Escape from Muddle Land
Escape from Muddle Land, Econospeak, Peter Dorman
Let’s get the up-or-down part of this review over with quickly: Escape from Model Land: How Mathematical Models Can Lead Us Astray and What We Can Do About It by Erica Thompson is a poorly written, mostly vacuous rumination on mathematical modeling, and you would do well to ignore it.
Now that that’s done, we can get on with the interesting aspect of this book, its adaptation of trendy radical subjectivism for the world of modeling and empirical analysis. The framework I’m referring to goes something like this: Each of us exists in our own bubble, a product of our experiences and perspectives. Our thoughts express this subjective world, and they are true in relation to it but false beyond its boundaries. This means no one has the right to speak for nor criticize anyone else. In some versions, bubbles can be shared among people with the same set of identities, but, as before, not across them. In some unspecified way, we will be happy and productive if we embrace the diversity of our incommensurable worlds and their corresponding truths. Oppression occurs when some privileged people think their bubbles are universal, the doctrine of false objectivity. We must be re-educated out of such a delusion.
This of course is a cartoon version, but I think it expresses the core of the cognitive bubble framework. Its adherents think it is very radical and liberating, and self-evidently correct. I won’t belabor the obvious contradiction between the no-objective-criteria-across-bubbles hypothesis and the claim that the cognitive bubble world is the one we all live in. The only other point I’ll make is that, true to their belief that each cognitive world is impervious to criticism from another, adherents never, and I mean never, acknowledge, much less grapple with, serious criticism of their worldview. Instead, they argue by authority: Author X, who is much admired by people like us, says thusly, and so we can use this insight as a basis for further analysis. “Argument” in this context tends to take the form of exemplary analogy: here is a good way to think about the topic at hand because something like it works in a situation that is analogous to it in some respects. Argument by analogy fits a subjectivist framework, since its salience derives from aha-ness, not the sort of reasoning or empirical evidence that depend on objective criteria.
So how can this framework be extended to the world of information sciences and mathematical modeling? Thompson’s insight is that each model can be thought of as existing within its own cognitive bubble, composed of the assumptions that structure it and the purposes it’s designed to serve. Each model is true within its own bubble, but we need to step outside them, into the world of social and cognitive diversity, to see their limitations and escape their claims to any broader truth or objectivity. And that’s sort of it. While (as you can see) Thompson didn’t persuade me of any of this, I think there’s a chance her book will be successful on its own terms: future writers of the radical subjectivist persuasion may cite her as the reason why we should all think about models in this way.
My own view is probably clear from the way I describe hers, but just to be complete, here is its own cartoon version: There really are better and worse models, based on criteria that apply across different social and intellectual divides. Our self-knowledge is imperfect, and others often understand things about us we don’t see. We benefit from their criticism. They can also represent us, sometimes better and sometimes worse than we would represent ourselves. Arguments that evade engagement with counterargument are generally weak and unreliable. Arguments based on reasoning and evidence are better than those based on some version of grokking, and those are the criteria we’ll need to use if we’re serious about positive social change. How much we share with one another, cognitively and otherwise, is not a matter for ex cathedra generalizations; it’s something we discover through interacting with others—or better, something we can create by building on what already connects us.
“How Mathematical Models Can Lead Us Astray”
The above caught my eye. What is interesting about that short phrase is the implementation of an AI Algorithm be Medicare Advantage to restrict, limit, or deny care to those who need such.
They measure their diagnosis against a pool of !6million people of varying ages. In which case, they determine an average and then apply it to present day patients. It go as far as denial of care.
Hump, “lead us astray . . .How far can it go? It has gone pretty far in this case.
Certainly models can be used for nefarious purposes, as in your example, and this is an area where I would probably agree with Thompson. Where I disagree is that this is really about the purposes, not the model as such. Evil insurers can do their dirty deeds haphazardly with poor models or hyper-efficiently with good ones. In a sense, the quality of the model is independent of the virtue of the purpose.
It’s important to make this distinction because virtuous types like you and me need effective models, and model effectiveness can be built upon precedents of all sorts (including models of insurance companies). If there was anything in Thompson’s book about model quality as such, I missed it.