ML and data science: read about the dark side

Just read Cathy O’Neil’s “Weapons of math destruction”. Highly recommended to all aficionados of ML and DS. A sobering read… Uncannily prescient (read the bits about the manipulation of elections).

1 Like

Nevertheless, the impact of big data in politics is still eclipsed by that of big money.

1 Like

Her point is that big data is subservient to big money.

They lobby on both sides, just in case.

  1. From a ML perspective(ensemble models), one person(a weak model) one vote with same weight is not a very good choice for meta learning, if your goal is to maximize the effectiveness of the voting result.

  2. Electoral College give some weight biases to each vote, but doesn’t change the fact that this biase is not tilting the system towards or against the optimization target. So having it or not don’t really matters.

  3. From non-ML perspective, having non professionals to decide professional things can have catastrophic result, that’s why for most real practical problems, we don’t use demarcrocy to solve it.

  4. Demarcrocy works when most people having commen sense of hating something, which is not the case when the difference of 2 candidates are only 1% difference. In such case, the democracy system don’t really have real effect, they just end up chosing a random one.

  5. The real effective factor for a country to develop, is how much power/resources does the professional have to solve real problems. I don’t think any social system design is actually optimizing for this goal, or any goal.
    Demarcrocy when it works, provide a lower bounds.
    The free market is some kinds of evolution algorithm, which is working but in a very slow way.
    We really need a backprop for social science.

1 Like

Democracy is a sham. Cogitate on that before worrying about election integrity.

1 Like