SoylentNews is people
SoylentNews
At The Guardian, Cathy O'Neil writes about why algorithms can be wrong. She classifies the reasons into four categories on a spectrum ranging from unintential errors to outright malfeasance. As algorithms now make a large portion of the decisions affecting our lives, scrutiny is ever more important and she provides multiple examples in each category of their impact.
This discussion has been archived.
No new comments can be posted.
How Can We Stop Algorithms From Lying?
|
Log In/Create an Account
| Top
| 74 comments
| Search Discussion
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
For recreational use only.
(Score: 2) by HiThere on Monday July 17 2017, @04:42PM (2 children)
This isn't inherently true, though it sure often it. But one can train a neural net to achieve a goal, and not notice that lying is a part of how it achieves the goal. In that case it would be the program that was lying, not the implementer...though the implementer would certainly be rightfully said to have caused the error.
Then there's the problem of biased test data sets. You can't eliminate bias, even biases you don't have. You can only eliminate biases you think (and bother) to check for. But EVERY finite dataset is biased except for a finite number of cases. To give an example, every collection of coin flipping data is biased whenever it contains an odd number of entries...unless you allow things like "on edge", or "fell down the gutter". It *has* to be biased. (I'm here assuming that you are flipping a fair coin.)
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.
(Score: 2) by Immerman on Tuesday July 18 2017, @04:23PM (1 child)
Heck, you don't even need an odd number of entries. One of the fundamental assumptions of statistical analysis is that your sample is *representative* of the larger population - for coin tosses that's only true if you have exactly the same number of heads and tails in your sample. Any imbalance introduces bias, and its actually extremely unlikely that your sample will be balanced, and not terribly unlikely that it will be *very* unbalanced.
(Score: 2) by HiThere on Tuesday July 18 2017, @04:58PM
While I agree with your point, my point was that with an odd number of entries you can see by inspection that the data is biased...and without much inspection at all. And that's half of all cases.
Javascript is what you use to allow unknown third parties to run software you have no idea about on your computer.