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This discussion has been archived.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.
You will have domestic happiness and faithful friends.
(Score: 2) by Runaway1956 on Monday July 17 2017, @04:13PM (2 children)
If you tape a duck's bill shut, he can't quack. Use some of that duck tape on your algorithms!!
Seriously though - algorithms can't lie. People frequently use the algorithms incorrectly, but that doesn't mean the algorithms are wrong, or lying, or anything of the sort. Fall back to high school geometry. You need to figure the area of a rectangular shape, so you dust off the old geometry book. The damned book is full of formulas, we pick one at random: A=πr2 No matter how we apply the formula, we don't come up with the same answer that (teacher, customer, carpenter, whatever) came up with.
It takes a little bit of familiarity with the formula or algorithm, as well a little bit of intelligence to apply them. The formula isn't flawed, in my example, but the user is definitely flawed.
(Score: 3, Insightful) by AthanasiusKircher on Monday July 17 2017, @10:17PM
Or to put it more broadly: just because something is expressed in numbers doesn't mean it's more accurate or more "factual."
Seriously -- I think that's the underlying problem today. Lots of people are numerically illiterate (or nearly so), and they view certain people who "get numbers" as if they are magic. Anything spit out in terms of a number is often trusted more (and there are psychological studies to this effect).
But any number or data requires interpretation. Sure, there can be an actual error (unintentional or deliberate) in an algorithm, but whatever data comes out of it needs to be interpreted. And that's generally where the problem lies. It goes back to stuff as simple as a high school kid typing something into a calculator and writing down 10 sig figs spit out by the calculator as if that were a meaningful answer when the student was basing the calculation on empirical measurements with only one digit accuracy. But the electronic beast spit out numbers, so they must be correct and meaningful.
This is one of the challenges of the current "big data" revolution. There's a lot of data and a lot of calculations that could be done, but that doesn't mean most of the possible data produced is meaningful. For a while in the late 90s I was temporarily employed at a company that was obsessed with efficiency. For all of our tasks, we had to fill out giant spreadsheets quantifying what we did and how -- it seemed like we spent at least 1/4 of our time filling out "efficiency" spreadsheets rather than real work. Anyhow, it rapidly became clear to me that upper management was making decisions based on flawed interpretations of the data coming from those spreadsheets, stuff that would be clear to any idiot who spent an hour or two actually supervising workers. I was praised for stuff that was basically an error of spreadsheet interpretation and criticized for not meeting goals on my spreadsheet that didn't make sense given the details of my tasks. In the last couple weeks I worked there, they instituted a policy of "logging into phones" while at your desk, supposedly to track people's breaks and again ensure efficiency. Except people frequently forgot to log in and out, etc. One day I skipped lunch and took a slightly longer afternoon break (17 minutes or something instead of 10); the company netted a lot of extra time from me, because I was at my desk for all of my lunch time -- but my name ended up on a company-wide email list of "delinquent workers" the next morning sent to all supervisors, because I had exceeeded my afternoon break time.
I quit on the spot. I simply couldn't stand the insane "data-based" decisions of the company based on flawed interpretations of "efficiency data." The entire department I was working in folded about 3 months later, from what I heard due to gross inefficiencies.
Again, data is only as good as the interpretation. Just because you measured something or plugged it into a calculation or algorithm doesn't guarantee that it's meaningful or useful -- and it can certainly be very misleading (again, both intentionally and unintentionally).
(Score: 2) by KGIII on Monday July 17 2017, @10:23PM
Hmm... Here's a fun one.
0.999... = 1
"So long and thanks for all the fish."