Data centers?

[pinkytoe]

These giant complexes are popping up everywhere since we and all of our devices (including cars) and services need to be "managed" by AI and technology. SIL and the small Texas town she lives in are fighting three different ones so she will have trouble selling her house if they get approval. We are all hooked now so what are your thoughts on how this will play out over time?

[catherine]

Yes, and many family farmers are being offered millions for their land. "Amber waves of grain" will soon be replaced by soulless industrial parks. I just went to the library to pick up one of the books that rosa and iris lilies recommended--but neither was available, so I brought home the book Abundance by Ezra Klein and Derek Thompson. I like Ezra Klein and I've wanted to read this book, but I am anticipating that it will paint a picture I'm not ready to be part of yet.. a very high tech world of AI, plentiful electrical power for driverless cars, and food from vertical farming in hydroponic high rises. .

[LDAHL]

I’ve read that some organizations are constructing underground or underwater data centers to reduce their footprint and power draw.

[bae]

Data centers, as they are envisioned and being built today, are problematic.

I believe many folks are ignoring a looming issue though. Building giant expensive datacenters for AI processing is fundamentally stupid(*), for technical reasons.

I have been doing AI consulting work for a while now with one of the major AI companies. I believe there are roughly half-a-dozen factors that influence the speed/efficiency/capability of an AI model. Each of those factors is on a *rapid* improvement curve, the likes of which we have rarely seen in technological transitions before. In my career I have lived through several of these curves (cpu speed, data storage costs, internet/communications speed/cost, ...). Look up Moore's Law. I have never seen this rate of improvement of a technology along so many fronts before.

These factors are mostly synergistic, and in a multiplicative way not a linear one. If you look at the curve of the rate of improvement, for most of the factors it is faster than we have experienced in other technological realms, by *a lot*. Fundamental improvements in algorithms happens nearly every day, producing not just a 1%-level improvement, but something much more significant. (I devote ~2 hours a day to just keeping up with the literature, it's not humanly possible to read it all, so much new work is generated, as so many people around the world with Big Brains are doing fundamental research right now.). Foundational improvements are being made nearly every week. (And a huge portion of these improvements are coming out of non-US research groups.)

Only one of these important factors is "I have a giant data center and am making it bigger by burning phantom investor dollars".

Short answer: people building the giant data centers are barking up the wrong tree, fundamental improvements to how AI works "under the hood" will quickly dwarf the gains to be had by turning Montana into a nuclear-powered datacenter farm. (I can already run AI models on my iPhone with similar performance to the "needs a giant datacenter" models of 2 years ago.). I suspect in a handful of years, insanely capable AIs will be able to run on your Apple Watch or iPhone. The models I can run locally in my lab on a current-production Apple silicon computer are pretty capable already.

Imagine, by way of analogy, if companies in the 1960s, seeing how capable giant mainframes were, the sort of mainframes that took up entire buildings and required water cooling towers (I used to use one of these), decided to go all-in in building lots of warehouse-sized datacenters for the super-awesome mainframes. And imagine if the pace of progress of computer hardware and software design was way faster than Moore's Law. Imagine if modern iPhones or Apple silicon MacMinis were just 3-4 years down the horizon from existing on the day you poured the foundation for your giant mainframe datacenter. You'd look pretty stupid, and lose a lot of money.

That, I believe, is where we are today.
(*) I understand the desire of some of the AI companies to not "surrender" the market and fall behind, and thus their mutually-silly decision to compete for now in the datacenter arms race...

[catherine]

So, another analogy would be like when every hotel around the world replaced their old CRT-TVs with flat screens and dumped them in landfills? Is the problem what will we do with millions of square miles of obsolete infrastructure in, say, 2050?

[early morning]

Or like tearing down old but lovely, well build schools because "it was too hard to cable in/ put in computers" when wireless tech was growing by leaps and bounds - and thus spending millions on poorly built new schools that within the first two years of their existence went with mostly wireless tech anyway? AND have major roofing, plumbing, and other structural issues. (This has been the main reason that no one in my immediate family will vote for a school levy in this district again)

[Alan]

Or like tearing down old but lovely, well build schools because "it was too hard to cable in/ put in computers" when wireless tech was growing by leaps and bounds - and thus spending millions on poorly built new schools that within the first two years of their existence went with mostly wireless tech anyway?

I went through that with my house. When we built our house in 1995 I thought it would be smart to run Cat 5 network cabling to several different locations in each room with 16 individual home runs terminating in one location in a basement utility room in preparation for a whole house wired LAN. While I was at it I also ran telephone wiring to each of the network jack locations with dual RJ 11 and RJ 45 wall jacks. That worked well for us for a few years but it didn't take too terribly long for wireless networking technology and cell phones to make it all pretty much obsolete.

[rosarugosa]

I might have mentioned this before, but I was a child guinea pig about 56 years ago (so about 1970), for an MIT project on teaching children with computers. The leader was Seymour Papert, who was a much more amazing person than I realized at the time: https://en.wikipedia.org/wiki/Seymour_Papert. I used a Logo Turtle and programmed it to draw mushrooms on the floor with a magic marker, among other things. The one we used looked kind of like a shop vac. The whole experience was a lot of fun. From our point of view (four pretty brainy 12-year-olds), we were just getting to play with computers, and interact with fun and interesting adults that paid us a lot of attention. Computers were a new and amazing thing that most regular people had not yet encountered. Anyway, short story long, there was a computer that I recall as being about the size of 3 - 4 refrigerators. I'm sure it did other things, but what we were shown was a little black and white screen with a little game where you blew up rockets. It was so "wow" then, and would be so utterly lame now compared to even a first-generation Game Boy let alone a contemporary smart phone.
Overall, a long-winded way of saying that I completely agree with your point, Bae.

[catherine]

Cool story, rosa! Yes, look at where computers have gone... I used to be amazed at my great-aunt, whose father owned a livery stable in Bridgeport, CT before cars even hit the road. When she was in her early nineties, I sat with her as we both watched Neil Armstrong walk on the moon. What a difference a lifetime makes in the world of technology. For her, it was transportation technology. For us, it's computer technology.

My computer story began when I worked at NBC. One of my first jobs there was to create "commercial routines" for certain shows. (My first show was Jeopardy). So I would get the information from the sales department about what commercials should be plugged into which blank commercial break spaces, and then I would type the information up, make a few Xerox copies, and walk them around to various departments--the video department, the film department (which was shrinking drastically in the TV world in the 70s), the sales department, and B.O.C. (Broadcast Operations and Control).

After a while, a big mainframe computer was built on our floor. We were trained on how to type our commercial data using the computer. This was such a big deal--a bunch of us production assistants organized and went to our boss to make the case for a raise, because of this new "skill" we had learned. Not surprisingly, we didn't get a raise, but we were praised for our initiative.

[Rogar]

I hope Bae is right. I've heard theories that AI may find solutions to the problems it creates.

I've wondered how much of the demand for AI is industrial and related to increased productivity or science and medical research and advancements, and how much is personal frivolous Tik-Toks, putting animal heads on friends bodies, computer generated videos, getting prime orders the same day, and the general widespread use of smart phones.

I can recall feeding punch cards into computers and learning COBAL. It seems like there was a time when we went to the moon using slide rulers. It's almost like a new era.