Are we overcomplicating progress?
Today I wanted to talk about software and sustainability. Not on a top-level basis like Ecosia is better than Google, but more from a data and programming point of view. The reason I've been in this rabbit hole is that one of my new clients is actually a university (the same one where I was doing part of my PhD), and I was asked to do a guest lecture in a course called ‘Sustainable digitalization in practice ’, and wow, did that open Pandora’s box!?
Because of that, I have been discussing things like ‘AI’s energy problem’ from the perspective of someone who started with computers when every byte and bit mattered (my dad).
In the 1970s and 80s, memory was expensive, and processors were slow. Programmers had to think carefully about how much data they used, how often they moved it, and how much work the processor actually needed to do. Efficient software wasn’t a nice-to-have- it was a necessity.
Today, we have almost the opposite situation. Computing power is abundant, and when we need more of it, we tend to add more hardware. AI has taken this approach to an enormous scale: more GPUs, more servers, bigger data centers, which means more electricity.
However, on modern computers, the actual calculation is often surprisingly cheap compared with moving the data around. Getting data from memory to a processor, moving it between GPUs, keeping thousands of machines synchronized and repeatedly transferring huge amounts of information can consume significant amounts of energy.
That makes some very old ideas relevant again. Smaller data (you have the same information, but each piece takes fewer bits for example: storing each number in 8 bits instead of 32 bits) can mean less memory traffic. Better algorithms can mean fewer calculations. Keeping data close to where it is being used can mean less movement. Using fewer bits can therefore mean less memory and less computation.
Sorry for the technicality, but please stay with me; I will get to the point (I promise), and you will move differently in the world when more people have a better understanding of this.
AI researchers are already working on many of these things, but what interests me is how little of this seems to enter the broader discussion about AI and energy. We talk a lot about how many GPUs a new data center will contain, how much electricity it will consume, and how much cooling it will require. Important questions, but why aren't more people asking: could better software reduce how much computing we need in the first place?
The historical irony here is that in the early days of computing, we optimized because we didn't have enough resources. Today, AI may be bringing us back to the same engineering mindset, except that the scarce resource is no longer a few kilobytes of memory. Instead, it is electricity, cooling capacity, and grid infrastructure.
All of this made me think of what I’m overcomplicating, what was better before, what is the compromise, and what am I optimizing for.
So often in modern life, we are short of time, and therefore optimize for efficiency, or at least so we think. Every problem often needs a new system, a new app, a new digital solution. We use audiobooks, digitally store our accounting, the kid’s school has yet another app (what was wrong with a note in the backpack?), and we create online shopping lists to share with our partners. None of it is bad per se, but maybe just maybe the compromise is too high? It is something that I have really taken to heart in my own business. I'm not optimizing for digital efficiency anymore if the task at hand wasn't that time-consuming to begin with, I actually liked doing it, or I believe that there is a secondary benefit to me for doing it myself (hello pretty stationery and hand lawn mower.)
Sometimes it feels like we got so busy with things like "don't print this email to save a tree,” but never took the time to reduce the actual emails we are sending- and here we are in an energy crisis instead (extremely simplified, but you get the point)
Would love your thoughts and topics around this? Just hit reply, and I'm here for it.
With love,
Alexandra