Vet Abstractions

Our brains are organized in substantial part by levels of abstraction. Brain layers go from those connected directly to outside, to layers that attend to fine details of both inferences and plans, to layers that attend to larger scale aggregates and distant sparsely-described things, up to layers that manage our most highest level summaries and plans. Construal level theory captures important elements of how these layers order our thinking.

A great deal of thinking well is about knowing when to think at what levels of abstraction. On the one hand, abstractions drop detail, and the reliability of inferences using them varies in complex ways with context. On the other hand, reasoning abstractly can be much faster and quicker, and can help us transfer understanding from better to less known cases via analogy. (Note: math isn’t really “abstract” in this sense; math inferences are quite precise and reliable, so dropping details causes few problems there.)

Folks like Socrates have made careers out of showing how easy it is to find contradictions in most people’s abstract reasoning. And many have long noted that the main way political reasoning goes wrong is over-reliance on simple abstractions. 

Most recent college grads are near a lifetime peak of their ability to write essays on abstract school-type topics, and are better than average at this. And yet those essays are pretty incoherent; even their teachers struggle to understand them. In the workplace, new graduates overestimate their ability to be productive using the abstractions they’ve learned at school. A few years later, their abstractions haven’t changed much, but the details they’ve learned have made them far more productive.

Also, new business ideas tend to be too abstract:

People often email me with big-sounding ideas (reinvent commerce, change the way people meet, transform scientific research), and the bigger the idea sounds, the less interested I am. Truly big ideas don't sound big initially. If someone reinvents commerce or changes the way people meet, it won't be because they set out to. It will be because they created something unexpectedly powerful that had that effect. - Paul Graham (more)

While in principle it is possible to rely too much or too little on abstractions, the above facts suggest to me that our educated classes, at least, trust their abstractions too much. Which makes sense as a signaling excess; we assign status in our world greatly according to abilities to understand and parrot school-taught abstractions. But we hand out status less for abilities to tell when abstract reasoning is sensible versus broken. So the educated overly trust school-taught abstractions. Even if their schooling has real benefits in terms of knowledge, flexibility, innovation, and mutual understanding. 

However, the academics who run schools are at least picky about which abstractions they accept. Abstractions proposed there are challenged, tested, and often refined or rejected. Outside of academia, in contrast, the well-educated form large worlds of informal conversation wherein new abstractions are often introduced, and subject to less testing. Arguments made there using abstractions are also less tested. 

In my lifetime I’ve enjoyed many such “fellow traveler” worlds of non-academic conversation, and my best one factor theory to explain the worlds I’ve liked best, such as “rationalists”, is that folks there have an unusually high taste for abstraction. As I too like to learn and consider diverse abstractions, I like to join such conversations. 

However, even though I’m far from being a typical academic, I’ve deeply internalized some academic values. And one of those is a wariness of unvetted abstractions. Thus my strongest advice for my fellow-traveler worlds of non-academic conversation is: vet your abstractions more. For example, this is my main criticism, which I’ve repeated often, of AI risk discussions. Don’t just accept proposed abstractions and applications. Yes, you are justifiably proud that you grok them and can write somewhat coherently using them, when others can’t. But that doesn’t mean you should trust them so much. 

So how can we vet abstractions? One can seek examples and test which abstract claims apply to them. One can look for patterns in examples and see which abstractions fit those. When abstractions are supposed to satisfy symmetries, one can check for those in claims and examples. When some abstractions subsume others, one can check applications using both. When related abstractions embody related insights or apply to the same things, one can check their inferences and examples against each other. 

But these are just a few examples of how to vet; there is no general answer. Different academic fields collect and teach different ways to test proposed abstractions in those fields. Many have whole integrated systems of abstractions, tests, and examples.

A key superpower of we polymaths is that, over a lifetime, we collect abstractions, tests, and examples from many fields. This is especially helpful in social science, where so many fields cover substantially overlapping territories. If you can’t yourself become an aged polymath soon, maybe give us a listen when we question abstractions and their applications. Remember: reality is actually concrete; it is only thoughts that are abstract. (Though yes, thoughts are real.)