So far, in spite of the (always) lofty promises of Big Tech, AI seems more used by Google to give worse results, Facebook to summarize comments I didn’t care about anyway, and making pictures of fish with birthday cakes on their heads.
I’m sure somewhere deep down in the bowels of AI there is somebody figuring out how to cure cancer or make better soup, but mostly we’re ramping up untold energy consumption in pursuit of shiny things with glittery edges.
It is not just AI that is creating the power demand. Cloud-based computing and computer storage requirements for many types of applications are expected to increase power generation requirements in the coming years.
Here is one solution that Amazon Web Services came up with…
From the link…
Power provider Talen Energy sold its data center campus, Cumulus Data Assets, to Amazon Web Services for $650 million. Amazon will develop an up to 960-megawatt (MW) data center at the Salem Township site in Luzerne County, Pennsylvania.
The 1,200-acre campus is directly powered by an adjacent 2.5 gigawatt (GW) nuclear power station also owned by Talen Energy.
~ ~ ~ ~ ~
960 MW is a lot of power! Each of the two nuclear plants at Susquehanna can produce 1250 MW. I’m guessing a portion of that 960 MW is just for air-conditioning to keep that many computers cool.
~ ~ ~ ~ ~
Looking at the total US electricity generation statistics for the first 4 months of 2024, electricity from natural gas is up 30 million megawatt-hours over last year. By my calculation, that extra natural gas consumption increased CO2 emissions by 12.3 million metric tons. There was a small decrease in coal use, so the net CO2 increase is more like 10 million tons.
This is where AI matters and is working in just about every aspect of medicine.
It is not the same AI that parties down with images of fish.
I asked ChatGPT to name a few successes of AI.
Of course the answer was completely biased, IMO
Fighting Illegal Fishing: OceanMind, a UK-based organization, uses AI to track thousands of vessels worldwide. By analyzing data from sources like collision-avoidance transmitters, radar, and satellite imagery, they ensure fishing vessels operate only in permitted areas, helping combat overfishing and illegal fishing¹.
Tackling Human Trafficking: The Global Emancipation Network (GEN) collaborates with Accenture, Splunk, and Graphistry to create an AI-powered tool called Artemis. It identifies businesses and individuals likely involved in human trafficking, sending alerts to law enforcement agencies for further investigation¹.
Medical Diagnostics: AI algorithms analyze medical images (like X-rays and MRIs) to detect diseases early. For instance, Google’s DeepMind developed an AI that identifies eye diseases like diabetic retinopathy with high accuracy.
Drug Discovery: AI accelerates drug development by predicting molecular interactions. Researchers use it to find potential treatments for diseases like cancer and Alzheimer’s.
Here is one more that I read about.
Researchers in Kenya recently used AI to discover that elephants have specific verbal names for each other as well as elephants in other herds that are friends.
Mike
Items 3 and 4 on the list are the most promising, not so much because AI is “smart” but rather because AI is quick and thorough. And items 3 and 4 will have their success easily measured. Items 1 and 2, well, how do you measure their success exactly?
Maybe maybe not. Depends if nonprofits are using AI. The AI results can not be patented. Big money won’t buy in perhaps.
Total US power consumption in 2023 was about 4,000 terrawatt hours of electricity. The 2030 forecast of 323 terrawatt hours is equal to 8.1% of the 2023 consumption.
If the total US consumption of electricty in 2030 remains at 4,000 terrawatt hours, then AI data centers will consume 323 terrawatt hours or 8.1% of the total US electricity.
However if 2030 total US electricity consumption increases to 4,200 terrawatt hours, then AI data centers will consume 323 terrawatt hours or 7.7% of total total US electricity.
Second law of thermodynamics: Entropy increases
Life decreases entropy locally because it captures energy from outside sources. AI is just one example. While not yet a life-form, wait until we build big enough data centers
AI is just a question of brute force, 100 billion neurons and over 100 trillion synaptic connections will do it. How much energy does a big enough data center require?
Elon Musk gets it:
Gaia will need air-conditioning to support AI without overheating.
The Captain
A ton. A gazillion tons. Running at it a bit and byte at a time is a fool’s errand, we need to have multidimensional planes of organization (as the brain apparently does) so the same synapses are used over and over for multiple things depending on how the connections are formed.
Think about how much information you have recorded, somewhere. You see some fairly nondescript movie that you saw 20 years ago and say “Oh, I’ve seen this.” Sure, you remember “The Godfather”, but some mediocre film you saw while making out in the back of a theater? All the TV shows you have watched. All the books you have read. People, places, things you have encountered and experienced. It’s like the internet in your head - of all of your own personal experience. Maybe you don’t remember everything perfectly, but it’s amazing to me how often I say “Oh, I’ve seen this” and realize that the flash memory of it is stored somewhere up there, inside all those other images.
The **original 3x3x3 Rubik’s cube** has **43 252 003 274 489 856 000 combinations** , or 43 quintillion.https://www.therubikzone.com/number-of-combinations/#
Think about that. A simple 3x3x3 matrix: 43 quintillion. AI has a long way to go - and countless energy to expend - until we get to a place where we’re not measuring things by off/on toggles on a piece of baked sand.
I’m sure someone in those areas could come up with a meaningful metric. Such as number of illegal fishing busts or increases in certain fish species populations. Maybe number of trafficking arrests made or a decrease in known abductions.
Mike