One of the most thought-provoking things that I’ve listened to over the past few months has been Derek Thompson’s discussion with Professor Agnes Callard from the University of Chicago about her concept of the ‘uni-context’. Derek describes this as a ‘one word theory for why the world feels so weird’, and it certainly feels like it underpins a lot of the stranger traits of the modern world.
The uni-context builds from the idea of ‘context collapse’ (ref Michael Wesch, Danah Boyd, Alice Marwick), which can be described as the flattening of different social circles and audiences into a single digital space. Post something online and it can simultaneously reach many different types of people you know from close friends to casual acquaintances, from family to someone we met at an event three years ago, from someone who is connected with someone we know online to our boss. Offline we may well act very differently when in the company of these different people or groups but online this collapses into a single context.
As a philosopher, Agnes Callard takes this idea much further. Starting with the word context, which we might define as a set of circumstances that tell us how we ought to act. For almost all of human history those circumstances may have been abundant but they were local, and it was the people there and the physical setting (a church, a bar, our own house) that determined how we should act, with barely any conscious effort. The uni-context is the condition in which that stops working because one set of norms now applies everywhere, regardless of where you happen to be.
She illustrates this with her own classroom. If a student turns up late, or interrupts someone, or raises their hand to say something unconnected with what’s being talked about, she can correct them. But if a student says something racist every other person in the room is in exactly the same position she is. Those norms are not role-dependent, but apply in an abundance of different situations. Once things sit in a single evaluative field they become comparable, and once they become comparable there is a reason to compare them. And sustained comparison pulls the things being compared towards each other. Social media, as an example, invites comparison but also placing all posts into the same feed with the same counters attached and they all start to feel reminiscent of the same techniques and patterns.
One of Callard’s key observations is that ‘goodness’ is much more context-dependent that ‘badness’. There’s almost nothing that is good for everyone all of the time, and happiness depends largely on who you are and where you are. But there are a small set of things that look and feel bad from whichever vantage point you’re looking at them from. Death, pain, violence, illness – you don’t need to know anything about someone’s particular circumstances to know that this will be bad for them. This means that in a space where everyone is present at once (like social media again) the things that spread are the things that need no context to be understood, and those things tend often to be negative.
There’s a parallel here with what happens in organisations. Positive local stories often stay local and when we try and scale good practice it can easily be reduced into formulaic playbooks. Which hints at Callard’s subsequent point about how sustained comparison leads to homogenisation. She has a nice example of this – two schools in two districts, each doing things slightly differently. If you live in district A you go to school A, and if you live in District B it’s school B for you. Then the system opens up so that children from both districts can go to either school and suddenly parents are comparing the schools – results, ratings, university places and so on. One school had been running a slightly more individualised curriculum, but when they start losing pupils to the other school they adapt their curriculum to be more like the other one so they can compete, and the two schools converge over time. Callard notes how this outcome was not inevitable – they could each have specialised, one taking younger pupils and one taking older for example – effectively rebuilding the boundary or differentiation between them. But without a new normative barrier of some kind, comparison will eventually converge.
There’s another relevant distinction that Callard mentions – that between character and identity. Character, she says, is a set of dispositions that shape how someone navigates their emotional life across a wide variety of circumstances. It shows up differently in different situations which can make it hard to read. An irascible person isn’t angry all the time, and in a situation where everyone is angry their irascibility is much less obvious because they don’t stand out. So in order to understand someone’s character you have to have known them for a while and to have observed them in a wide range of situations so that you can generalise effectively. Identity however, works in the opposite way. You are a member of an identity category regardless of the situation (it’s a hat you never take off) which means that identity is very suited to a uni-context world but character is not.
Back to the organisational parallel and I think this is why values, purpose statements, and brand missions all start to sound the same (the Wankernomics comedians did a brilliant take on this) – they are identity, not character. But culture in any sense of the word that is worth having is about character, an expression of what is unique about the company, not what it shares with so many others. Perhaps this is why organisational culture so often disappoints. We import identity when the organisations with the most distinctive cultures are often the ones that are least articulate about it. Character resists summary. But over time, with things like Glassdoor, LinkedIn, engagement scores and best-place-to-work rankings, employer culture has become comparable, and every measure that makes something comparable also makes it converge.
Which brings us to AI, the latest and most efficient commensuration technology, and the question of whether it is additive to the uni-context, or whether it actually helps us deal with it. Strategy, of all the things that an organisation does, is probably one of the things that is most context dependent. A good strategy is a set of choices that take account of the organisation’s unique position, history, capabilities and constraints in a way that it wouldn’t make sense if it was applied anywhere else. Strategy is character, not identity. So it matters that when you ask an LLM for strategic inputs the answers come from the ‘universal room’. In HBR earlier this year Angelo Romasanta, Llewellyn Thomas and Natalia Levina published research showing what LLMs actually recommend when asked for strategy. Across seven core business tensions the models almost uniformly favoured differentiation over commoditisation, augmentation over automation, and long-term over short-term, regardless of the company context that was provided. The researchers called it ‘strategy trendslop’. And it was notable that attempts to correct it through better prompting and richer context had minimal impact on the bias.
This echoes work done by Jon Kleinberg and Manish Raghavan on what they call ‘algorithmic monoculture’. They found that when many decision-makers rely on the same algorithm the overall quality of decisions across all of them can fall, even when that algorithm is more accurate than any alternative available to them individually. It makes you wonder what could happen to brand differentiation in a category when everyone is running a similar analysis using similar models and potentially getting similar outputs.
All of this is to say that this makes the role of the human strategist in bringing character and local colour to the strategic process essential, but this is not an argument for using AI less. Nathanael Jo and Raghavan more recently ran a pre-registered trial in which participants rewarded for their originality relative to their peers produced collectively more diverse work than those rewarded for quality alone. Participants that produced diverse results didn’t abandon the use of AI but they did take fewer of its suggestions verbatim, and they used it more selectively to help them come up with more ideas, to help with proofreading and targeted edits. The pull of homogenisation dissipated when the tools were used with more deliberate consideration. This, again, is why humans need to stay the driver in strategic processes.
AI is simultaneously the most powerful localising tool and commensurating one that has ever been built. The risk to strategy is that the second one wins because it’s the one that we become dependent on without thinking. It doesn’t have to. The local good, the character, the distinctive practice are valuable for exactly the reason that they are difficult – they were never portable, and that was the point of them.
A version of this post appeared on my weekly Substack of AI and digital trends, and transformation insights. To join our community of over thirteen thousand subscribers you can sign up to that here.
To get posts like this delivered straight to your inbox, drop your email into the box below.
Photo by Katie Mont on Unsplash
Leave a Reply