Except the existence of laws against racial/age discrimination in housing and employment are not arbitrary. Those laws were passed in response to actual and widespread discriminatory behavior in the mid 20th century.
OK, maybe arbitrary isn't exactly the right word. I meant that there's no indisputable link between ethics/morality and law.
You say that they "were passed in response to actual and widespread discriminatory behavior". That's true. But they wouldn't have been passed, notwithstanding discrimination, without enough political support (of one sort or another).
I mean, there's also been discrimination in health insurance rates based on preexisting conditions. And gender-based discrimination in vehicle insurance rates. The Affordable Care Act more-or-less restricted the first. But the second is still the norm in the US.
There is plenty of ways to lawfully discriminate in this country. You can even still have discriminatory policies in employment if you can show it directly relates to the job.
> What non-arbitrary rule should we use as a society to determine if a product/service can be targeted or not?
Actual past experience with specific, widespread, and demonstrably harmful discriminatory practices.
Widespread discrimination in housing during the 20th century -- and the negative effects that had on certain communities -- resulted in laws prohibiting discrimination in housing ads.
Widespread discrimination in employment during the 20th century -- and the negative effects that had on certain communities -- resulted in laws prohibiting discrimination in employment ads.
BTW, these categories also make sense. Housing (i.e., schooling) and employment have a huge impact on your life outcomes in the USA. Choice of hair product, not so much.
It dodges the legal issue if you're very careful with your paper trail and get a sympathetic jury.
You won't get a sympathetic jury. People don't have a lot of empathy for algorithms. Maybe empathy for the people who write them, and then only maybe. But you'll be up against a huge slate of expert witnesses explaining how we already have lots of open sourced methods for teasing out these sorts of indirect indicators.
The paper trail won't matter if your "accidentally" discriminatory policies are not directly related to the position you are hiring for. It is the effect of the employment policies that matter under US law.
> The laws enforced by EEOC prohibit an employer or other covered entity from using neutral employment policies and practices that have a disproportionately negative effect on applicants or employees of a particular race, color, religion, sex (including gender identity, sexual orientation, and pregnancy), or national origin, or on an individual with a disability or class of individuals with disabilities, if the polices or practices at issue are not job-related and necessary to the operation of the business. [0]
And if you have an algorithm, that algorithm can be dragged into court itself. The prosecutor can show what happens when inputs are fed into it that are identical except for specific information (age, race, sex).
The ad must have discriminatory intent [1]. So if you're selecting candidates with a black box ML model and you didn't explicitly include racial/gender preferences...
But "the algorithm was sexist, not me" is probably going to be a losing argument in any court case.
[1] Well, not entirely, but that's another can of worms.
Yeah, I was asking what the requisite intent is. Is it The intent to knowingly discriminate, intent to do the thing that is discriminatory (a lesser standard), or merely a disparate impact (which does not require any particular state of mind)?
Yeah, I started off super into self-learning, even before the whole MOOC thing took off. I loved the MIT Open Courseware courses in high school, and I even really liked the central thesis of Illich's Deschooling Society.
Then I started tutoring and teaching. That's when I realized that I'm an extreme outlier. Most people are not particularly motivated and won't put in the hours upon hours of struggle.
The average experience of learning CS in university is just completely alien if you teach yourself how to program as a child. The biggest differences is the emotional labor. There are lots of things that my students would describe as "frustrating" that I have literally never thought of as frustrating (e.g., reading compiler errors). I think it's similar to learning a new language as an adult vs. as a small child.
A Ph.D.'s worth of first authored NeurIPS/ICML papers will get you a very good job pretty easily still. But AI slices papers very thin and author lists are inflated relative to other subfields of CS. A single paper in one of the major conferences is a pretty marginal contribution, especially if you're in the middle of a long author list.
Also, NeurIPS reviewing has gone to absolute hell. I mean, peer review everywhere has problems. But I've never seen something quite this bad. At this point I think it's safe to say that most reviewers wouldn't even make it to an on-site interview for a faculty position at a research university. That's definitely nowhere near normal. You can't really blame anyone, I guess; the community is growing way too quickly for any real quality control.
Frankly, I think those conferences have outlived their usefulness as anything except marquee marketing events. I'm now mostly attending smaller and more specialized conferences.
I've gone in a similar direction. Only at smaller conferences can you have any kind of confidence that your reviewers are people with actual expertise in the field. That's pretty useful, not only because it makes it less likely you'll get reviews that are very annoying, but also because a review by a knowledgeable person can be genuinely valuable. The big conferences are full of reviews written by 2nd-year grad students, because with this many submissions, any warm body with anything approaching credentials is needed.
Besides just "quality" in the general sense, one thing this has really hurt, I think, is any sense of history or continuity. There are a ton of reviewers who have basically no familiarity with the pre-2010 ML literature, and it kind of shows in both the reviews and the papers that get published. I mean I get that deep learning beats a lot of older methods on major benchmarks, but it's still not the case that literally every problem, controversy, and technique was first studied post-2010.
These "is college necessary" posts suffer the same problem as "why are programmers making 200K right out of school" posts. The job title covers everything from hacking out WordPress plugins, to leading a team building an MVP, to building large distributes systems, to designing the software that goes into medical devices and autonomous vehicles.
Some programmers really do need a CS degree with at least a few years of math and a couple years of physics. Sometimes programmers need to know a minimal amount of PHP and JavaScript. The job title captures a huge range of actual jobs.
> Sometimes programmers need to know a minimal amount of PHP and JavaScript
I'd say that covers the majority of programming jobs. The problem, though, is that the are a vast number of people who can meet that bar and salaries are bound to collapse to reflect that at some point. What happens to them then?
I don't know about that. I think salaries for web developers without some differentiating skill set have collapsed, or were never really that high to begin with. The average salary coming out of my undergrad's CS dept is about $60K.
And UNC is a top tier CS school, not even close to representative of "local branch campus".
My undergrad OS course didn't even have us implement malloc or free... CMU's sophomore systems course is a lot more rigorus, let alone their OS course.
And that's in systems courses where the variance is actually a lot smaller than in other subfields. There was simply no course in my undergraduate institutions curriculum where you could learn what a Turing machine or even DFA was! We never saw real asymptotic complexity analysis (our algorithms course spent most of the semester reviewing basic data structures as interview prep). Obviously no AI or ML course. Even where there was coverage the material covered was so different. For example, the PL course was all about using a functional PL and didn't include any of the type theory, logic programming, language implementation work, etc. It was a watered down version of SICP, which is an intro course.
If you think the difference in quality between UNC and CMU is large, consider the massive difference between branch campuses and top-ranked public research universities. For all practical purposes, "CS" means something very different at regional branch campuses than at top tier research universities like UNC and CMU.
> The CS program for undergrad in Havard is exactly the same as your closer public college.
No. Not even close. Unless you're taking about flagships like uiuc and Berkeley (which are not local for 99%).
This may be true in the humanities but is nowhere near true in the mathematical sciences.
My final year analysis course at college only covered a fraction of the content in the FIRST year advanced math sequence at Harvard.
As someone who went to one of those truly local state schools then taught at a #1 CS school: the quality difference in CS educations is absolutely insane. See also the "Java Schools" rant.
I searched through your comment history for the "Java schools rant" but to no avail. Pointer?
I went to school just as Java hit .. saw the old C days and then Java (in my OS course .. the horror!). I feel Java for OS stunted me to this day. Curious what your thoughts are on this.
3x to 4x depending on col. Grad student stipends are only 30k, sometimes lower
ML isn't much higher paid than standard engineering, but new PhDs from top schools who would've been competitive faculty applicants tend to get jobs at top paying firms and enter toward the top of the non-executive ranks.
