The following conversation is fictional.
---------------------------
ALICE: (Looking over a student's record.) Hmm, let's see. She wants to work in quantum computing. She's had the year-long quantum sequence in the physics department and has taken a course in quantum computing in the computer science department. She has done a project in quantum computing in an REU program. Grades good, letters good. I think we should admit her.
BOB: Wait! Did she get a minor in Physics? This is very important!
--------------------------
When looking over a student's record the questions
Does she have a minor in X? or
Did she double major? or
Did she graduate with honors?
never dawn on me.
1) When I am on an admissions committee I look at:
a) Transcript: What did they take? The grades are generally good so that's a minor factor.
b) Letters that tell me what they did within STEM. I don't care about ballroom dancing or moral character.
c) Papers they've written whether or not they have been published.
d) Their personal statement. They need to tell me:
i) Why they want to get a PhD. When Ted Kennedy challenged Jimmy Carter for the presidential nomination in 1980, Ted Kennedy was asked Why do you want to be president? See here for his rambling and incoherent answer.
Despite his background in proving lower bounds on approximation contingent on the Unique Games Conjecture, Ted Kennedy would not have gotten into our graduate program.
ii) What they are interested in (this may have been covered in part (i)).
iii) Why they are qualified.
2) Do I care what the major is? No. I care that they know computer science which I can get off of their transcript.
3) Do I care if they double major in (say) Math. No. I can look at the transcript and see what math courses they took. I don't care what (possibly arbitrary) rules their school has for double majoring.
4) Do I care if they minored in (say) physics? Not even a little. If they want to do quantum computing I care if they have taken courses in that area. I don't care what (likely arbitrary) rules their school has for minors. I took five courses in Philosophy as an undergraduate. Did I get a minor? I don't recall. Two of them were in logic so I don't think I deserve a minor.
5) Are they in their school's CS honors program? Some other honor program? Are they on track to graduate with CS honors? Some other honors? I don't care what (definitely arbitrary) rules their school has for honors programs. If they are writing a paper, honors thesis or not, I will want to hear about it from their letter writer and from their personal statement.
6) Do I care if they are in phi-beta-kappa? Sigma-Xi? Tau-Beta-Pi? The last two I only know about since I googled is there an analog of phi-beta-kappa geared toward STEM for this post. You can probably guess that I don't care about any of those things.
7) The point is that these formal criteria: major, minor, honors are not important when I am doing admissions.
a) Are they important to students?
I've heard that high school students who are honors students get a bumper sticker for their parents car that says:
My kid is an honors student at blah high school.
I would be more impressed if the bumper sticker said
My kid can prove the polynomial van der Waerden theorem.
b) Are they important to other people on the admissions committee?
8) Has the scenario I paint at the beginning of this post ever happened?
Many years ago, university administration was pushing us to reduce the number of required courses in the CS degree.
ReplyDeleteThat mean many courses we wanted to include didn't make the cut.
That was pretty contentious, because various faculty members wanted to ensure that students have a path towards those courses, otherwise they would become random courses and most students don't know enough to make informed decisions on what to pick, and they go default plus random
We ended up creating specializations within CS. The intention was two folks: first, it gives students a small number of good default selection of courses (e.g. CS BS with specialization on their, specialization is not required but they are good default), and second we wanted this to be on the transcript to indicate that this student has taken a solid selection of courses in a particular area to make grad admission committees see that clearly.
I believe minors can play a similar role.
Any good university gets way more applications than it can handle and uses signals like GPA etc. to trim down the list to a list that can be reviewed by committee members in depth.
None of these side signals will make someone admitted without a careful review, but how can you create a shortlist, in the pre-AI world we needed simple heuristics, and you can decide what signals are most useful for the filtering.
Does AI change this? It is complicated, particularly because AI is not reliable nor fair (see EU's AIACT). It might drop good students unfairly and without explainability (which is important, when you use AI/ML for making decisions about people) and auditablity. So I believe we will need these signals still.
Your question is a good one: why we need these side signals? But my answer is different from yours.
Also character aspects matter to me. I want to get students off good character. I'm the long run, those students are more likely to succeed than the higher IQ students that with bad character, and even worse I would regret training students that cause harm to society.
Lots of interesting aspects here, would love to hear how others think about these issues.
Smart, focused, mature aged or young nimble, producing papers in focs and wants to become prof somewhere like so many others? Of course young ones like Jacob Lurie are special but most young ones are not consequential even as professors.
ReplyDelete@gasarch, wait ... wait ... Ted did not prove any lowerbound! you got me there for a good minute! It would have made a better satirical joke if you had aligned the timing. For one, I CLAIM computational complexity did not really exist during the time that Teddie attended Harvard -- CORRECT OR FALSE?
ReplyDeleteFor another, if we had sticked with lower bounds what was a good conjecture to prove during that time? Probably circuit lower bounds? Along the lines, How many basic logic gates (relays, switches, or AND/OR gates) are strictly necessary to compute arbitrary Boolean functions? ... in 1949 Shannon answered this with \Omega(2^n/n). But I remember reading about efforts by David E. Muller (1956) who was formalizing and closing the constants on these bounds. Another problem was probably analysing complexity of MergeSort. I wonder if he had proved those lowerbounds I Betcha you would have laved this guy.