More on this book
Community
Kindle Notes & Highlights
might encourage opportunistic behavior,
might be necessary to prevent wrongdoing.
A system might tolerate noise as a way of producing extra deterrence.
people do not want to be treated as if they are mere things,
With these points in mind, our general conclusion is that even when the objections are given their due, noise reduction remains a worthy and even an urgent goal.
too expensive.
There is a legitimate concern here, but it is easily overstated, and it is often just an excuse.
perhaps the educator could make sure to read each essay at the same time of day, so as to reduce occasion noise.
We could easily imagine a limit on how much to invest in noise reduction.
In the case of divergent medical diagnoses, efforts to reduce noise have particular appeal; they might save lives.
an institution might think that elaborate corrective steps are not worth the effort. Sometimes that conclusion is shortsighted, self-serving, and wrong, even catastrophically so.
the belief that it is too expensive to reduce noise is not always wrong.
some noise-reduction efforts might themselves produce unacceptably high levels of error.
some efforts at noise reduction might even increase bias.
false positives are a directional error—a bias.
but some cures are worse than the disease.
three common objections to reform efforts.
aggravate the very problem they are inte...
This highlight has been truncated due to consecutive passage length restrictions.
might be ...
This highlight has been truncated due to consecutive passage length restrictions.
put other important values ...
This highlight has been truncated due to consecutive passage length restrictions.
Quoting Václav Havel, they insisted, “We have to abandon the arrogant belief that the world is merely a puzzle to be solved,
a machine with instructions for use waiting to be discovered, a body of information to be fed into a computer in the hope that, sooner or later, it will spit out a universal solution.”
Eliminating noise is the central point of the three-strikes legislation.
the price of this success is too high.
the price of that noise-reduction strategy is too high.
Supreme Court, a serious constitutional shortcoming of the mandatory death sentence is that it “treats all persons convicted of a designated offense not as uniquely individual human beings, but as members of a faceless, undifferentiated mass to be subjected to the blind infliction of the penalty of death.”
all these people might make mistakes if they apply overly rigid, noise-reducing rules.
noise-free scoring system that fails to take significant variables into account might be worse than reliance on (noisy) individual judgments.
Some noise-reduction strategies ensure too many mistakes.
help reduce noise without creating intolerably high costs (or bias).
introduce other forms of decision hygiene,
growing objections to “algorithmic bias.”
much of this book might be taken as an argument for greater reliance on algorithms, simply because they are noiseless.
In Weapons of Math Destruction, mathematician Cathy O’Neil urges that reliance on big data and decision by algorithm can embed prejudice, increase inequality, and threaten democracy itself.
these and other cases, algorithms could eliminate unwanted variability in judgment but also embed unacceptable bias.
algorithm could discriminate and, in that sense, turn out to be biased, even when it does not overtly use race and gender as predictors.
predictors that are highly correlated with race or gender.
discrimination could also come from the source data.
bias in the training data, it is quite possible to design, intentionally or unintentionally, an algorithm that encodes discrimination.
algorithms could be worse: since they eliminate noise, they could be more reliably biased than human judges.
Exactly how to test for disparate impact, and how to decide what constitutes discrimination, bias, or fairness for an algorithm, are surprisingly complex topics, well beyond the scope of this book.
same kind of scrutiny; people sometimes discriminate unconsciously and in ways that outside observers, including the legal system, cannot easily see.
criteria that matter: accuracy and noise reduction, and nondiscrimination and fairness.
of criteria we select. (Note that we said can and not will.)
producing less racial discrimination than human beings do.
predictive algorithm in an uncertain world is unlikely to be perfect, it can be far less imperfect than noisy and often-biased human judgment.
broader conclusions are simple and extend well beyond the topic of algorithms.
In that case we have a serious problem, but the solution is not to abandon noise-reduction efforts; it is to come up with better ones.
should ask is, can we design algorithms that are both noise-free and less biased?”
And if one effort to reduce noise is too crude—if we end up with guidelines or rules that
