Home
> Uncategorized > The Era of Plausible Deniability in Big Data Continues
The Era of Plausible Deniability in Big Data Continues
October 16, 2018
Today I published a new Bloomberg Opinion piece on how Amazon’s sexist recruiting algorithm is not a surprise to anyone, but is framed as one because the tech bros are trying to maintain plausible deniability:
Amazon’s Gender-Biased Algorithm Is Not Alone
For my other Bloomberg pieces, go here.
Categories: Uncategorized
mathbabe 
I agree with your posted article and your video. I worked for many years as a innovations technologist & software architect in the clinical space where there has been a tendency to move from rule-based clinical software to probability based algorithms looking at retrospective data. The idea was to find computational approaches to automate the delivery of “evidence based medicine”.
So translating the models of the two approaches into human language we approached the first by saying if the patient is an with indications history etc… then we should be doing for this patient.
The second approach says “how well does this patient resemble patients along parameters ,,? If the alignment is > some value over some subset of patients, then do for this patient what was done for those patients.
Both approaches have weaknesses. The first is that determinative rules, even if perfectly current and perfectly sound, can become infinitely complex when confronting real patients all of whom are different.
The second approach has two weaknesses, the lack of sufficient parameters on which to make the decision (similar to the rules problem), and the second is that it propagates medicine as it was practiced in the past. And ironically, the deeper the data set, the more antiquated are the medical decisions.
So the older, more deterministic model was, in fact better than the probabilistic one. There’s an interesting tweak of rule based systems using deontic logic which has some potential in this space.
LikeLike