Another 34000 People Are About to Put Their Future in the Hands of an Algorithm | Hacker News Reader

Another 34000 People Are About to Put Their Future in the Hands of an Algorithm

fivethirtyeight.com

66 pointsevanb24 comments

An algorithm is still an algorithm if a human is running it. Therefore, humans put their futures in the hands of algorithms being run by impartial humans all the time; partiality is only nice when it's acting in your favor, after all, and we don't want to be at the mercy of people who might not like us for some undefinable reason.

I agree, even the presidential election is decided by an algorithm.

> we don't want to be at the mercy of people who might not like us for some undefinable reason

I keep restating this, but people still hate bureaucracy. Go figure.

This is the kind of title that I would edit. The article is about how the National Resident Matching Program will use an implementation of the Gale-Shapley algorithm for matching medical students and residencies. The title makes it sound much more significant than it is; virtually all of us "put our future in the hands of an algorithm" countless times every day. Everything we do that leaves a trace in a database somewhere will affect our future to varying degrees.

The article does not overstate the significance. Have you known people going through Match? It is quite significant for those 34,000 people.

I don't follow your argument. The match process has very little in common with people leaving traces of information in a database.

I'll compare the NRMP with something in the same ballpark: a credit score. There are three services (last I checked) involved there. You might have a mix-up, but a credit score can be disputed and corrected. To my knowledge, for all practical purposes, there is little recourse if the Match does not go as you hoped other than trying again the following year.

One might also compare the NRMP with the standardized tests for college admissions; that would make for an interesting discussion.

Numberphile has an episode on the Stable Marriage Problem: https://www.youtube.com/watch?v=Qcv1IqHWAzg

The "matching program" isn't new--goes back to the 60's if not further. Sounds like the idea hasn't changed even if the "algorithm" is machine-implemented in recent times, no surprise there.

Used to be the student would put on his/her list the residencies in order of priority, and so would the institutions rank the candidates. Let the matching begin!

A thing to note is that not all residencies were in the mix and there were always a good number of students who weren't selected or maybe didn't even participate. That left room for students to find vacancies to fill.

AFAIK there weren't any students who couldn't get into a residency somewhere even if not among their top choices. Besides by the end of the first year of training there were always plenty of residents shifting to other programs so "holes" migrated around to an extent.

Chances are the situation is not at all dire for the current crop of graduating medical students. It tends to all work out.

>where the students will do their residencies — three to seven years or more of supervised training, usually in a hospital. Since you can’t get a license to practice medicine in the United States without completing a residency, this is a high-stakes game.

at least in CA (and majority of other states) the requirement is only 1 year of post-med-school training:

http://www.mbc.ca.gov/Applicants/Physicians_and_Surgeons/

" How much training must I complete before I am eligible for licensure?

A US/Can must complete 12 continuous months of training in a single program to be eligible for licensure. Further, a US/Can must be licensed by the end of the 24th month of training. "

So going into 3rd year of residency one must already be licensed.

The rest of states:

http://www.fsmb.org/policy/public-resources/state_specific

Note: That is not to dismiss the importance of residency. As one can see from the above, its importance goes way beyond of just getting the license.

The last section of the article was really interesting to me. Some points from it:

1) Students need a residency to become a Doctor. 2) Residencies are funded by congress. 3) More doctors -> cheaper healthcare

So why don't we fund more residencies? And why is the government the one funding them?

I'm on my way out, so this'll be quick, but a few other restrictions on supply:

* There can notionally only be as many fill-able residencies as there are graduating medical students -- this number would probably be the most easy to increase (since schools are fine with making more money), but it's actually quite difficult to find enough instructors to teach the students because the job pays so poorly compared to many other avenues a physician can take (and you need a certain instructor/student ratio for both accreditation and to attract students).

* Each residency program can only admit as many students as they are licensed to do, and this licensing generally is tied to the program volume (so a program seeing X patients per year would be granted Y residency spots). This gets tied to the professional organizations for the individual specialties (an oversimplification), who actually have an incentive to restrict the number of residency spots in order to maintain their own job security.

* There are plenty of available residency spots in the fields we need most, such as primary care -- it's just that unless people feel particularly compelled, they don't go into those specialties, because compensation and work/life balance are so poor compared to many specializations.

? Primary care is the easiest work-life balance outside of radiology. 0 stress 28-hr weeks.

Zero stress? 28 hour weeks? Where are you getting this from?

http://medicaleconomics.modernmedicine.com/medical-economics...

> Median hours per week remained unchanged from 2011 to 2012 for family/general practitioners at 50 hours, though visits per week increased slightly from 98 in 2011 to 99 in 2012.

Why is this an article anyway? They use matching algorithms here in Canada for university co-op (intern) jobs as well. Lots of people don't get jobs, but that's no fault of the matching algorithm.

My hunch is that a lot of value could be added -- and perhaps money made -- by rethinking the resident matching process. This would be much more than an algorithm redesign.

There are, for the record, often unfilled residencies in infectious diseases. This specialty tends to end up on the bottom of physician pay scales however.

Based on work of "Lloyd Shapley and David Gale". Can somebody explain why is it better than just solving good old "Assignment Problem"?

I'd be interested to know if the algorithm can be audited, and if so, by whom and why what process.

A bit click-baity. The "algorithm" is the same vein as the "algorithm" that runs our voting system. (add up the votes and choose the winner).

I don't think the "matchmaking algorithm" in the blog post actually matches the colloquial meaning of the word "Algorithm".

It isn't really counting and choosing a winner though. It's an example of the gale-shapley algorithm for the stable marriage problem - http://en.wikipedia.org/wiki/Stable_marriage_problem

http://en.wikipedia.org/wiki/Stable_marriage_problem

Actually a relatively interesting algorithm, but I don't think the author explains it fully. The men must also rank the females. The key point is that for a match between people A and B, A has no person ranked higher than B who also ranks A higher than their current partner. it is optimal for every participant.

I'm not too familiar with residency- I wonder if hospitals submit a list of potential students they accept? How does the length of a hospital's acceptable candidates list compare to the number of hospitals a student will apply to? In the Stable Marriage Problem, each set member must rank every member of the other set.

> I'm not too familiar with residency- I wonder if hospitals submit a list of potential students they accept?

The article does mention that, "Med students and hospitals are currently ranking their top choices from these interviews." I would assume that the hospital only ranks students they interview and would consider acceptable for the program.

> Each set member must rank every member of the other set.

I would assume that they just fill in the blanks with some value that prevents matching. I don't think a hospital or a resident would be very happy with what would amount to random assignment of the leftovers.

Yes, they only rank students they interview. Those students who don't match before the rankings are exhausted "scramble"; they speed-date programs that still have openings in the hope of matching, so the leftovers are not randomly assigned. (Whether the outcome for them is better than random is hard to say.)

But isn't this exactly the Gale Shapely algorithm for stable matching? I was reading the Algorithm Design book by Kleinberg and Tardos and they introduce both the algorithm and also the fact that it's used for matching students with residencies. If it's algorithm enough for a book on algorithms I can't imagine it not matching the way it's used here.

I agree. In my algorithms class in undergrad, the Gale-Shapley algorithm (and its cousin, the Boston-Pool algorithm) were used as motivating examples of how algorithms affect our lives at the very beginning of the semester.

Reply on news.ycombinator.com