I see news stories that will give examples of how much energy a type of technology uses (usually AI or crypto). They’ll claim very big numbers like the whole ecosystem using “as much as a small country” or one instance of use being “as much as an average home uses in a year.”
With the crypto ecosystem being so big and I’m less inclined to defend it, I haven’t thought as much about the claims. But with AI while it still has problematic aspects, it also has a lot of useful applications. When I run a single query the idea it’s the same energy as driving my car ten miles or whatever doesn’t seem to pass the smell test.
How are these numbers generated? Historically media doesn’t do great with science reporting (“a cure for cancer was just invented” etc) so just trying to get some context/perspective.
This is a specific (and probably the most well-known) case, but for details about how the Bitcoin Energy Consumption Index is calculated, you can hit https://digiconomist.net/bitcoin-energy-consumption and look under the header “Energy consumption model and key assumptions”. There is a summary and a link to a paper detailing their methods. Skepticism about science reporting is super important! And I’m certainly willing to have my mind changed, but it looks like there’s a lot of substance to these claims.
Thanks for sharing this, just the kind of thing I was thinking about.
Are you aware of anything related to AI energy usage?
This is a paper about it by the same guy: https://archive.is/AQhiC#selection-3573.66-3573.67. I found it while reading an NYT article about AI consumption.
deleted by creator
With prejudice.
Someone already answered how the data is obtained, but another way to look at the question is how we should evaluate the amount and type of energy a technology uses. The important bit is comparing it to the alternative, for example electric heaters are 100% efficient, that sounds tough to beat, until you learn that burning gas is over 100%, however it produces more CO2 if your electric heater is running from a clean source of energy, finally heat pumps are at least 300% efficient and can be ran with clean energy, so something that looked good at the beginning is actually quite bad when you look at alternatives.
I see this argument about cryptocurrency raised over and over again by people who don’t understand the technology, so first of all I’m going to explain a little bit on cryptocurrency and why they consume ungodly amounts of energy, if you already understand the terms peer-to-peer electronic currency and PoW feel free to skip the next paragraph.
The first important question is what IS a cryptocurrency, the shorter answer is that cryptocurrencies are money that can be used electronically without a centralised organisation regulating it and without needing to trust in anyone. This is very different from using your credit card where you have several layers of organisations controlling things over, and you need to trust that every single one of them will do their job properly. Every time you buy/sell with a credit card you’re trusting that the machine will not clone your card, the value displayed is actually what’s being sent, MasterCard/Visa/etc will process the payment properly and in the correct amount, your bank will remove the correct amount from your account, the government will ensure money has some value. If you go to a shoddy seller one of these might not be accurate, and if you live in certain countries others of them might not. But how does cryptocurrency achieve being that secure? In short essentially it processes lots of transactions that happened over some time and produces a sort of puzzle with it, to try to make it human understandable let’s call them crosswords, imagine that to produce a block of transactions you needed to arrange them in a grid of a given size, it’s quite hard to do and you probably will need to try multiple times to make it fit properly, but verifying that it’s a proper crossword is quite easy. The more people you have trying random combinations, the faster you’ll get to a result, the problem is that others are also trying to find the same crossword, and the first one to find it gets a price. To make matters worse if people are consistently finding the crosswords in less than a given time the rules change and the crossword becomes more difficult to solve to try to bring it back to the same time, i.e. the fact that you were able to find a crossword is a Proof that you Worked on that for a while, called Proof of Work or PoW for short.
Ok, I know that was a lot, but bear with me a little longer. First of all Cryptocurrency are replacing money, so the alternative is money, how much (environmentally) it costs to print it, ship it, store it, transport it, secure it, etc, including how much it costs for banks and credit cards to run servers and machines for it. When you put it all in perspective the amount of electricity spent by cryptocurrencies is WAY more environmentally cheap. Especially when you consider that most crypto farms are near green sources of energy because they sell their excess electricity very cheaply. Finally the amount of electricity a cryptocurrency uses is their security standard, because for you to generate the crosswords faster than everyone else with any certainty you need to at least personally spend 51% of that electricity amount which is a ludicrous thing to maintain, especially for larger and well known cryptocurrencies. Even so some other cryptocurrencies are migrating away from PoW, to Proof of Stake, where people stake their cryptocurrency that their block is valid, if it’s not they lose that money, so drastically less amount of electricity is needed.
Thanks for your comment. I’m familiar with crypto and built my own mining machine at one point in like 2014.
It’s not accurate to say the energy cost of conventional finance is higher than proof of work blockchain. Sure maybe in absolute terms right this minute (although even that I’m skeptical of) but the per transaction energy cost is much much higher for blockchain. If we scaled Bitcoin to handle every transaction in the world the energy costs would massively eclipse conventional finance, and that’s before accounting for the corresponding massive increase in mining difficulty.
I stopped using my mining rig when the cost of the required electricity outweighed the value of the coins generated. I was also mining an altcoin so the difficulty was lower than Bitcoin. Even so, I wouldn’t invest in a proof of work chain at this point because of its energy consumption. I think proof of stake is fascinating, but have mostly checked out of crypto these days.
I said the environmental cost, let’s just look at one tiny fraction of it, money needs armoured trucks to move around, an armoured truck does 4 Km/L according to a quick google search, every bank has at least one armoured truck that gets driven around daily, let’s assume they use a single litre of fuel each day to account for the possibility of some days not being driven around. This means a single bank uses around 200L a year. Just in the US there are over 72k bank branches, that means ober 14 Million Litres of fuel. A diesel engine produces 2.7Kg of CO2 per Litre, i.e. a very low estimate is that 39 thousands Tons of CO2 emitted just in the US only by moving money around.
When you add all of the small stuff, for all of the physical currency in the world, plus all of the stuff for the digital version of that same currency I think it surpasses Cryptocurrency usage by a long shot, especially because crypto farms are almost exclusively located near green power plants because for example during raining season electricity is much cheaper since they will generate it regardless of usage.
but the per transaction energy cost is much much higher for blockchain. If we scaled Bitcoin to handle every transaction in the world the energy costs would massively eclipse conventional finance, and that’s before accounting for the corresponding massive increase in mining difficulty.
That’s the point that everyone misses in cryptocurrency, the technology can already process global levels of transactions, the amount of energy that a crypto farm consumes is independent of number of transactions processed, finding a block with one or thousands of transactions is essentially the same difficulty, and consumes the same amount of energy. The cost of energy is only directly related to the difficulty, and the difficulty is only a matter of how fast blocks are being found, essentially the system tries to stabilise so that you get one block every 10 minutes regardless of amount of computers trying to find the next block. If there are more computers the difficulty goes up, less computers less difficulty, but the size of a block, i.e. the limit of transactions that can be processed during a period of time, is virtually infinite (not technically because you need to account for network speed, storage capabilities and a bunch of other stuff, but it’s independent of electricity consumption). To prove that take a look at Bitcoin Cash (BCH), it’s essentially a fork from Bitcoin that increased the block size but where the underlying technology remains the same. It uses a lot less energy BTC but that’s only because less miners are interested in it. Even using 10% of the energy, currently BCH can process 100 transactions per second, to put that into perspective MasterCard processes 5 to 6k transactions per second, while Visa a whopping 65k, so BCH can’t currently compete with either, but it outperforms BTCs 4 TPS by a LOT while consuming a lot less energy even though it’s the same technology. So again, the cost of energy of a cryptocurrency is a function that depends only on the amount of miners, not what those miners are doing.
Even IF amount of transactions had a significant impact on time to find a block, that would mean that more adoption would diminish the difficulty to maintain balance. E.g. if 100 computers take 10 min to find a block of size 1MB, but they take 20 min to find a 2MB block, the difficulty will drop by half of there are twice the transaction to maintain the average of one block every 10 minutes.
Finally, none of this has anything whatsoever to do with profitability. A coin being profitable to mine dictates how many computers will be trying to mine it, but the system self regulates, if miners migrate the difficulty goes down to compensate. But less difficulty makes it more profitable so miners come back, so on and so forth.
I don’t think the goal is to replace cash, it’s to replace electronic transactions, so I don’t see that as an energy savings. (also every branch of every bank does not have an armored truck)
The reason I suggest difficulty of mining would increase at global adoption is because we already see the price of a single Bitcoin went way up with very little adoption. If the entire world was using it the price would go up much higher. The higher the value of Bitcoin the more people will be drawn to mining, the higher the difficulty of mining goes, the greater energy required to sustain the blockchain.
You are wrong, cryptocurrency replaces money, not digital transactions. Bitcoin has its own value, independent of dollar or gold, it’s not meant to be used instead of MasterCard, but instead of dollars. If you only look at a small fraction of what cryptocurrencies do the energy consumption will look huge, it’s like using a supercomputer as a calculator and complaining it consumes too much power.
That’s not exactly how it works, more adoption doesn’t necessarily increases price, ethereum and Bitcoin Cash are a lot more used than BTC yet their price is lower. Price is determined by speculation, and the vast majority of people have only heard of Bitcoin, so that’s what they invest in. Also Bitcoin can’t grow any more, essentially everyone with an ASIC profitable in Bitcoin is already pointing it to Bitcoin, so it becoming more popular couldn’t possibly make more ASICS point to it, and if difficulty starts to increase the less profitable ones would mine something else and difficulty would fall again. If on the other hand any other coin became the most popular then part of those ASICS would look to that.
Don’t you think if some other currency surpassed Bitcoin as the best known and universally adopted currency that people would start speculating on the new currency instead of Bitcoin?
The issue is both proof of work and limited number of coins. As more people use a limited number of coins, the price will go up and so mining difficulty will go up, and so energy use goes up.
I think the future of cryptocurrency is inflationary coins that use proof of stake. (Also with anonymity built in instead of pseudonymity)
If a coin is used the speculating goes down, same reason people don’t speculate with dollar value, if X coins buys you a hot dog it gets a lot harder to speculate, the price becomes what people who use it give it. This is similar to gold, while it was used like money it was quite stable, but as people started to use dollars and the dollar decoupled from gold, actual gold became a lot more speculative.
Again no, the more coins people hold the more the value goes up, people actually using the coins doesn’t interfere in their price, unless you get to a point where there aren’t enough coins for everyone. But even if it got to the point that the price increases because everyone wants to use it and there aren’t enough coins for everyone to use, I don’t see how that would make mining more expensive. Like I said, processing 1 or 1000 transactions has a similar difficulty. The price of the coin going up could affect profitability, but the amount of ASICS is constant, so the power is already being consumed to mine other coins and would just be pointed here instead of there but would still have been consumed regardless.
I agree inflationary PoS coins, especially ones with smart contract capabilities such as Ethereum are way more useful. I understand the appeal for anonymity, but I disagree on the importance of it, I think most cryptocurrencies are anonymous enough even though like you pointed out they’re not fully anonymous.
But with AI while it still has problematic aspects, it also has a lot of useful applications.
Ah yes, stealing content en masse and polluting the whole internet with junk content in the hopes of being able to monopolize entire industries. Peak usefulness.
(There are of course many useful applications of AI in general. But they also tend to not burn through as much energy and processing power as LLMs)
I use an LLM pretty much every day to assist with software development. I find it to be very useful.
That’s good for you, however, content generation from these models has still polluted the internet and using Google’s Image search is impossible.
Thought I replied to this but don’t see it here:
If you look at my original comment I said that AI has problematic aspects, and also has real uses. I don’t think what you said is inconsistent with what I said.
I’m not sure what original comment you mean.
The reason for my comment was that the comment I replied to wasn’t addressing what the comment you were replying to, brought up. They bring up issues with using Google now has worsened because of LLMs and your reply said “well it’s useful for me in software development”. Like I said, that’s good for you but doesn’t address actual issues.
Yeah, yeah, that’s what they said about the steam engine, and… Oh, it did exactly what you said. Nevermind.
After posting this I just read this Ars Technica article where the headline says crypto uses over 2% of U.S. electric generation, but then if you click through to the source, it says a preliminary estimate is between .6% and 2.3%.
And this is why scientists tend to dislike science journalism. Whether intentional or not, it misses out the nuance that scientific findings usually require. And then those misrepresented facts start getting quoted elsewhere until you get a web of “sources” that just point in a circle and can eventually cut out the actual source that includes the nuance.
Always be skeptical of journalism headlines and check the body. Then be skeptical of that and check the source. Then still remain skeptical of that because one study doesn’t determine scientific consensus. Before long you’ll be in the rabbit hole of the replication crisis.
But don’t write off science. It’s flawed because humans are flawed, but it’s still our best tool for determining truth.
The worst part about the headlines is that they’re usually not made by anyone scientifically literate, but by editors who just want the site to get more clicks. Even if the article itself articulates fantastically what the scientists are saying, all it takes is an editor putting a clickbait headline on it for the average reader (let’s be honest, skimmer) to ignore the good reporting and only remember the inaccurate headline.
For me, the energy consumption estimates are meaningless without a comparison to alternatives. For example, the headline “2% of electricity is used for crypto” is meaningless without an honest comparison against alternatives, like energy cost of gold mining, banking institutions, and financial transaction networks.
After all, a single Google search has an ecological footprint.
I think a lot of headlines are not exactly honest or at least diligent with that side of the story.
Crypto is an alternative to none of those things. Its a ponzy scheme with extra steps.
