be extremely important for how much control we'll get over AI datacenter expansion.

That's all well and good from a supply side. There's the demand, side, too. Those of us writing and using software can impact the demand side. If there's lower demand for AI services, there's less justification for building out more data centers.

Demand is currently partially supported by unsustainable price models (actual costs aren't being borne by those using the services) and hype, IMO.

Ya, but based on Anthropic revenue consumer AI is a small chunk. It's mostly enterprise subs and API charges (which I assume are also enterprise)

@didibus Wouldn't we (as software developers, some of who are employed by or run those enterprises) be part of the enterprise subs and API charges rather than the consumer AI? That's how I see it, and puts us squarely in the primary demand space.

That's also why this particularly concerns me: my usage as part of my work directly contributes to the primary demand. It's not something the everyday consumer is directly contributing to.

(The everyday consumer potentially indirectly uses AI as consumers of the products/services I provide, but the direct consumer would be me.)

Ya, but I don't really control the decision of if we must use it or not. It's a good point though. I will say, I use AI personally as well, and I simply don't have the self-control to not use it 😛. I also don't have the risk taking guts to be the one not using it at work. That's why I like the regulation angle, sometimes you need rules that apply systematically. Like I can both be using AI all the time, and support an AI ban 😛 Just like I have imposed myself usage restrictions of apps on my smartphone, or how I try not to buy cookies instead of believing I will resist eating them.

While I'm sympathetic and feel some of those risks as well, that's also the abdication of personal responsibility that contributes to a lot of the societal problems, IMO. It's a collective action issue. I'm not a big fan of regulation/having laws for everything. I would like us to be able to/to have the courage of our convictions to make conscious decisions to do this on our own (I see some parallels to the mess we've gotten ourselves into with social media/networks, FWIW.)

(I'm not opposed to regulation in this case, I just would like us to be able to come to some kind of agreement without it being forced on us by the state)

I feel regulation is a conscious choice.

It's perhaps a collective choice, but for those opposed to the regulation, it's not a choice. There are a plenty of people that feel restricted by regulation just because it's a regulation.

Right, but say you don't use AI and live an environmentally sound life as much as possible, and climate change still happens, drought still accelerate, forest fires are more rampant, etc., because everyone else did. What's the point of that?

I think the "not a choice" angle is also wrong. It's a social agreement, an arrangement. It can change, evolve, it considers many voices, it often tries to strike a balance, etc. With AI especially, I feel we can have it both ways. There are net-zero emission data-centers, with fully renewable water consumption, etc. You can replace old fossil power plants with renewable ones. And for the environment, it's best to think of it as a common good, imagine we all own 1-share of it. But it's like owning a house with someone else, they can't just decide to burn their half of it, so it needs to be preserved and maintained until it's agreed jointly to be "sold off".

Sure, but I also would like to have a common understanding of what's communally beneficial so that (a) compliance with regulation wouldn't be such a fierce battle (and swing widely every few years); (b) people don't need to be forced to do things they don't want to do; and (c) I'm not thought of as a weirdo and can still have a reasonable number of options working in industry in a way that aligns with my beliefs.

Isn't regulation a communally beneficial understanding? Assuming a somewhat functioning democracy I guess, but I mean, it's kind of a decision that needs a large support behind it or it won't happen no?

Isn't regulation a communally beneficial understanding?

It seems there's some traction at the localities like Sean was saying of his. People don't want AI data-centers. I think some people though do think AI is worth its external cost to the environment.

Anyways, it's not an either-or. I commend someone who can live by their principles and have the willpower to act how they'd like to act, and not how they feel like acting 😛

I get the impression that many making the argument of the benefits of AI are motivated by making money (the loudest industry voices) and actually don't care about the environment. It's not a trade off for them because they place little to no value on the environmental costs.

Not many actually value environmental costs. Modern life was built on destroying the environment.

Ya for sure. But I think greed is also like hunger, sometimes you can't help yourself. And also you have to acknowledge the system, if every one company feel like they will lost business wise to their competitor if they try and do "the right thing", even if they'd like too, they won't. But it you make the rule of the game such that this is not the case, then you won't feel compelled in order to remain competitive that you have to ignore the environment for example.

I think that’d help. I don't know where the endpoint is, though. Right now it seems like unless you're actually convicted of something, it’s actually accepted as okay. I think that's bad. I don't want every aspect of life to be regulated by law. There’s no place for virtue or morality if that's the case. All actions end up being the result of fear of force/consequences, as opposed to choice.

(The counter example would be true civil disobedience, accepting the consequences because of your beliefs.)

One of the arguments for the free market is to allow people to drive some of this by their choices. I think the market should play a role.

I guess I see virtue and morality in choosing to put in place the mechanisms that will result in the virtuous/moral outcome. Would sure be nice if we could all act as we talk of our own will like I said, but we're fighting thousands of years of evolution with built-in chemical reward mechanisms 😛

I agree with putting in place mechanisms to help. I just don't know where the endpoint is.

I feel one issue with regulation as it is done today is that it tends to be rigid and evolve slowly. I feel it be nice to allow more experimentation with regulation, you can course correct it, optimize it for the outcome you want.

Also, if "free markets" is more your interest. I was reading that in the past the market tended to also maximize for stakeholders, like customers and society, not just shareholders. If you read about Friedman doctrine, neoliberalism, and the Chicago economic school. That was the catalyst when capitalism decided that all costs can be paid by society and it was the government job to figure out that problem, where-as corporations would focus exclusively on maximizing shareholder value, without any balance or care for employees, customers, communities, suppliers, etc.

I can see local legislation happening and maybe state-level -- in some states -- but federal legislation tends to lag far behind the technology curve, and most lawmakers at all levels do not understand technology and are not experts on ecological issues. That makes it much harder to "experiment with regulation" at any level beyond the state and, as we've seen, sometimes the federal administration tries to short-circuit the ability of states to actually "do the right thing".

I mean, experimenting with regulation should be, do something, see the metric go down, roll it back 😛 Re-evaluate, try something else. It should work even if you don't know anything. But ya, I know it's difficult to implement that level of iteration and each change destabilizes things

Hahaha... that presupposes governments actually measure things! 🙂