On the one hand, we shall bring some napkin math to the table.
A human brain consumes something around 20W (Balasubramanian V. Brain power. Proc Natl Acad Sci U S A. 2021 Aug 10;118(32):e2107022118. doi: 10.1073/pnas.2107022118. PMID: 34341108; PMCID: PMC8364152).
One hour = 20 Wh, or 20 × 3.6 = 72 kJ
One day = 72 kJ × 24 hours = 1728 kJ or 1.728 MJ
One year = 1.728 MJ × 365.25 ~= 631.152 MJ
20 years ~= 12.6 GJ
The entire world population in 2024 (you’ll understand soon why I’m using 2024) was estimated as 8,141,808,945 (World Bank Group, World Development Indicators)
Rough brain power consumption for all humans who were alive in 2024 (I’m using values for one year instead of 20yo bc the 8 bi. accounts for all ages) = 8,141,808,945 × 631.152 MJ ~= 5.14 EJ (Exajoules)
In Joules it’s 415 TWh × 3.6 = 1.494 EJ (Exajoules)
My napkin math may be heavily inaccurate (hence “napkin”) but, yeah, Math tells us humans (roughly) consumed more than all non-cryptocurrency data centers (1.49 EJ is less than the 5.14 EJ required by 8 billion Homo sapiens for thinking).
And I’m only considering brain power. The number would certainly be bigger if I were to consider the rest of metabolic consumption, this would further consolidate the entire humanity, when taken together, as indeed consuming more energy than AI data centers worldwide.
On the other hand, hell no! I’m not gonna agree with Sam Altman! Especially bc they’re ignoring several factors.
For starters, the fact that AI and their data centers required humans, so the “human energetic bill” is shared with AIs, not disconnected from them. After all, AI is not something existing in a vacuum.
Fossil fuel, the elephant in the room, is another factor in play: I didn’t research a CO2 side-by-side comparison between human-emitted CO2 (from biological processes such as respiration) and the the amount of CO2 emitted to keep said data centers running, but this can’t be ignored.
Homo sapiens (usually) don’t ingest fossil fuels (i.e. in normal situations, we don’t drink gasoline… nor we eat coal).
Meanwhile, global data centers seem far from achieving green energy (e.g. hydro power), they rely heavily on fossil fuels, therefore they’re expected to be breathing out more CO2 than humans.
Tables would only turn regarding CO2 when (and if, a big if, considering how AI is currently at the hands of corps who, in turn, deny and ignore the climate change because “line must go up”) data centers pivoted to full (and true, not the “green-washing” creative accounting that tech corps usually do) green energy.
Because cryptocurrency data centers (normally) don’t deal with AI, and the object of comparison was all about AIs vs humans energetic consumption. In their specific speech, Sam Altman was trying to justify (albeit in a very twisted manner) the energetic thirst from their ChatGPT and the alike. So my napkin math focused on this specific comparison they made, hence why I tried to leave crypto and other non-AI-related data centers out of the equation.
If I were to include cryptocurrency into this equation, surely the entire comparison would lean heavily towards data centers, because things like crypto mining are highly energetically demanding.
And very polluting indeed. Really. If we consider the chronological aspect, crypto data centers did pollute and consume more than all AI data centers: Bitcoin is functioning since 2010 (when the block 0, aka Genesis block, was mined), it’s been 16 years, uninterruptedly (I don’t remember seeing news headlines such as “Bitcoin operations are currently down”, so it’s been operating for 16 years in a row), while ChatGPT, the one to open the Dantesque gates we’ve been facing nowadays, was released to the public in 2022, only 4 years ago and with several moments of interruption and downtime.
There is a bit of a feedback loop though. As value and work is often expressed in the amount of labour a person can do. With digital currency this is shifted to computer labor. But the basic cost of human labor is still being valued. In fact this might give us an exchange ratefor computer labor vs human labor.
I know your comment is about intelligence but that is difficult to measure. This is projected to be the bubble bursting bit of AI as it intelligence is limited but the amout of actual work it replaces is measurable.
Im not saying your are wrong, by the way, im just trying to understand the metrics of this. If you want any type of economic evaluation one doesn’t only need to involve the cost, but also the revenue. I don’t think we have the figures yet but I think that the production side of AI is still a bit underwhelming.
I do think that crypto data cost needs a place in this evaluation, just like heat management is a thing in automotive engineering. Even thout it is an unwanted byproduct it needs to be accounted for.
@yogthos@lemmy.ml @usa@lemmy.ml
On the one hand, we shall bring some napkin math to the table.
A human brain consumes something around 20W (Balasubramanian V. Brain power. Proc Natl Acad Sci U S A. 2021 Aug 10;118(32):e2107022118. doi: 10.1073/pnas.2107022118. PMID: 34341108; PMCID: PMC8364152).
One hour = 20 Wh, or 20 × 3.6 = 72 kJ
One day = 72 kJ × 24 hours = 1728 kJ or 1.728 MJ
One year = 1.728 MJ × 365.25 ~= 631.152 MJ
20 years ~= 12.6 GJ
The entire world population in 2024 (you’ll understand soon why I’m using 2024) was estimated as 8,141,808,945 (World Bank Group, World Development Indicators)
Rough brain power consumption for all humans who were alive in 2024 (I’m using values for one year instead of 20yo bc the 8 bi. accounts for all ages) = 8,141,808,945 × 631.152 MJ ~= 5.14 EJ (Exajoules)
In Joules it’s 415 TWh × 3.6 = 1.494 EJ (Exajoules)
My napkin math may be heavily inaccurate (hence “napkin”) but, yeah, Math tells us humans (roughly) consumed more than all non-cryptocurrency data centers (1.49 EJ is less than the 5.14 EJ required by 8 billion Homo sapiens for thinking).
And I’m only considering brain power. The number would certainly be bigger if I were to consider the rest of metabolic consumption, this would further consolidate the entire humanity, when taken together, as indeed consuming more energy than AI data centers worldwide.
On the other hand, hell no! I’m not gonna agree with Sam Altman! Especially bc they’re ignoring several factors.
For starters, the fact that AI and their data centers required humans, so the “human energetic bill” is shared with AIs, not disconnected from them. After all, AI is not something existing in a vacuum.
Fossil fuel, the elephant in the room, is another factor in play: I didn’t research a CO2 side-by-side comparison between human-emitted CO2 (from biological processes such as respiration) and the the amount of CO2 emitted to keep said data centers running, but this can’t be ignored.
Homo sapiens (usually) don’t ingest fossil fuels (i.e. in normal situations, we don’t drink gasoline… nor we eat coal).
Meanwhile, global data centers seem far from achieving green energy (e.g. hydro power), they rely heavily on fossil fuels, therefore they’re expected to be breathing out more CO2 than humans.
Tables would only turn regarding CO2 when (and if, a big if, considering how AI is currently at the hands of corps who, in turn, deny and ignore the climate change because “line must go up”) data centers pivoted to full (and true, not the “green-washing” creative accounting that tech corps usually do) green energy.
Why would crypto currency datacenters be left out of the comparison?
@Akasazh@lemmy.world @usa@lemmy.ml
Because cryptocurrency data centers (normally) don’t deal with AI, and the object of comparison was all about AIs vs humans energetic consumption. In their specific speech, Sam Altman was trying to justify (albeit in a very twisted manner) the energetic thirst from their ChatGPT and the alike. So my napkin math focused on this specific comparison they made, hence why I tried to leave crypto and other non-AI-related data centers out of the equation.
If I were to include cryptocurrency into this equation, surely the entire comparison would lean heavily towards data centers, because things like crypto mining are highly energetically demanding.
And very polluting indeed. Really. If we consider the chronological aspect, crypto data centers did pollute and consume more than all AI data centers: Bitcoin is functioning since 2010 (when the block 0, aka Genesis block, was mined), it’s been 16 years, uninterruptedly (I don’t remember seeing news headlines such as “Bitcoin operations are currently down”, so it’s been operating for 16 years in a row), while ChatGPT, the one to open the Dantesque gates we’ve been facing nowadays, was released to the public in 2022, only 4 years ago and with several moments of interruption and downtime.
There is a bit of a feedback loop though. As value and work is often expressed in the amount of labour a person can do. With digital currency this is shifted to computer labor. But the basic cost of human labor is still being valued. In fact this might give us an exchange ratefor computer labor vs human labor.
I know your comment is about intelligence but that is difficult to measure. This is projected to be the bubble bursting bit of AI as it intelligence is limited but the amout of actual work it replaces is measurable.
Im not saying your are wrong, by the way, im just trying to understand the metrics of this. If you want any type of economic evaluation one doesn’t only need to involve the cost, but also the revenue. I don’t think we have the figures yet but I think that the production side of AI is still a bit underwhelming.
I do think that crypto data cost needs a place in this evaluation, just like heat management is a thing in automotive engineering. Even thout it is an unwanted byproduct it needs to be accounted for.