The aim of frugal computing is to keep a liveable planet supporting an inclusive global society.
My Frugal Computing article is meanwhile five years old. A lot has happened in computing, and from the point of greenhouse gas emissions, all of it has been worse than expected.
It is hard to imagine that the notion of “Frugal Computing” predates the AI craze, but even without AI, back in 2020 emissions from computing were rising steeply. AI just made it infinitely worse.
For some reason, since its publication, several people have told me that frugal computing is all about optimising efficiency. Clearly, I did not communicate my vision well, as this is not at all what I had in mind.
What I wrote in my original position paper was:
Therefore, as a society we need to start treating computational resources as finite and precious, to be utilised only when necessary, and as frugally as possible.
So it is about sufficiency first: not to use more than needed.
I added:
And as computing scientists, we need to ensure that computing has the lowest possible energy consumption.
This is because inefficient computation means squandering of resources, so to be frugal also implies to compute efficiently. But crucially, what I am concerned with is the overall energy and resource expenditure, which includes that of manufacturing the devices and the entire supply chain it requires.
But efficiency on its own does not reduce energy use and therefore emissions: the entire history of the industrial revolution is one of efficiency improvement, and this is what led to the growth in emissions. Jevons already observed this in 1865, and this paradoxical increase of resource use as a result of efficiency gains bears his name. Closer to date, it was the increased efficiency of GPUs that enabled the commercialisation of generative AI, which is now leading to an unprecedented increase in material use, energy consumption and emissions of computing.
I concluded with
I would like to call this “frugal computing”: achieving the same results for less energy by being more frugal with our computing resources.
I see now that “achieving the same results for less energy” might muddle the waters somewhat. A better phrasing is “achieving the same results while using fewer computing resources, meaning less use of energy and materials, and therefore lower operational and embodied emissions.” I have now updated the wording in the original article to clarify this.
Let me give a few examples:
As a first example, instead of computing, let’s consider leaf clearing. The efficient way is to use a petrol leaf blower. The frugal way is not to optimise the energy efficiency of the leaf blower but to use a broom. It is arguably even more effective.
Next let’s consider some actual computing: calculate the rest after division by 9 of 301415904225571808. The frugal way is to recognise that you don’t need a computer or even a calculator for this at all: you can do it on paper or even in your head [1]. The result is the same. Not using a computer at all when there is no need, i.e. when you can obtain your desired result without it, is the ultimate frugal computing.
One can argue that most of the tasks for which we use a computer can often be done without a computer. So is there such a thing as “frugal computing”? The reductionist view would be that from a perspective of frugality, we should not have computers. But it is a matter of degree: given that we have computers, we need to use them frugally. We can recognise that this is not the most frugal way. But I am not advocating for frugality for its own sake. I am concerned with keeping a liveable planet supporting an inclusive global society. Business-as-usual, non-frugal computing already is harmful to society in many ways, of which its greenhouse gas emissions is only one aspect.
Profit, harm and usefulness
The consequence of advocating for frugality as a means to keep a liveable planet supporting an inclusive global society is that the term acquires an additional connotation: being frugal should not result in harm to our society, which includes harm to our environment. Until now I have taken this as read, but considering the twists put on the meaning of “frugal computing”, it appears it is important to spell this out.
Profit does not justify harm
The most common rationale for the adoption of a technology is that it will increase efficiency and productivity and therefore increase profits or at least reduce costs.
Increasing profit is not a morally acceptable justification of the use of any technology. The puirsuit of profit in itself is not a force for good in the world. By extension, neither is “saving costs”. At the very least, these pursuits should not cause additional harm.
For example, wars increase the profits of many companies (including the “AI” companies). Yet this is not a morally acceptable reason to start a war (whether there ever is such a reason is a different question). Closer to home, most governments have introduced laws that curb industrial pollution, even though the industries involved always argue that it cuts into their profits. Often the harm is more subtle, e.g. reducing the quality of a product to increase profits by limiting its lifespan and so forcing people to discard it and buy a replacement, or making hardware obsolete by redrawing software support for it.
Being useful is not sufficient
For any product or service, “being useful” is simply a necessary condition, and therefore the argument that something is useful is by itself meaningless. If it was useless it would not be used. We need to consider the sufficient conditions, which take into account the externalities of the use. When someone says that something is useful, they usually mean “useful for me”, and the fact that they need to make this claim is because the thing is worse than useless for others: it is harmful.
For example, for the person using it, a car is useful. But considering the extent of harm done by the widespread adoption of cars, it is clear that, to be sustainable, car ownership should be drastically reduced. Most people are aware of this, hence their need to justify why for them a car is useful, even necessary. If cars were entirely harm-free, there would be not beed for such a justification.
Conclusion
The aim of frugal computing is to keep a liveable planet supporting an inclusive global society. It is about using less energy and resources for computing, because the current trajectory is unsustainable. The key notion is therefore sufficiency. Efficiency is a means to reduce energy and resource use.
Coda: quid “frugal AI”?
The above discussion is regardless of the type of compute. Because of the dominance of “AI”-related compute, I discuss the notion of “frugal AI” in a separate article.
[1] Remove all zeroes and all groups of digits whose sums make 9. Do this until only one digit remains. So the result is 2.