Pre-preface: I’ve been doing some AI content grading for one of the big labs. The job is to detect AI slop, so that their AI produces less of it. The work is known as RLHF, which stands for Reinforcement Learning from Human Feedback. It’s giving me renewed appreciation for the quality of simple language, rather than quasi-intellectual and flowery rubbish.
Something that has amused me with this is how much quasi-intellectual jargon is used on Substack by writers who just write that way by default. Including, too often, me. No wonder AI ended up writing so horribly. It was trained on us.
Now I’m trying to wean myself off using phrases like ‘downstream of’ for things that come after something else, or ‘path dependent’ for how things ended up where they are. One of the hardest ones to relinquish is ‘failure mode’ to describe a common fault or weakness in something, but I’m going all in on this, so expect simple language from here on.
Language like ‘first-order effect’ (ie a direct consequence) and ‘instrumentalise’ (use something to get what you want) and ‘operationalise’ (start using) has a kind of superficially satisfying quality because it makes you look and feel cleverer than you necessarily are. It’s intellectual padding and I see way too much of it around here.
But I digress. Ultimately this piece aims to show that the frame through which people encounter AI can determine what they even recognise as relevant to discussions about it. And that is what makes it feel impossible to intelligently discuss with certain people. There will be a bit of jargon, but I’ll keep it simple.
Preface.
So this piece is about stories and ‘frames’. It’s an exploration of reactions to AI and how they shape popular conversation (too often annoyingly called ‘the discourse’), but the ideas transfer to other areas of controversy, so you don’t necessarily need an interest in AI to get something out of it.
What sparked my thoughts about all this was remembering two confident predictions from 20 years ago that were clearly motivated by sentiment of the same kind I see today on Substack, from what I see as a kind of anti-AI Taliban. One claim was that Amazon would never turn a profit. The other - commonplace in my own professional field around the same time - was that no one would ever buy a secondhand car, unseen, from a website. I remember thinking at the time that predictions like that were based on fear and motivated reasoning among people who felt threatened by the expansion of e-commerce. The structure of the arguments then and now - and how the ‘evidence’ to support them was cherry-picked - are sometimes similar.
This will not be a why they’re all wrong about AI argument. I’m agnostic on that score. ‘They’ might all be right that AI will ultimately produce nothing of real value to humanity and maybe even cause a lot of harm. I expect that it will play out in currently unexpected ways, but the point is that I’m not interested in that argument. What I’m interested in is the shape this discussion seems to take and what might be going on, structurally, in the minds of people who are clearly unhappy about the prospects of AI influencing their lives and the world. Specifically I’m interesting in the holes that seem to exist in their reasoning.
That’s why it’s about stories and frames, rather than here’s some evidence about why AI is so great / terrible. And I’m focusing on the anti-AI side deliberately, because it appears to be far noisier in my world - such as on Substack and in everyday social interactions - than the AI-enthusiast segment of amateur and professional punditry. That’s Rarely Certain’s schtick. The squeaky wheel gets the grease. The noisiest tribe gets the attention from this blog.
The perspectives below are transferable to other domains. So if you’re not interested in AI it doesn’t matter. The same dynamics play out around all hot button topics. What I hope is that you’ll read some of this and think ah, yes, maybe that also helps to explain why I can’t talk to uncle Roger about immigration and transgenderism or some of my friends about Gaza without coming away thinking they’re just too stupid to see what’s obvious to me.
Terminology you’ll see in this piece.
Yes, sometimes jargon is useful and I’ll be using some.
Frames
A frame can be pictured as a mental structure comprising assumptions that you use to interpret incoming data in a specific context, which you then use to make sense of the world.
Having a frame in place strips raw facts of their neutrality and warps them into a form that makes them useful for maintaining the frame. This makes frames like a feedback loop, or a self-reinforcing system, and we all have them. If you look at the classic cognitive biases that people talk about, many of them are mainly the manifestation of frames. You might say that biases are generally just foundations for the assumptions that make up your frame on any particular subject.
One of the reasons why frames are especially influential on our thoughts is that they determine what even counts as relevant data.
There’s nothing sophisticated about this concept of frames. You walk into a cafe and the waiter doesn’t smile or greet you. You think he’s rude. After a while you overhear him telling a colleague that he’s reeling from some terrible news. His behaviour no longer seems rude. Nothing about his behaviour changed. It was the frame through which you interpreted it that changed.
Paradigms
Another handy bit of jargon.
People often talk about paradigms too. These are a ‘bigger picture’ version of frames. Paradigms are more durable than frames because they come first. A helpful way to see the difference is paradigm as overarching worldview and frame as a mode of interpreting a specific set of circumstances or data that’s consistent with the paradigm.
Paradigm is a posh word but again it’s nothing particularly sophisticated. You see someone giving money to a homeless person. You think that people are fundamentally responsible for their own circumstances while the other person thinks that people’s lives are largely shaped by forces beyond our control and that this is unjust. You see a reward or an incentive offered to an irresponsible person while the money-giver sees themselves as acting in compassion and a righteous desire to relieve misfortune.
Note that you might both be ‘morally good’ people, in the everyday use of that word. Both of you would prefer that the homeless person was doing better. But your different paradigms shape how you think this is best achieved. You might think that relieving their immediate crisis creates an incentive to continue begging. The other person might think that immediate relief is a no-brainer and that more of it should be done. This is a big part of why economic conservatives are branded as heartless and socialist-leaning people are considered unrealistic idealists. Both sides actually want the homeless person to be better off. Where they disagree is on the how. Growing up during Britain’s Thatcher economic revolution I thought that the Conservatives didn’t care about poverty. My interpretation of every decision they took was through a simple paradigm which assumed the moral superiority of instant compassion and relief of immediate suffering. I had very student politics then.
Incidentally, this is why I argued previously that ‘values’ are unhelpful in explaining the differences that sometimes come between us.
Last year I wrote about paradigms as a helpful way to recognise the roots of some disagreements. I just lifted the paywall on this piece, which explains how we often find ourselves talking past each other. It seems under-recognised as one basis for common forms of intellectual/moral conflict.
We're mostly divided by ethical paradigms, not value differences
Housekeeping: everything is paywalled here, these days. But you don't have to be a permanent subscriber to read past the break. You can donate on a per-month basis via Buy Me A Coffee where you also have the option to offer ongoing support. One coffee lifts the paywall for a month, scaled pro rata. If you…
One thing I would add today to that account of frames is to the metaphor of glasses which filter out certain objects or perspectives. Frames also seem like armour, off which incoming evidence bounces.
Paradigms are more fixed than frames because they establish how you think the world works, rather than what happened in a specific moment.
Frames and paradigms support the next thing that shapes how we perceive events.
Stories
Common features in Stories are often called ‘narrative’ but I’ve come to roll my eyes at that word, now that it’s so popular.
So, by ‘story’ I mean an interpretation of events and facts that are in relation such that they produce a comfortable and emotionally satisfying sense of order. It’s worth noting that one might easily mistake something that’s intellectually satisfying as different from a story. But intellectual satisfaction is also an emotional affect, so emotionally satisfying covers such cases too. Obviously, a story need not be true, as we all learn from an early age. It just has to have a recognisable structure and components like protagonist, hero,and things happening in a progression that forms a beginning-to-end arc.
Stories can be so powerful in our minds that challenging them can carry bad social consequences. Take the story that women are treated unjustly in payment for similar work, or in this case, sport.
It’s hard to argue with many propositions that have taken root as stories, because introducing a new category of thing like evidence represents a transgression of widely accepted norms. In progressive circles you are expected to just believe that capitalism is bad and leaves women under-rewarded compared with men, for comparable effort. Try suggesting in polite society that if women really are cheaper in the workplace, all efficient capitalist enterprises would hire no one but women, and you’ll see what I mean. I’m not saying that’s a slam dunk argument, but it’s an interesting one to confront.
Now to why any of this matters. It illustrates the points I just discussed, while providing a bit of light relief, because it’s so ridiculous.
Here’s a story that’s been doing the rounds in France recently. It’s a fictitious story presented and typically received as fact.
Celine Dion recently gave some concert performances, much of which were mimed to audio playback, but media and social media were awash with her ‘triumph’ because she’d returned after being off the music scene with a serious illness. Reactions were so fervent that one journalist posted to Facebook that she would one day be telling her grandchildren about the moment Celine sustained a particular famous vocal crescendo. When you watch this clip you’ll see Celine also affecting relief at having managed this feat.
This is a story. She wasn’t singing live here.
But the way it was presented - and received - became part of the story arc of Celine’s deservedly respected and loved musical history, her illness and her reappearance in public.
