📚 Quinn's Highlights

Join 📚 Quinn's Highlights

A batch of the best highlights from what Quinn's read, .

How Measurability/Mathematical Bias Limits the Scope of Scientific Inquiry and Human Discovery Transcript: Speaker 1 So there's this old paper from the, I think, 1960s by Eugene Vigner, the Nobel Prize physicist. It's called something like, on the unreasonable effectiveness of mathematics. The fun paper, and he's like, there's no good reason why mathematics should work as well as it does. And there's no good reason why there should be a tool that allows humans to predict things as well as math does. There's no good reason. It's kind of nuts. And we should all just be grateful. And he says some other things, but he's basically just kind of being all about how great mathematics is and how there's no good reason why it should be. And it's pretty cool that it does work so well. I think that there's a counter to that, which is that not everything is that easily described that mathematics. And there's lots of things for which mathematics is not that effective at describing. And it's actually just the things that were well described or easily described by mathematics are the things that were discovered using mathematical tools. They're the things that lend themselves that were amenable to mathematical inquiry. And a lot of the things that we're interested in terms of social science and cognitive science and the related philosophical inquiry are things that are much less tangible in terms Of this kind of specification. And you can see it like in a physics equation, right, a physical theory, whether it's about mass or electricity or something else, right, you have a theory about how things work. And then you can write out equations. And all the terms in the equations have units. And they are all directly related to the things that are measurable. The theories are directly about relationships between things that are measured. And in social theories and cognitive theories, so often our theories are about relating constructs. And then we have proxy measurements, but the theory isn't about the relationship between the proxy measures. The theory is about the constructs and the relationships between the constructs that are social in nature, that are cognitive in nature, but aren't the things that are being measured. And so there's this gap. And I don't know the extent to which that gap can be overcome.

Paul Smaldino & C. Thi Nguyen on Problems With Value Metrics & Governance at Scale

COMPLEXITY: Physics of Life

While Algorithmic Decision-Making Does Suffer From Bias, It Offers the Potential for Unparalleled Transparency In the Decision-Making Process Summary: Algorithms offer a transparent and accountable way for decision making. They can detect bias and perpetuated patterns, but must be transparent, independently audited, and not proprietary or snake oil. Transcript: Speaker 1 And then the response comes back saying yes but if you're basing it on historical data then you're feeding in biases of the past which you're going to propagate into the future there Is a kind of new attitude about all this which is kind of orthogonal to these two axes which I personally find pretty compelling and it's come up in from a couple of different places independently I could drop a few names but let me just say that the attitude is that algorithms at their best offer a new way for decision making to be transparent and accountable that's at their best So you know if an algorithm is something that everyone understands how it works everyone understands why we are chose to use this algorithm how it was trained and it's something which Can be independently audited it's even something which could be tinkered with to see if it could be made more fair and more accurate that kind of algorithm could raise the standard of Decision making in many areas and let us detect bias where it crops up and also help us detect where historical patterns are being perpetuated and what we might do to fix that but the big But is they have to be transparent they have to be independently audited they can't be proprietary and opaque and hidden behind veils of intellectual property and they also can't just Be snake oil right so there is a lot of snake oil out there there's a lot of products being put out to market which have not in any sense been independently verified or validated and where Their users and customers frankly don't really know whether their results ought to be interpreted the way they ought to be interpreted and so there needs to be a lot more critical thinking Aimed at these

Glen Weyl & Cris Moore on Plurality, Governance, and Decentralized Society

COMPLEXITY: Physics of Life

The Tension Between Organized Behavior at Scale and Individual Needs Summary: Large-scale organizations aim for legibility and coherence, but this may lead to a lack of diversity and individual needs. The educational system's emphasis on GPA overlooks other important skills and qualities. Transcript: Speaker 2 One of the most influential ideas for me recently has been from James South's book Seeing Like a State. And Scott has this idea that like what large-hill organizations wants its legibility and legibility is a kind of clear coherence that's aggregatable to a kind of higher level view. So a simple version might be like look if you're a CEO you can't have every department have its own obscure little value system. You need a single collective value system or something close to it so you can get production and profit measures and aggregate them in what Scott says is bring the whole organization Into view. So one way to put my worry is that what would be good for human life is an incredible diversity of bottlenecks which work on different often non-metrified systems. If Scott is right large-scale institutions will tend towards is a kind of monolithic measurement system that moves towards let's have a small number of bottlenecks and let's have A unified measure. And so like the heart of my worry is that organized behavior at scale is inevitably in tension with what a diverse population of individuals needs. And that's just an unfixable problem. Let me just give one quick example. In the educational system the dominant measure is GPA. You can add other like I can write in my notes all kinds of other shit about what students are good at. That barely matters because that's not aggregatable. When a law school admissions officer is doing their spreadsheet to do the first main cutoff nothing in my weird little notes is going to make it into that first level cutoff. The big moving forces just look at GPA.

Paul Smaldino & C. Thi Nguyen on Problems With Value Metrics & Governance at Scale

COMPLEXITY: Physics of Life

...catch up on these, and many more highlights

Readwise.

Join 📚 Quinn's Highlights

A batch of the best highlights from what Quinn's read, .

LEARN

EXPLORE

CONTACT