Working backwards from a PhD in ethics in AI
October 13, 2018 4:03 AM Subscribe
Say in 6-7 years, I want to do a PhD in AI ethics, basically how can we teach ethics to AIs, what ethics would AIs develop, etc. I have access to a university with a good AI research center that funds PhDs.
I need to complete my undergrad at another university for financial reasons then transfer to this university for my masters and PhD. I'm doing a science degree now.
What undergrad degree that isn't computer science or maths (I like my math papers but not enough to only do them) should I switch over for a masters/PhD in this field?
Assume I plan to take a coding course on my own separately (I do not think decade-old perl still counts).
I need to complete my undergrad at another university for financial reasons then transfer to this university for my masters and PhD. I'm doing a science degree now.
What undergrad degree that isn't computer science or maths (I like my math papers but not enough to only do them) should I switch over for a masters/PhD in this field?
Assume I plan to take a coding course on my own separately (I do not think decade-old perl still counts).
I have a friend who did a similar PhD and did his undergrad in Cognitive Science.
posted by ITheCosmos at 5:03 AM on October 13, 2018 [1 favorite]
posted by ITheCosmos at 5:03 AM on October 13, 2018 [1 favorite]
Digital humanities (if your university offers it). Or cultural studies if your university's cultural studies program has a strong focus on digital cultures.
posted by lollusc at 5:04 AM on October 13, 2018
posted by lollusc at 5:04 AM on October 13, 2018
Best answer: I’m not sure you’d be a very good masters or PhD applicant at an AI program at this university if your coding experience is minimal and separate from your normal undergrad curriculum. You say ‘transfer’ but you will be applying from scratch to a program that is likely fairly competitive. Most people in the sciences apply to several grad schools, hoping to get into one that will also make them a decent offer.
If you won’t major in CS or math, I think a major in philosophy with lots of documented coursework (ideally a minor) in CS or Math would be the way to go.
posted by SaltySalticid at 5:37 AM on October 13, 2018 [7 favorites]
If you won’t major in CS or math, I think a major in philosophy with lots of documented coursework (ideally a minor) in CS or Math would be the way to go.
posted by SaltySalticid at 5:37 AM on October 13, 2018 [7 favorites]
Contact the graduate program you’re interested in and ask what majors their admitted applicants have had. Way more useful than asking us.
posted by deludingmyself at 6:07 AM on October 13, 2018 [18 favorites]
posted by deludingmyself at 6:07 AM on October 13, 2018 [18 favorites]
Psychology? Ethics can not be taught *to* AI's, it may be possible to teach some to software developers and academic researchers. Have you read Weapons of Math Destruction: How Big Data Increases Inequality and Threatens Democracy? The author had a long running blog so would also be an approachable source for this question.
posted by sammyo at 6:21 AM on October 13, 2018 [2 favorites]
posted by sammyo at 6:21 AM on October 13, 2018 [2 favorites]
Economics.
During the last year several events i worked on brought together topics of AI, automation, and impact on the economy and the labour market.
posted by 15L06 at 6:29 AM on October 13, 2018 [1 favorite]
During the last year several events i worked on brought together topics of AI, automation, and impact on the economy and the labour market.
posted by 15L06 at 6:29 AM on October 13, 2018 [1 favorite]
Agree that "I took a coding course on my own" is not likely to make you a competitive grad student applicant for an AI program. What are the requirements for the graduate program you are interested in? What undergraduate coursework do they require? Further areas for inquiry: who is doing this kind of scholarship now and what is their academic background? Do you want to have a more technical focus or a more philosophical focus?
posted by jeoc at 6:44 AM on October 13, 2018 [4 favorites]
posted by jeoc at 6:44 AM on October 13, 2018 [4 favorites]
Best answer: If you could double major in philosophy and computer science, that would be perfect. But definitely take philosophy and try and take some formal philosophy courses if they're offered, or advanced logic courses (non-classical logics, modal and temporal logics) so you have the requisite technical background on your CV.
posted by dis_integration at 7:06 AM on October 13, 2018 [2 favorites]
posted by dis_integration at 7:06 AM on October 13, 2018 [2 favorites]
CS is a lot more than coding. That you think it could replace the need for CS coursework stool out as a yellow flag to me.
Look for a CS prof with a dual appointment in psych, cog sci, neuroscience, or if you’re lucky philosophy, make an appointment, and ask about opportunities. I firmly believe you should work with someone as an undergrad in the proper field before even thinking about a PhD.
posted by supercres at 7:11 AM on October 13, 2018 [2 favorites]
Look for a CS prof with a dual appointment in psych, cog sci, neuroscience, or if you’re lucky philosophy, make an appointment, and ask about opportunities. I firmly believe you should work with someone as an undergrad in the proper field before even thinking about a PhD.
posted by supercres at 7:11 AM on October 13, 2018 [2 favorites]
A friend of mine did his PhD in a psychology dept a while back and he was basically using neural networks to model human learning, so may be some possibilities in that discipline.
posted by biffa at 7:53 AM on October 13, 2018 [1 favorite]
posted by biffa at 7:53 AM on October 13, 2018 [1 favorite]
Why do you want to get a PhD? Work backwards from that question as well. Are you thinking you want an academic career? Many PhDs in CS /AI don't wind up working in academia, but take corporate positions.
Are there people specifically known for ethics of AI working at the university you want to apply to, and are they interested in taking on grad students who want to study ethics as the primary topic of a disseratation?
Because you don't love math or science enough to do an UG major I think you should carefully decide whether you would be better off doing the AI degree focusing on ethics or a humanities degree focusing on the topic of AI (ie philosophy, cultural studies or even medical anthropology).
posted by nantucket at 8:20 AM on October 13, 2018 [4 favorites]
Are there people specifically known for ethics of AI working at the university you want to apply to, and are they interested in taking on grad students who want to study ethics as the primary topic of a disseratation?
Because you don't love math or science enough to do an UG major I think you should carefully decide whether you would be better off doing the AI degree focusing on ethics or a humanities degree focusing on the topic of AI (ie philosophy, cultural studies or even medical anthropology).
posted by nantucket at 8:20 AM on October 13, 2018 [4 favorites]
Response by poster: Geographically and financially, the university I want to go to for my masters has a fantastic undergraduate comp-sci and maths program, and the one I'm at now has a comparatively bleh one. Even if I got an honours degree in comp-sci or maths at my undergrad place, they would look pretty lousy compared to an honours degree from the same university.
I love math and science, but I'm not confident that I can do an all-math degree at 40 the way I could have at 20.
I want to go into academic research, but I've got alternative back-up options. I have to switch majors this month because my current major requires in-country fieldwork that's not possible for me now, and I just don't want to end up with a degree that doesn't have a possible path to ethics in AI research as the only topic that's kept my interest over the past decade consistantly.
posted by dorothyisunderwood at 10:33 AM on October 13, 2018
I love math and science, but I'm not confident that I can do an all-math degree at 40 the way I could have at 20.
I want to go into academic research, but I've got alternative back-up options. I have to switch majors this month because my current major requires in-country fieldwork that's not possible for me now, and I just don't want to end up with a degree that doesn't have a possible path to ethics in AI research as the only topic that's kept my interest over the past decade consistantly.
posted by dorothyisunderwood at 10:33 AM on October 13, 2018
Do the CS/Phil. dual degree. If you want to be taken seriously in the field, the degree will matter, even in just the gatekeeping ways. Like academic CS is full of retrograde dillbeans but the way to improve it within the academy starts with a CS undergrad even tho it’s nit appealing
posted by dame at 10:38 AM on October 13, 2018 [1 favorite]
posted by dame at 10:38 AM on October 13, 2018 [1 favorite]
Best answer: What are the backgrounds of the PhD students in the program you're targeting? What are the backgrounds of the people (particularly early career people) publishing relevant papers or presenting at relevant conferences?
Depending on the community, not having a CS degree (or perhaps math or stats) may be a a serious hindrance or there may be people coming from cognitive science, psychology, philosophy and so on. I say focus on early career people because they're a more reliable indicator of the paths that are actually open.
It's unclear how much geographic flexibility you'll have, but keep in mind also that there are programs/departments that are more open to students who don't fit the template (and are therefore seen as "risky"). It's not a "good" or "selective" vs "bad" or "weak" program thing, it's a culture thing.
posted by hoyland at 11:06 AM on October 13, 2018
Depending on the community, not having a CS degree (or perhaps math or stats) may be a a serious hindrance or there may be people coming from cognitive science, psychology, philosophy and so on. I say focus on early career people because they're a more reliable indicator of the paths that are actually open.
It's unclear how much geographic flexibility you'll have, but keep in mind also that there are programs/departments that are more open to students who don't fit the template (and are therefore seen as "risky"). It's not a "good" or "selective" vs "bad" or "weak" program thing, it's a culture thing.
posted by hoyland at 11:06 AM on October 13, 2018
Best answer: I'm sort of at the edge of work on this topic (a collaborator's lab does a bunch of it) and I don't think one could seriously engage in actually currently relevant issues in AI ethics without the technical skills and first-hand experience in machine learning, and probably a good handle on the currently massive and rapidly-changing literature on applied machine learning. I don't even really think it would be possible to do a good job of figuring out what questions to ask without all this. You will find it hard to get these outside of a CS department or perhaps a cogsci program that has a big computational component [disclaimer, I teach in one such program].
If you want to actually do philosophical ethics or metaethics, the answer would have to be different, because it's really hard to get traction in philosophy departments without philosophy degrees, but I just can't see doing any kind of actually relevant AI ethics inside a philosophy program without at least a pretty heavy CS background before that.
However, this is an area that is developing and changing extremely rapidly right now, and it also seems really hard to anticipate what's going to be going on in 6-7 years, it's very dependent on where exactly ML goes from here...
posted by advil at 12:52 PM on October 13, 2018 [5 favorites]
If you want to actually do philosophical ethics or metaethics, the answer would have to be different, because it's really hard to get traction in philosophy departments without philosophy degrees, but I just can't see doing any kind of actually relevant AI ethics inside a philosophy program without at least a pretty heavy CS background before that.
However, this is an area that is developing and changing extremely rapidly right now, and it also seems really hard to anticipate what's going to be going on in 6-7 years, it's very dependent on where exactly ML goes from here...
posted by advil at 12:52 PM on October 13, 2018 [5 favorites]
Best answer: There's a very important distinction that I'm not sure about from your question, because there's two very different ways to interpret what someone might mean by "AI ethics".
-- are you interested in problems of bias and fairness in current, actually-existing (or near future) machine learning? like, stuff in the wild right now that is actually guiding people's lending and hiring and sentencing decisions?
If so, I agree that it will be tough without a math or CS degree, or at least something very quantitative and computational -- physics, comp bio, statistics, or something like that, anything that would give you the programming skills and applied mathematical fluency to turn yourself into a machine learning practitioner very quickly.
Another path might be economics -- economists or people trained in econ do actually participate in research in this area (after all, econ already tries to address issues of fairness). I don't know that it would be any less math-heavy but it might be different math.
-- or are you interested in more of the Nick Bostrom type issue of ensuring ethical behavior in a hypothetical superintelligent AI?
(I don't personally find the extreme fear some people have over superintelligence to be reasonable, but plenty of people way smarter than me would disagree.) If this is what you want, I think philosophy is your best bet. Most of the people I've seen who actively do research on these problems (as opposed to just worrying about them late at night) are philosophers.
posted by vogon_poet at 8:39 PM on October 13, 2018
-- are you interested in problems of bias and fairness in current, actually-existing (or near future) machine learning? like, stuff in the wild right now that is actually guiding people's lending and hiring and sentencing decisions?
If so, I agree that it will be tough without a math or CS degree, or at least something very quantitative and computational -- physics, comp bio, statistics, or something like that, anything that would give you the programming skills and applied mathematical fluency to turn yourself into a machine learning practitioner very quickly.
Another path might be economics -- economists or people trained in econ do actually participate in research in this area (after all, econ already tries to address issues of fairness). I don't know that it would be any less math-heavy but it might be different math.
-- or are you interested in more of the Nick Bostrom type issue of ensuring ethical behavior in a hypothetical superintelligent AI?
(I don't personally find the extreme fear some people have over superintelligence to be reasonable, but plenty of people way smarter than me would disagree.) If this is what you want, I think philosophy is your best bet. Most of the people I've seen who actively do research on these problems (as opposed to just worrying about them late at night) are philosophers.
posted by vogon_poet at 8:39 PM on October 13, 2018
Response by poster: Okay, I've applied to change my major to philosophy and data science (comp-sci). Thank you hivemind! I went through my favourite AI ethics papers and tracked backwards for their grad students to see where they came from and had a look through the current grad students at the university I'm interested in to find possible paths, and which undergrad programs were strongest at my current school.
posted by dorothyisunderwood at 11:51 PM on October 13, 2018 [3 favorites]
posted by dorothyisunderwood at 11:51 PM on October 13, 2018 [3 favorites]
« Older What to expect from a polygraph? | Weird Feelings About Volunteering at a For-Profit... Newer »
This thread is closed to new comments.
posted by kestrel251 at 4:20 AM on October 13, 2018 [9 favorites]