Timeline for Avoiding social discrimination in model building
Current License: CC BY-SA 4.0
51 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 17 at 16:21 | history | edited | Stephan Kolassa |
edited tags
|
|
| Jan 22, 2019 at 22:47 | comment | added | Alexis | @Eff "I'm pointing out that the book is not about mathematical models but is mostly a political/ideological/activist book." Got it, was not clear from the folks who were discussing it. | |
| Jan 22, 2019 at 21:38 | comment | added | Eff | @Alexis ...Ehmm.. what? I'm pointing out that the book is not about mathematical models but is mostly a political/ideological/activist book. It's not a book for learning mathematical models. If you think that all books about mathematical models are essentially books about political activism, then no, I don't agree. That's absurd. | |
| Jan 22, 2019 at 20:00 | comment | added | Alexis | @Eff All mathematical models embody worldview, and there is no 'value-free' mathematical model. But as I implied I have not read that book, just passing along a recommendation by math friends with explicit social values. | |
| Jan 22, 2019 at 13:05 | comment | added | seanv507 | I am no expert on the topic, but I would look at topic of causal inference, eg in 'Mostly harmless Econometrics'[mostlyharmlesseconometrics.com/] | |
| Jan 22, 2019 at 10:12 | comment | added | Lucas Morin | @seanv507: I'd like to hear more about that... I already have 8 answers and only one of them talk about controlling with respect to some variable (without providing enough methodology). | |
| Jan 22, 2019 at 8:54 | comment | added | Eff | @Alexis Nah.. That book is a political book first and foremost (and doesn't adequately deal with all the obvious critiques of the arguments she puts forth). It's a work of (biased) political activism. It is certainly not a book you should read if you want to know mathematical methods to avoid bias. | |
| Jan 22, 2019 at 7:04 | comment | added | seanv507 | I think this is covered by any applied statistics course in social sciences trying to determine causal effects... Identifying omitted variable bias etc, ' controlling for factor X' etc | |
| Jan 22, 2019 at 2:27 | comment | added | Alexis | Finally, Cathy O'Neil's Weapons of Math Destruction has been praised by people I know interested in the intersections of data science and justice. | |
| Jan 22, 2019 at 2:24 | comment | added | Alexis | Also relevant: An ethics checklist for data scientists | |
| Jan 22, 2019 at 2:23 | comment | added | Alexis | Relevant keynote address at the 2017 Neural Processing Systems conference by Kate Crawford on youtube The Trouble With Bias | |
| Jan 21, 2019 at 15:45 | comment | added | Lucas Morin | No the goal is not to do positive discrimination either. May I suggest you look at the accepted answer ? | |
| Jan 21, 2019 at 15:31 | comment | added | Lucas Morin | My question relate to the reduction of gender effect given two identical profile, not the origin of the historical bias. I am under the impression you are drifting appart the topic to express very personnal beliefs.Please refrain to do so. Or write an answer if you can link your toughts to the original subject. | |
| Jan 21, 2019 at 15:19 | comment | added | Lucas Morin | Or maybe there is a historical bias against women in the field and it make it more difficult to enter the field... I don't know honestly and this was not my question. Maybe the IT field was not a good exemple as it is historically skewed, but the Amazon exemple was the only one I had in mind. My domain of application is more balanced. Ps: I am not the one slapping the constraint. | |
| Jan 21, 2019 at 11:06 | comment | added | Lucas Morin | @Aksakal: why do you think it's not ? I don't see an obvious reason why there would be any significant difference in fitness between men and women in programming jobs. This kind of consideration have brought the autorities in my country to write the equality in rights in the constitution. Our representatives are currently writing laws to ensure equality in the professional world, in terms of recruitment, training, promotion, salary... etc. Why not do the same about jail you ask ? I'd say "politics", but this not in my hands and not the subject of my question. | |
| Jan 21, 2019 at 0:59 | comment | added | Aksakal | @Lcrmorin, why only physical? Why do you think the programmer is a best fit for both women and men? I don't see the obvious reason. I'm suspicious because only high paid jobs are in question. For instance ,men get jailed more, why not push to get women jailed at the same rate? A good definition of what constitutes a discrimination is paramount | |
| S Jan 20, 2019 at 12:19 | history | bounty ended | Lucas Morin | ||
| S Jan 20, 2019 at 12:19 | history | notice removed | Lucas Morin | ||
| Jan 20, 2019 at 12:18 | vote | accept | Lucas Morin | ||
| Jan 20, 2019 at 8:30 | answer | added | Carl | timeline score: 1 | |
| Jan 16, 2019 at 8:28 | comment | added | Lucas Morin | Plus if the model take into account the physical fitness, "absence of impact of the gender on the output of the model" would be a good definition. | |
| Jan 16, 2019 at 8:19 | comment | added | Lucas Morin | I can see your point with physical jobs but this is not the case here. | |
| Jan 16, 2019 at 3:26 | comment | added | Aksakal | "absence of impact of the gender on the output of the model" - I don't see why is this a good definition. Suppose, that you don't have a feature called gender in the model, which it probably doesn't anyway. Then model "figures out" by looking at other attributed that it's a woman, and downgrades because a woman is less fit for the job then what? There are many jobs where women would be poor fit, e.g. marines. It doesn't mean you can;t find a woman that would be great, but most of them won't fit. | |
| Jan 16, 2019 at 2:59 | answer | added | Ben | timeline score: 5 | |
| Jan 15, 2019 at 16:35 | comment | added | Lucas Morin | Plus I don't believe all what I read on the internet. | |
| Jan 15, 2019 at 16:35 | comment | added | Lucas Morin | I tried to avoid this issue by giving an edulcorated definition of fairness (absence of impact of the gender on the output of the model), instead of the whole fairness ordeal. This simple definition is rather practical and fit some big ethical considerations (ex. the 1st article of the constitution of my country newly state that "it shall ensure the equality of all citizens before the law, without distinction of [...] sex"). I felt that this simple definition of fairness would help get more answer than more general considerations about fairness. | |
| Jan 14, 2019 at 19:05 | answer | added | Dave Harris | timeline score: 8 | |
| Jan 14, 2019 at 18:45 | answer | added | Alex R. | timeline score: 7 | |
| Jan 14, 2019 at 8:49 | answer | added | Tim | timeline score: 5 | |
| Jan 13, 2019 at 17:49 | comment | added | gung - Reinstate Monica | Well, the article quotes a top AI researcher at Carnegie Mellon who says that the field is a long way from being able to ensure that algorithms are fair. Your question is how to ensure that the algorithm is fair. So, to me, that implies this may not be answerable right now. | |
| Jan 13, 2019 at 17:44 | answer | added | Karolis Koncevičius | timeline score: 4 | |
| Jan 13, 2019 at 17:21 | comment | added | Lucas Morin | How would i know if the question is not answerable ? I was under the impression that this kind of questions have already been delt with. But it is quite possible that there is no canonical answer yet or in general. I can accept and reward any answer that would provide a convincing reason why it is not possible yet and in general... | |
| Jan 13, 2019 at 16:55 | comment | added | gung - Reinstate Monica | What about the possibility that a canonical answer is not possible here? Can you address that? I'm considering reclosing. | |
| Jan 13, 2019 at 16:32 | comment | added | gung - Reinstate Monica | I reopened, & edited, but in the process of providing the necessary context in the body of the Q, I notice that a leading computer scientist from Carnegie Mellon says that, "How to ensure that the algorithm is fair [ie, your question], [is] still quite far off". That makes me worry that this isn't really going to be answerable here. In light of that, you might want to clarify your question further. | |
| Jan 13, 2019 at 16:24 | history | edited | gung - Reinstate Monica | CC BY-SA 4.0 |
quoted article for context; formatted; removed thanks
|
| S Jan 13, 2019 at 16:21 | history | bounty started | Lucas Morin | ||
| S Jan 13, 2019 at 16:21 | history | notice added | Lucas Morin | Canonical answer required | |
| Jan 13, 2019 at 16:02 | history | reopened |
Juho Kokkala kjetil b halvorsen♦ dsaxton Lucas Morin gung - Reinstate Monica |
||
| Jan 11, 2019 at 11:15 | history | edited | Lucas Morin | CC BY-SA 4.0 |
Added amazon story link, reworded the questions for clarity.
|
| Jan 8, 2019 at 17:35 | review | Reopen votes | |||
| Jan 9, 2019 at 10:36 | |||||
| Jan 8, 2019 at 17:16 | history | edited | Lucas Morin | CC BY-SA 4.0 |
greatly narrowed my question
|
| Jan 8, 2019 at 11:21 | history | closed |
Xi'an mdewey Peter Flom |
Needs more focus | |
| Jan 8, 2019 at 10:40 | history | edited | Lucas Morin | CC BY-SA 4.0 |
added 102 characters in body
|
| Jan 8, 2019 at 10:20 | history | edited | Lucas Morin | CC BY-SA 4.0 |
added 123 characters in body
|
| Jan 8, 2019 at 10:02 | history | edited | Lucas Morin | CC BY-SA 4.0 |
added 256 characters in body
|
| Jan 8, 2019 at 9:37 | comment | added | Lucas Morin | The story have been reported by reuter without much detail about the recruiting tool more than " It penalized resumes that included the word "women's," as in "women's chess club captain."" to provide a 1-5 star mark. Amazon obvisouly declined to comment. Here is a transcript : reuters.com/article/amazoncom-jobs-automation/… I didn't include the link on purpose as the matter is sensitive and I wanted to keep it general and avoid sterile discussions. | |
| Jan 8, 2019 at 7:25 | review | Close votes | |||
| Jan 8, 2019 at 11:21 | |||||
| Jan 8, 2019 at 3:01 | history | tweeted | twitter.com/StackStats/status/1082472226010615808 | ||
| Jan 8, 2019 at 2:59 | answer | added | dsaxton | timeline score: 9 | |
| Jan 8, 2019 at 1:30 | comment | added | StatsStudent | I think a reference to an article about Amazon's supposed recruitment scandal would be important to have in the body of you question. Some could argue there was no "discrimination" (depending on how the term is defined) at all but a simple imbalance in men vs. women hired, so maybe provide your definition of "discrimination." | |
| Jan 7, 2019 at 23:52 | history | asked | Lucas Morin | CC BY-SA 4.0 |