3 ms·

Algos may fail to reflect reality. That's easy to measure. But insofar as they don't, there is no way reason to believe their errors go in one direction or the

Algos may fail to reflect reality. That's easy to measure. But insofar as they don't, there is no way reason to believe their errors go in one direction or the other. For example, in our supposedly anti-black society, test scores are biased in favor of blacks (and against women). The existence of unknowns simply does not support a case against algorithms or formalism - informal decision-making has the same unknowns and more (namely you can't even know the model). In short, you are pulling the postmodern shuffle - demonstrating that there are unknowns, and assuming those unknowns support your position.

dmreedy 10y ago
I don't think it's always so easy to measure as you're asserting; to really measure it, you have to have some idea of what the reality is, which is the whole problem in the first place. You can get some of the way there by approximating, but again, that's begging the question. The existence of unknowns absolutely does not support a case -against- formalism, agreed, but it does support a case against the absolute support of formalism. I agree that informal decision-making is just as bad (and often worse) but it has the quasi-benefit (I should really say, 'comforting effect') of a massive, historied infrastructure of ethics and judgement/punishment to make it look like it's fixing itself when a problem happens. Trying to avoid looking like I'm defending that system; it's mostly clear that it's pretty bad as well. And obviously all the different subsystems disagree as well, but they are somewhat accommodating of disagreement. So I'll agree half-way that I'm pulling a postmodern shuffle, but I don't want it to come off that I'm asserting that unknowns support my position. I'm trying to take the weaker position that the existence of unknowns makes it currently impossible to objectively pick one position over the other, and that extended defense of either will eventually reduce to dogmatism. EDIT: Something else that's been nagging at me.... There's a further category of unknown that is unlearnable by pure signal-based statistical modeling. A purely statistical approach to such things as ethical decision making, even in a world with more consumable data than our own currently, would still not be able to solve that problem. I'm not sure how many functions of society fall into that category. My hunch is that it's a lot of them. I don't have the math for that, though.
- yummyfajitas 10y ago
  I believe that you call the "historied infrastructure of ethics and judgement" is the same thing as what I call "political power". It's basically a system by which most decisionmakers will act towards certain hidden goals even while pretending they favor other more palatable ones. And I stand by my claim that this threatens the extant power structures. Many of those power structures maintain power by pretending to serve one palatable goal (e.g. equal treatment) while actually serving a far less palatable one (e.g. quota systems). Concretely, I believe that if existing power structures actually implement the principles they claim, algorithms will support them. And if they don't, algorithms will threaten them. So far the latter seems to be what is happening.

dmreedy 10y ago

I don't think it's always so easy to measure as you're asserting; to really measure it, you have to have some idea of what the reality is, which is the whole problem in the first place. You can get some of the way there by approximating, but again, that's begging the question. The existence of unknowns absolutely does not support a case -against- formalism, agreed, but it does support a case against the absolute support of formalism. I agree that informal decision-making is just as bad (and often worse) but it has the quasi-benefit (I should really say, 'comforting effect') of a massive, historied infrastructure of ethics and judgement/punishment to make it look like it's fixing itself when a problem happens. Trying to avoid looking like I'm defending that system; it's mostly clear that it's pretty bad as well. And obviously all the different subsystems disagree as well, but they are somewhat accommodating of disagreement. So I'll agree half-way that I'm pulling a postmodern shuffle, but I don't want it to come off that I'm asserting that unknowns support my position. I'm trying to take the weaker position that the existence of unknowns makes it currently impossible to objectively pick one position over the other, and that extended defense of either will eventually reduce to dogmatism. EDIT: Something else that's been nagging at me.... There's a further category of unknown that is unlearnable by pure signal-based statistical modeling. A purely statistical approach to such things as ethical decision making, even in a world with more consumable data than our own currently, would still not be able to solve that problem. I'm not sure how many functions of society fall into that category. My hunch is that it's a lot of them. I don't have the math for that, though.

yummyfajitas 10y ago
I believe that you call the "historied infrastructure of ethics and judgement" is the same thing as what I call "political power". It's basically a system by which most decisionmakers will act towards certain hidden goals even while pretending they favor other more palatable ones. And I stand by my claim that this threatens the extant power structures. Many of those power structures maintain power by pretending to serve one palatable goal (e.g. equal treatment) while actually serving a far less palatable one (e.g. quota systems). Concretely, I believe that if existing power structures actually implement the principles they claim, algorithms will support them. And if they don't, algorithms will threaten them. So far the latter seems to be what is happening.