5 ms·
Weapons of Math Destruction is good book on the topic. Using ML to do tasks like evaluate employee performance (and trigger firings), issue loans, insurance etc
by MonkeyMalarky 4y ago
Weapons of Math Destruction is good book on the topic. Using ML to do tasks like evaluate employee performance (and trigger firings), issue loans, insurance etc. is affecting peoples lives in the real world today.
- dekhn 4y agoInsurance has always been a statistically driven process and ML is just an advanced form of stats, so it makes perfect sense that the industry should use this technology.
- twobitshifter 4y agoI know of models that use things like typing speed and birth month to make loan decisions. Feed the AI inputs like race and it will make racist decisions.
- MonkeyMalarky 4y agoFeed a model zipcode information and you've got a pretty decent proxy for race too.
- dekhn 4y agoI am not aware of models that use birth month to make loan decisions, but if that's a useful predictor, why not include it? I'm not really sure exactly what you mean by "feed AI inputs like race and it will make racist decisions", as that's an absurdly general claim that doesn't even really say anything useful. If you mean the model reflects the underlying stats that correlate with race, class, gender, then sure, yes, I agree the predictor will do that. To say that it's racist is an entirely subjective claim. THe point of insurance isn't to avoid social ills, but to statistically minimize the impact of risk on individuals (while making a tidy profit for the insurer).
- PuppyTailWags 4y agoAn insurance model that uses data on financial stability in a society where a specific race of people have been, through bigotry alone, forced into a less stable position by default, will be bound to perpetuate the original bigoted patterns of behavior. It's not inaccurate to describe the model as therefore racist because it's modeling itself off a risk assessment where certain groups have historically been forced into the fringe and are therefore inherently risky. I'm sure if black wall street wasn't razed to the ground by a racist white mob (among several other attempts to gain stability, wealth, etc. by black people being destroyed by racist white anger) then maybe the model wouldn't need to "reflect the underlying stats that correlate with race...". But those underlying stats, and those correlations, didn't just happen in a vacuum, hewn out of the aether like a magical and consequentless thing.
- dekhn 4y agoThe points you are making are entirely political and apply generally, not just to insurance. Insurance is a business and it's not the job of the insurer to correct long-standing social ills.
- PuppyTailWags 4y agoWhether or not my points apply generally does not mean it does not apply to insurance. In fact, it very likely means insurance is a subset of generally, and therefore insurance also has to deal with racism. You cannot extricate models of society from their ills by pretending the ills aren't relevant.
- mrguyorama 4y agoIt's not political to point out bad data will make bad stats.
- dekhn 4y agoIs it bad data if it reflects the underlying observed frequencies regardless of their social origins? I don't think so.
- otabdeveloper4 4y ago> and birth month Holy cow, so literally a horoscope? Lol.
- terminal_d 4y agoI'll have you know that crypto astrologers exist.
- mattmcknight 4y agoI thought it was one of the most deceptive books I ever read! It contradicts itself enough that it is often obviously wrong even without conceptual knowledge of the field.
- WaffleIronMaker 4y agoCan you elaborate on this? I found it to be a very insightful read?