[yummyfajitas]

I'm not taking any normative position. I'm simply pointing out that the implicit assumption underlying lots of modern beliefs - that racism/other evil beliefs lead to factually wrong beliefs - is being challenged by "racist" and "sexist" machine learning algorithms.

If you want to make normative arguments, go ahead. My first principles tend to be very individualistic (I view individual humans as being the sole carriers of moral consideration), so our normative claims will likely disagree.

Also, describing the results of any code/program written by a human as a "completely inhuman intelligence" is a tenuous claim at best.

Clearly you've never written such systems. If they behaved remotely the way humans think my job (building them and making them usable by humans) would be vastly easier.

[rmxt]

I understand the difference between a normative claim and a positive claim. You might not be taking a normative position explicitly, but your distaste is showing through: "allegedly come from," "obfuscate this point". I'd like to see where you go with this (both your distaste and positive statements) -- even if what you're saying is accurate/factual, what implications does that have for society at large? For the intersection of machine learning and society?

Appeals to authority and accomplishments aside, I don't need to have written such systems to understand, infer, and conclude things about aspects of their behavior. My point is this: something created by humans cannot be, by definition, inhuman. Two methodologies, the "human approach" and the "ML approach", might have radically different steps but come to the same conclusions. It would appear from your comments that you are OK with these conclusions ("An unbiased methodology produced these results, therefore, it's OK!"). Are you morally satisfied by the conclusions discussed above? Do the results of "such systems" influence your satisfaction?

[yummyfajitas]

I have distaste for the anti-intellectual behavior/ideology underlying the NYT/Salon articles, and which is also visible in this comment thread. Doubly so since it's so dominant in our culture and since it is being used as a rhetorical weapon against tech.

The implication for society (assuming these findings generalize) is that most likely, we cannot solve statistical disparities via unbiased processes - we can have fair treatment of individuals or statistically representative distribution of spoils, but not both. As noted above, I'm very individualistic, so I favor fair treatment of individual humans.

Appeals to authority and accomplishments aside, I don't need to have written such systems to understand, infer, and conclude things about aspects of their behavior. My point is this: something created by humans cannot be, by definition, inhuman.

I don't know what you mean by "inhuman". It sounds like you mean the term to be "never tainted by the ephemeral emanations of humanity". I merely mean "inhuman" as "thought processes so radically different that intuitions about a human mind are completely useless".

Concretely, do you believe a random forest can somehow infer that the variable x[27] represents gender, and that to make it's sexist creator happy it should reduce the objective function in order to screw some women over? If you look at the internals of sklearn, that's just not what random forests do.

Two methodologies, the "human approach" and the "ML approach", might have radically different steps but come to the same conclusions. It would appear from your comments that you are OK with these conclusions ("An unbiased methodology produced these results, therefore, it's OK!"). Are you morally satisfied by the conclusions discussed above? Do the results of "such systems" influence your satisfaction?

I don't know what you mean by "morally satisfied". A fact about the world is either true or false. In computer science terms, I believe "morally satisfied" has type `satisfied: HumanAction -> Boolean`. Your question consists of applying `satisfied` to a value of type `WorldState` - it's a type error. In human terms, your question doesn't make sense.

In terms of my own individual happiness (as distinguished from moral satisfaction), this fact reduces my happiness. Because I believe many of these facts to be true, I'm forced to either lie about my beliefs (which causes me disutility) or suffer social opprobrium from anti-intellectual types and lazy thinkers influenced by them.

[AnimalMuppet]

But you can have distortions in the data, even if the algorithm is neutral, can't you? The data says "men are more likely to commit rape than women"; OK, that's probably not just that the data encodes a bias. But if your program says "blacks are more likely to be charged with violent crimes", say, is that because blacks are more likely to commit violent crimes, or because blacks are more likely to be charged with violent crimes because the justice system is (or historically has been) skewed?

Even an unbiased analysis system can reach bad conclusions from bad data, and a biased justice system can produce bad data. So the conclusions can be biased even if the program is unbiased.

[yummyfajitas]

Yes, I agree that not all learning algorithms are perfect.

Is it your belief that we can cook up better algos/data collection methods/etc and all the people complaining about "bias" in algorithms will be satisfied? I don't believe that is the case, given that no one is actually complaining that the algorithms are getting the (factually) wrong answer.

[AnimalMuppet]

My assertion was that a perfect algorithm, fed biased data, produces biased results. Given Ferguson, etc., data from the justice system may be biased, at least for some locations.

If the arresting officers are biased, the number of arrests are probably biased. I'm not sure that you can fix that with better data collection methods; the problem isn't with the data collection. The problem also isn't with the analysis.

It may be possible, at least in theory, to create an algorithm that will determine whether the officers are biased, or whether the race in question actually commits more crimes. I'm not optimistic about that working in reality, though.