Hacker News Reader

Comment by AvSaba | Hacker News Reader

Saying a particular sector (AI/Crypto/...) consumes a lost of energy and we should do something about it, is like treating the symptoms and ignoring the root causes. The problem is mostly how we produce energy, not how we consume it. Consuming less energy on a particular sector means people in that sector have more money to spend on other things and those other things also cost energy (ok maybe less). To be clear I'm not saying we should totally ignore how much energy a sector consumes but most of our focus should be on how the energy is produced.

View on HN

I don't think this is correct. I think that in essence what we've learned over the past 100 years is that we need to be more conscious of the totality of the cost benefit analysis, because what happens is that a lot of industry can can make private profits by exploiting public goods. So it's actually encouraging that we're starting to more objectively look at different activities and "Ok, here are the private profits of this industry and here are the public costs". Energy is a really simple element of that, and in AI it's probably not the most important, but it is very easy to measure. It's very easy to say "If you want to do this AI stuff, it's going to mean X tonnes of carbon or Y tonnes of rare earth metals or Z square miles of wind turbines".

Having said that, in Crypto it's a much more compelling argument because it's like "Here are some obvious, big costs that are fundamental to the system, and there are almost no benefits" whereas in AI the nature of these algorithms is that we're likely to improve them, so comparing training GPT-3 is kind of like measuring the power draw of the LHC. It's certainly big and it's certainly something we can talk about, but no one thinks we're going to be building 100,000 LHCs.

> Saying a particular sector (AI/Crypto/...) consumes a lost of energy and we should do something about it, is like treating the symptoms and ignoring the root causes. The problem is mostly how we produce energy, not how we consume it.

For AI, sure, for crypto (if PoW), no.

If energy was a tenth of the cost, PoW crypto would use 10 times more.

You do have a point, and I agree that making the production-side clean should be a large focus, but there are valid arguments for reducing consumption, too, I think.

Supply of energy is limited, and so there is a market effect: overconsumption by some players can increase the cost to other consumers.

Market economics should mean that the resulting increase in profit margin turns into an opportunity for more-cost-effective competitive entrants.

Most of this assumes no collusion between players in the market. If, for example, company A provides a product that is knowingly energy-inefficient and half of their board members also sit on the board of energy company B that is experiencing record profits -- and could easily acquire company C that theoretically has a cheaper energy production solution -- then perhaps problems could occur.

(note: the shared-board-seats is mostly a lazy, hypothetical and probably unlikely example. similar issues could probably occur due to vocal activist investors, capital funds, potentially even pools of shareholders aligned on similar goals. these theories rest on the idea that consumer adoption of products isn't entirely organic and rational, and that demand can be manufactured/guided)

Totally agree. 5% of power plants are responsible for 73% of emissions [0]. That's like 1500 power plants. Rather than fix the emission issue at this small source, many would rather blame the population at large. Rather than fix 1500 big issues, they'd rather change the behavior of billions of individuals. It makes no sense.

0: https://www.smithsonianmag.com/smart-news/five-percent-power...

I think the main problem is the unaccounted externalities in the transactions.

Reply on news.ycombinator.com