Over me

Over you

Over everybody

Too much information running through my brain

Too much information driving me insane

— “Too Much Information,” Ghost in the Machine,

Gordon Sumner,

The Police, 1981

Awhile ago I wrote about bitcoin and its absurd energy consumption vs. function. Now there is AI, which is in its global infancy but is neither absurd, nor is its function. The potential is like that of the Big Bang. As follows, so will be its energy consumption.

Consider this analogy, the brain is only about 2% of the average person’s weight but consumes 20-40% of the energy of the entire body. Multiply this by just under 2 1/2 and you have the percentage of energy of the chips that power the brains of AI. Now multiply this by an ever-expanding supply of data and you have ... a very big bang!

Lincoln Laboratory at MIT is developing ways to reduce power, train efficiently and make energy use transparent. For example if you search for flights on Google you will notice that their carbon emission is listed next to their cost. A similar transparency does not exist yet for the computing industry. AI models like ChatGPT are examples of huge data centers that will consume large amounts of electricity.

MIT says that data centers will draw up to 21% of the world’s electricity by 2030. I think it will be much higher than that. The Lincoln Laboratory SuperComputing Center is developing techniques to help data centers control their energy use. The Lincoln Laboratories are powered from 100% carbon free energy.

Energy use in ‘training’ — an AI model

This is the process by which an AI model, like CHAT’s AI, learns patterns from huge data sets using graphics processing units (GPUs), which are power hungry hardware. As one example, the GPUs trained GPT-3 (the precursor to ChatGPT) are estimated to have consumed 1,300 megawatt hours of electricity, roughly equal to that used by 1,450 average U.S. households per month. Now consider that with an ever-increasing flow of data that needs to be updated what electricity it will take. This is yet to be determined.

A perfect storm is brewingAI and its ever increasing users will change the face of everything, but there will be a cost. Energy today is on a collision course with a perfect storm. Here’s the energy storm trifecta: No. 1, the world is changing — more people want more things and want to go to more places, etc; No. 2, there is the need to transition from fossil fuels to wind and solar; and No. 3, to top it off, meeting energy needs of new technology, today being AI and who knows what else.

For example Microsoft introduced Microsoft Flow in 2016 and it has recently been changed to MicroSoft Power Automate. This software inter alia can take care of redundant processes that were the mainstay of process control. These systems require a massive amount of storage, which continually adds data for these systems. At this writing, Microsoft just saw a surge in stock price due to its cloud AI income.

Energy costs with cloud data vs. personal storage

According to Stanford University Magazine (https://stanfordmag.org/contents/carbon-and-the-cloud) a Carnegie Mellon University study concluded that the energy cost of data transfer and storage is about 7 kWh per gigabyte. An assessment at a conference of the American Council for an Energy-Efficient Economy reached a lower number: 3.1 kWh per gigabyte. (A gigabyte is enough data to save a few hundred high-resolution photos or an hour of video.)

Compared with your personal hard disk, which requires about 0.000005 kWh per gigabyte to save your data, this is a huge amount of energy. Saving and storing 100 gigabytes of data in the cloud per year would result in a carbon footprint of about 0.2 tons of CO2, based on the usual U.S. electric mix.

AI going forward

— Phenomes and why Medical will be the largest single consumer of data energy.

First what are phenomes:

The phenome contains a set of traits that respond to the environmental changes occurring in and around an organism and modify its response for better growth and survival, according to the book Omnic in Horticultural Crops. Jeremy Abbate, VP of Scientific American, says we now are on the next iteration of the post-Human-Genome era, using data-gathering bioinformatic tools to harness insights into the complex interactions of our genes, behavior, environment, gut microbes, metabolites and other disease-influencing actors.

How much data for just health data are we talking about?

AI is front and center stage and will be for decades to come. Energy consumption by AI is based on the computing power of the chips and the amount of data that feeds it. In the future, I believe medicine will initially be the biggest single user of AI data. It’s no secret that healthcare is a data-driven business.

According to HealthTech magazine, “the average hospital produces roughly 50 petabytes of data every year. That’s more than twice the amount of data housed in the Library of Congress, and it amounts to 137 terabytes per day. Due in large part to the proliferation of medical devices, genetic testing and patient-generated health data, coupled with near-universal use of electronic health record systems, the amount of data generated in healthcare has been increasing at a rate of 47 percent per year.”

The numbers speak for themselves.

In the newest versions of most word-processing software, saving a document for the first time presents you with two options: saving to your computer or to a cloud storage service. As your finger hovers over the mouse button and you make your choice, it’s important to realize how different the two processes truly are. Saving a document to the cloud takes it out of your hands, so you don’t have to worry about losing it and can access it anywhere. Cloud data is stored not in actual clouds but in buildings — massive structures filled with thousands of hard drive-bearing racks using a mind-boggling amount of energy. There are millions of data centers around the world, some taking up nearly 200 acres apiece.

Conclusion

In my very humble opinion:

No one can dispute that we are experiencing a whirlwind of technology change that is affecting us individually. The problem with change is not it inevitability but our ability to adapt. Energy use must be accompanied by reasonableness of applications. By this I mean energy and the advancements of technology must be compatible and that they are interconnected. For more reading on this check out LeRoy Hood.