Alexandria Ocasio-Cortez Is Absolutely Right About Racist Algorithms

What sort of critique are you trying to make of "this community?" Are the words "he is right" in our vocabulary?

Machine learning, as well as human learning, largely work off of correlations observed in the source data. In some cases, the data itself is a flawed source for the learning it's being used for, because human biases have already influenced the dataset. An example is arrest records, which is the result of discretionary decisions made by officers whether or not to pursue an arrest.

But in other cases, the source data is less in question, but the results of the learning are nonetheless undesirable. We ought to be able to distinguish, in our professions and our political lives, between the two scenarios. It's counterproductive to conflate the two to strive for a goal, because we risk using mechanisms that never discover and won't correct the true origin factors.

Ok. Lets assume this is true. (Note: I'm biased in that I accept this as true)

So, as someone who has possibly implemented racist/sexist/ageist algorithms, how do I:

     1. Detect if I'm running a said algorithm (whats the % racism/sexism/ageism I can do before bad?)
     2. Run an open dataset to detect said problems
     3. Prevent overfit with the proposed dataset from #2
     4. Correct said algorithm to reduce bias

What's my way forward here? How do I do my part and take part in the solution? What percent of unintended racisms/sexism/ageism is allowed before being considered illegal?

(worried this article will be flagged, but the nuts and bolts implementation discussion needs done.. and we're the implementers )

The second half of the article that talks about Yudkowsky, rationalists, and Roko's Basilisk is absolute trash and a gross mischaracterization of the actual positions taken.

>Yudkowski has for more than a decade pursued the possibility of perfect human reasoning

His website is literally called "Less Wrong". It's fundamentally a quest for improvement, not perfection.

>His system of coldly logical reason, it turned out, was by many accounts completely undone by a logical paradox known as Roko’s Basilisk.

Roko's Basilisk is a thought experiment designed to import your "don't negotiate with terrorists" intuitions into a really weird corner of decision theory. It's possibly why prior decision theories held were rejected as flawed, but it's not a current issue with their logic. Roko's Basilisk has, unfortunately, gotten way more coverage and fame than it deserves in terms of actual importance. This is because of the regretful (but understandable) decision to censor discussion of it on Less Wrong, which backfired spectacularly.

>For super-nerd bonus points, it’s also arguably a spin on Godel’s incompleteness theorem, which argues that no purely rational algorithmic system can completely and consistently model reality, or prove its own rationality.

That's not at all what Godel's incompleteness theorem argues. That's much more narrowly about formal logic.

The problem here is, we don't even know what "fair" means. What do we mean by fair? Say we want to base fair on group outcomes, in which ways should we divide people into groups in order to perform some sort of balancing? Is hair color or texture a relevant way of subdividing people for instance? Are we being "fair" to red haired people? What if red haired people for some reason seem to be less likely to pay back loans? Will we need to adjust our credit rating model to bump the credit scores of red haired people slightly? Will we need to agonize over other features fed to our model to ensure that they can't be used as a proxy for red hair?

Maybe???? But it seems very dishonest.

I think, if anything, the models are probably too true for our tastes as social creatures who evolved to avoid social conflict (being 100% honest with your unruly neighbors can lead to a loss of harmony that is much more damaging than just putting up with some disturbances). Our idea of "unbiasing" the models is more likely actually making them biased in a way that pleases our monkey brains.