Economists love scale. Start-ups love scaling up (and so do investors). Politicians love large-scale, global stuff. And in all things digital, economies of scale are assumed to be a common feature of markets – high fixed cost, low marginal cost, network effects, yada yada yada. BUT: There is a very important difference between a minimum scale for things to be viable (i.e. pay for their cost) and the maximum scale all those economists, entrepreneurs and politicians refer to. Scale comes with concentration comes with power. Good for those who have it, (mostly) bad for everyone else.
In order to allow for a more diverse internet and digital technologies in general, we need to break free from the massive scale that Big Tech companies have achieved. Unfortunately, that does not scale at all. We still need to do it, we just need to understand the task ahead of us. And: AI is the ultimate scaling machine, so it cannot be part of the solution. We cannot automate our way out of this, but need to manually unplug. The following explains this in some more detail.
Scale with everything
I started to think about scale again after talking with Paris Marx about Europe and digital sovereignty on his podcast and I realised I had not quite thought things through. We spoke about local initiatives to break free from Big Tech influence like the Dutch domain registry and I said something like “unfortunately, this doesn’t scale”, implicitly suggesting that it should scale. And then I thought twice and really was not sure about what I meant and, more importantly, what I should mean. And then I saw a random guy’s post on LinkedIn citing from an article in which I was cited alongside others; the guy used quotes from all other “experts” (who all happen to be male) but me (so this is just jealousy speaking ;)) because they all highlighted the challenges to scale in Europe – internal market barriers, fragmented and illiquid capital markets.
“Why don’t we have European Big Tech?” The question often implies this to be a failure, a shortcoming. But actually, it is also a feature: There is an inherent tension between local, diverse structures and big, global ones. When people talk about alternatives to Big Tech, they usually imply something big, something that scales. But that would lead down the very same path. Hence, it is important to engage in a careful weighing exercise.
The cost of scale
The problem with scale is that its benefits accrue to those that have scale (usually power translated into profits), while the cost is externalised to everyone else. As a big cloud infrastructure provider, companies can just decide to reinstate out-of-fashion energy sources such as nuclear or gas. On social networks, scale comes with context collapse whereby – as the name implies – the context of messages is lost when they go viral and/or are reproduced without sufficient context. At the same time, there is such an overload of content that people become dependent on algorithms whom they cannot trust (I still remember when Facebook was just a chronological feed with people I knew; in the meantime Facebook had to admit that when buying Instagram, they spaced out time between content from family and friends so as to increase the time spent on the app which increases ads which increases profitability).
Generative AI creates content at massive scale with minimum input. Let loose on the internet, it creates utter chaos as Jacob Silverman describes for the FT, “Navigating the chaos exacts its own toll. It breeds mistrust and inefficiency, a slowdown in the smooth movement of things as we find ourselves crossing the digital street to avoid another obstacle. It reduces attempts at genuine communication to a mere yelling into the void. We are faced, now, with a digital world defined by madness and hostility.” He also cites an article by Max Read in the NYMag, “This is the most widespread use yet found for generative-AI apps: creating stuff that can take up space and be counted. When you look through the reams of slop across the internet, AI seems less like a terrifying apocalyptic machine-god, ready to drag us into a new era of tech, and more like the apotheosis of the smartphone age — the perfect internet marketer’s tool, precision-built to serve the disposable, lowest-common-denominator demands of the infinite scroll.”
Emphasising human agency, putting human needs at the centre and ensuring democratic governance does not scale. It is also very hard work. But it is an effort worth making.
Scale is one-size-fits-all, it homogenises because it applies one recipe to everyone. It is the lowest common denominator (usually taking white men from the US as the default) and, at the same time, it also reduces the lowest common denominator: Because, with algorithmic personalisation, we no longer see what others see. It is not the scale of books, not of linear TV, but of a personalised experience determined by some invisible workings in the background. This applies to online content on social networks, but also to government workers who see generative AI being inserted into all the digital tools they use – because their software provider just decided to do so or because their manager was dying of FOMO.
The scale we want: scaling down is the opposite of efficiency
Once we acknowledge that scale is a trade-off, we need to figure out what level of scale makes sense. Of course, digital infrastructures must cover their costs somehow – but when looking at the current Big Tech quarterly earnings (Google, Facebook and Microsoft all have profit margins way above 30%!!), it is clear that they are doing far more than that, and environmental and other concerns also cast doubt on whether the current size of infrastructure is desirable at all (as also Goldman Sachs comments on AI overbuilding).
How does smaller scale look like? First, platforms do not have to be global. Mastodon proves the point with its distributed instances and its features that discourage virality (such as the so-far unavailable quote-post feature). For something like social networks or, say, streaming providers, mobility platforms, e-commerce providers and others, it makes sense to be linked to some kind of community, possibly defined by region, language or other aspects. Second, infrastructure control should not be global. Cloud is the main example right now with “cloudification” being underway across sectors. And, since “nobody ever got fired for buying AWS” (i.e. choosing big providers is the default for non-technical experts including managers), the biggest providers increase their scale. This is a dangerous, harmful path: I find Bert Hubert’s insights very valuable, and as he puts it, “‘Repatriating’ your workloads back onto infrastructure you control is going to be very hard work.” This will already be necessary in many instances. It will be very hard work.
To get to smaller scale, it will not be enough to just build alternatives that should have inbuilt mechanisms to balance scale with its costs. They should not flood the zone with useless AI features or content that can be sold to advertisers and keeps people glued to their screens. We also need to scale down the big players – by leaving them, by getting them to internalise the cost of scale, by regulating them, by breaking them up. This will make them less “efficient”, but that is precisely the point.
I hope I have made it clear that I am really not the first to think about these points, but I wanted to bring them together and link them to the economic debate where I personally started. I also very much appreciated tante’s take: “As people interested in a vibrant, humane, life-affirming social web I think we need to break the rut. And our very well built solutions are tying us down a bit too much I fear. When building for others we need to think less about “building the one right solution that everyone can use” but about building a thing that works for a specific set of people or that allows a community to actually change things in structural ways to serve their needs and wants, their practices and in-jokes.”
Emphasising human agency, putting human needs at the centre and ensuring democratic governance does not scale. It is also very hard work. But it is an effort worth making. And regarding those European market barriers: Yes, many of them may be mostly bureaucratic obstacles. But many of them may reflect differences in needs rooted in regional diversity and local histories. In these cases, scale means applying one-size-fits-all and applying one dominant model. Sufficient scale should be our aim.