I’d be interested in the zero-carbon path that the software industry anticipates, if there was such a thing. A professional guild of coders may be concerned for the havoc their energy consumptions are unleashing on the world. Unfortunately, coding is a new profession, with no guild or central craft, having sprung up quite recently due to the unreasonable success of a few mathematical logicians.
Instead, the people and the profession are cogs in a wider corporate machine. Our climate action plans are written for us, by sales most likely.
Coder climate action
What might a coder call to climate action look like?
It would firstly recognise and seek to own the energy pathways that coders lay out, and add up just how much of a problem the industry is, in relation to everyone else. The most popular number is around 6% of total energy usage1.
This is a number from Gartner 2007, however, uncalculated since. The number, if you include household computation and allow for global growth in computation energy cost might be around 10%. The
computation complex that includes software development, computers, the internet and media streams is about the same size as Aviation. As big a problem as planes.
And yet, if you look through the IEA2 reports and models, the industry is classified ambiguously, as brown goods and what is value-added services. Software energy usage patterns are hidden within the fabric of the matrix, someone else’s problems to solve.
This makes software a bit special because it is, in reality, a coherent project and industry, one of the few that can move the dial. And we’ve just seen the first ever reduction in human energy usage, the first of many perhaps, driven by software facilitating pandemic-response behavioural change. And at a speed no-one alive imagined possible. Software is talking a bow and leading the way.
This combination makes the industry powerful indeed. It can, for example, participatein and celebrate the outrageous energy consumption plans of Bitcoin, some utopian vanity project. Name another big energy vampire prepared to flaunt and moon us in the here and now. Big Software, as part of the Big Tech posse, will not only eat your world; but burn it too, because we can.
Helped by this invisible power source, the sum total of all green commitments across the entire computational complex is a bug.
Take Google3. There’s not a single commitment to actually reducing energy footprint in any way. When you commit to buying renewable energy rather than carbon, you help build confidence in the new technology at the margin, but you also push prices and people around so that the overall excess energy consumption just shifts somewhere else. Beyond this, they offer machine learning to the masses, helping cities and forests by providing ever more data to them, and branded machines who are learning how to help, via consumption of massive heterogeneous data sets (MHDSs).
Welcome to the Sunlight Age.
Since Software doesn’t seem to be on the field, they may not understand recent evolutions in potential climate action pathways, worked on by our best and brightest. Just like mathematical logic, solar PV and wind to a lesser extent has been unreasoaby sucessful. So much so that, across the scenarios a consistent line is forming, a milestone in development, where energy usage will no longer be a problem. We will turn off the meter readers and allow anyone access to energy if they want it. Not because it will be free but because a metered system costs too much versus the societal base price.
Welcome to the Sunlight Age! That’s where we want to get to and if we get to there, huzzah! Just like computation before it, the renewable endgame will close off another epoch - the one we are living in right now.
The Great Carbon Squeeze
When people right now talk about trusting the science behind the climate crisis what they’re really speaking to is our measurements of size, scope and the morose certainty of it all. The knowledge that this is happening at such scale gives it reliability, and extra oomph.
What will our era be like? Let’s assume for a moment that our best and brightest are working on climate response pathways, and not trying to understand MHDSs. If you place trust in the pathways they map and evolve, between us and the Sunlight Age lies a gap. The gap is a shrinking carbon supply - the distance between total energy demanded and renewable energy available. There is an energy usage squeeze coming, beyond our experiences, as the fossil fuel economy is shut down.
So whether restricted by government decree, or by market forces, or famine and pestilence, or food security, a reliable prediction is that global energy usage is going down a lot. Trust the certainty of our Gaian measurements; the obvious consequences and likely action and response.
Within the Great Carbon Squeeze, how does the Software industry reduce energy use and be most helpful? As much fun as they will be, nanotech, quantum computers and other thermodynamic adventures towards the the von Neumann–Landauer bound will be more likely Sunlight Age past-times. We may have to cobble together something from what we have at hand.
We have Moore’s law, in the physical infrastructure world, but it’s looking shaky. Bottlenecks along multiple production paths and recent shortages signal wide systemic malaise. An entrepeneurial engine run out of puff.
Is that it? We’re as good as you get and you’ll just have to live with less computation? From Walport’s tour de force survey4 of the intersection between thermodynamics and computation, here’s a hint of another path:
quantum computers as currently understood could only reduce the costs (and increase the speed) of an extremely limited set of possible computations, like factoring products of prime numbers. More importantly, many of the major challenges facing modern computing, both in terms of speed and thermodynamic costs, do not arise in the operation of the CPU (which is where quantum computation might be helpful), but rather in the i/o. In applications ranging from widely popularized applications of deep learning to search engines to simple back-office database management, modern computation exploits massive, heterogeneous data sets
Across a wide range of computations; management reporting, AI media selection, data mining and machine learnt business logic, MHDSs congregate. And in every case, you can liken them to a form of computational composting.
From inside looking out, the informational content in MHDS’s is incredibly sparse. The likelihood that any particular piece of data in them matters, that it will ever effect the world, is vanishingly small. Machine learning tools like python and SQL run big sieves repeatedly through this low-nutrient soup in the hope of constructing a meal.
Big State not Big Data
My main Machine Learning interface suggestion looks a bit like this:
-- | extracts information from lots of Stuff. -- -- DEPRECATED: Current usage tends towards inputing lists (or streams) of Stuff -- that are somewhat less than infinite in size, but only just. -- -- Do not use this function if energy cost is at all an issue, or if you are working on dynamic or somewhat closed systems, where some of the Stuff happened far away and long ago. unsafeExtract :: [Stuff] -> Summary Stuff -- | safely traverse some Stuff to extract information. -- -- Enforces on-the-fly compression of lots of Stuff into a hopefully snappier Summary. safeExtract :: (Traversable T) => T Stuff -> (Stuff -> Summary Stuff -> Summary Stuff) -> Summary Stuff
If you muddle through an MHDS it becomes obvious that vast oceans of data have been collected shabbily and in haste. A record of historical summary has long been seen as a weakness in data science. An MHDS is low-level memory stripped of context and purpose. Machine learning needs to go live, becoming Summary State updates for other computations happening around them. It wouldn’t need as much data, would be much faster and more timely, and may prove more useful.
In this world, MHDSs would undergo
compression at source; take these large volume but ephemeral data streams and capture a current Summary, with the wider context and purpose to the best you can, so that the Summary can be improved over time. Let the rest go down the Sink.
The coding professions and craft have only a short window to recognise they’re on the wrong side of the carbon divide, are laggards as a profession in real climate action, and could even be the bad guys. Saving all the Stuff, assuming near-infinite computational capacity and confidence in high fidelity reproduction of the past maybe a good approach in the Sunlight Age, but we’re not there yet, and we may have to be more clever. We need to count our cycles and conserve our capacities, in the coming great carbon squeeze that we can now just feel.
Wolpert, David (2019). The stochastic thermodynamics of computation. Journal of Physics A: Mathematical and Theoretical, (), –. doi:10.1088/1751-8121/ab0850 ↩︎