post inter view

I've interviewed, perhaps, at many a company for someone my age, and after skimming through what should have been a Lee Kai-Fu (the Legend of) something-but-not-exactly-hagio 'graphy and coming across an article Niniane posted a while back, I wanted to serialize some of the thoughts I have had about the modern interview process.

Personally, I have done okay at interviewing, although it has been a sometimes rocky road in learning the (hard ways the) importance of "it's a small (pass as white) world" presentation. I've gotten better, but who knows what would have happened if I had access to one of LKF's books when I was growing up ABC?

I don't know, but the struggle is part of me now, like some darkly symbiote.

(I might as well try to be blunt now, since I don't know an easy way to cushion the guile behind the (I note, invisible) sonic boom that you don't hear when someone drops the p- word.)

And as I look back, I feel the importance of acknowledging the difficult-to-estimate impact of white-p, male-p, CMU/MIT-ish privilege (and its dual (duel?), my own internalized not~white, not~male, not~CMU oppression).

These are hard things to broach, and, I can imagine, hard things to hear. But they must be spoken some day, by someone, once more.

I've come to know some of my "strengths" and "weaknesses" as judged by, well, the people that did (or didn't) want to hire me.

My "weaknesses" include my soft-spoken nature and my humility. If I have a talent for anything, it is a talent for understatement. (I leave the problem of figuring out my other talents, if they exist, as an exercise for the reader ;)

More abstractly, another "weakness" is my penchant for non-linear/speculative execution. I often like to solve a problem by jumping directly to the answer in a single visible leap (like some super girl), as my intution is not bad...

(How bad is my intuition? See previous previous paragraph on my most in.visible superpower.)

I think superscalar thinking has a number of benefits, and Lee Kai-Fu reportedly seems to agree (given his comments in some book). Non-linearity may have the same power as a random bit. That is, it often gives an easier way to find or guess the answer with some possibly-high probability.

For those used to thinking-in-the-small and extremely linear thinking, superscalar execution has no place in an interview, as only deterministic proof certificates matter. But, with sufficient discipline after the fact, and a little known idea called "collaboration", speculative branching can be a pleasant ally.

Another "weakness" of mine is that I will sometimes hand wave things as being low-level wrt my current goal, and therefore just details. Something I can figure out later, by reading in between the lines of TaoCP. I think this has to do, implicitly with power dynamics. I think I agree with McCarthy's argument as to why there are separate sciences (or at least why there have been, in retrospect, even though boundaries always shift and change in size, like just-add-water boyfriends or shrinky dinks).

When I say power dynamics, I'm talking about intellectual power dynamics. The power to say that this area of study...

(whether it be latte-sipping Dinosaurs or an ugly variant of CBCL that needs to be "reevaluated" in light of human+computer-friendly YAML. If there's one thing I picked up from my brief interactions with Rudder-san, it's that Standards are sometimes dumb. When I think about it now, saying that we should use something only because it is Standard and the proposed alternative is not-yet-Standard is like going back in time and trying to convince merchants to stick with Roman numerals as a data interchange format, because they're Standard and it doesn't matter how hard it is for a median person to easily manipulate Roman numerics, arithmetically or geometrically.)

...is worth more study as it is the basis for Everything blah blah never mind the fact that my shoes have holes in them don't look behind the ring 0 curtain. Yep, some people haven't learned that reductionism is, as we know it today, an incomplete system of thought.

If the details can be worked out later, it doesn't mean that the details (how it can be implemented in your favorite system (cough API) of incomplete thought) are somehow, well, "inadequate" it's just that for a given task, let's say trying to help humans and computers simultaneously (gee what a concept), issues that relate to a finer level of granularity (a way to put it, I suppose, that may not as quickly invoke the swollen head syndrome in some unknowingly infected by privilege) can be dealt with later (some of the time), or abstracted out, to use a fancy new term that I admit can be difficult to truly understand.

Finally, understanding itself must be analyzed from an intersectional social justice lens.

Modern interviewing theory often says that "A people hire A's" but "B people hire C's". That is, you assume a strict, deterministic hierarchy, and make sure that those below you don't get uppity.

While I agree that you can sometimes tell when you see someone younger than you making the mistakes that you once made (an "A" person being able to estimate someone else as "B"), that does not imply that as an "A" person, you sample from The Right Distribution (rather than Way, could be a Way of Distribution, hmm), and estimate the (ugh) static level of someone smarter than you.

In my travels, I have come to see that react differently to new ideas and other-ly diverse new acquaintances. Some can check their privilege, listen for once or twice in their life, and maybe even learn something new. Some others, implicitly take "think different" to mean "think worse". If I don't understand this person, they must be less worthy than I blahblah blah blahblahblah.

The static hierarchy of intelligence rather than the dynamics of hard work raises many questions, but if we stay within that frame, let me more simply state that while smart people can often tell that someone is not as smart as they are, they sometimes get it "wrong" when they come across someone smarter.

That is, by posing the problem as "how can we avoid hiring people dumb-r than we are" you can easily fall into the trap where you miss some fraction of the people smarter than some smartest subset of your bunchiness. Newsflash, someone tell the dotcoms that their interviewing pipeline is hosed.

With that in mind, I'd like to think that some academics get this right, and have an inkling of when they come across somebody whose random variable of intelligence sometimes exceeds their own random variable of xQ for x in {E, I} and any other still undiscovered letters of intel privilege.

(This can be argued as follows. In industry, smart people get used to seeing folks that are not as smart, along certain dimensions, as they are. Thus, they build systems to filter out dumb people, forgetting to check their own egos at the door. Bright academics at some universities seem to have a different experience. They've seen a student or two that showed promise and later proved to, perhaps, even exceed their expectations, and this gravy brain train of reasoning leads them, given sufficient time to mature, to realize that once in a while, however unlikely, a kid comes along and helps to change the way things are seen in some previously unexpected way. Hmm...)

Enough beefing. Let me just end with one way that interviewing systems may be improved, in theory. I call it, extreme interviewing. Just kidding, please don't use that term. I love Smalltalk and Smalltalk-inspired people, but let us just call it pairing, or more descriptively, "good cop bad cop".

You (might) know the drill. You're shoved into a little room without a window. The light is high and fluorescent. Some angry man comes in and asks you to explain Krylov subspace iteration, Taboo-style, without using the words vector, point, and (sub)space. Ouch, you've forgotten your numerics.

He leaves, still angry, and someone else comes in. This person offers you a Krispy Kreme donut and takes you to the free as in beer beverage room, and asks you to implement bidirectional iterators over randomized BSTs.

Phew.

My improvement is not suggesting this style of guerilla interviewing, but to point out that interviewers can be paired, in a good cop bad cop way (!). Note that this, like pair programming, may reduce defects and produce more reasonable results (in this case people). Not only that, but note that the human bandwidth is effectively almost doubled! Twice as many eyeballs. You do the math, how much better will your hires be?

I haven't figured out the (sociocombinatorial) details (waving hands gently) as I have yet to see pair-interview-ing implemented in practice, but it might go as follows.

Bad cop comes into the room. She doesn't have to be wearing a uniform. Her first question has to do with generalizing the pancake flipping problem for arbitrary d. The candidate stumbles. Bad cop presses a red button (on her laptop), and in comes good cop, bearing a small cornucopia of healthy snacks.

Would you like a donut? asks good cop. A Luna|Power- bar?

The candidate happily agrees. (Nobody has given the candidate food as of yet.)

Good cop asks bad cop what bad cop is asking of the candidate, in full view of the candidate. Bad cop explains and good cop gasps.

Hey, isn't that too difficult? says good cop.

Well, maybe, but what, do you want me to solve it for them? replies Bad Bobo-Cop. We only hire the Bestest (TM).

Good cop looks at the candidate sympathetically, and wrinkles a lip or two.

Well, continues Goodie Cop, why do we try to solve it together and you tell us if we're on the right track. If the candidate doesn't mind, that is, but we'll be sure to evaluate the candidate fairly, as we're about teamwork here (or at least that's what we're supposed to say), and we want to know how people work on a team (even if, surprise, we never ever ever nevernever usually make people :o work on a team in the interview process (oh no, how scandalous!)).

Bad cop hems and haws like a cheesy B-grade book, but finally agrees.

Short story shorter, bad cop plays the adversary and good cop the sympathetic oracle. The problem really should be hard enough so that good cop and candidate need to work hard, together, and communicate, if they want to get a decent answer, and perhaps even the bad cop is soo Strong + Bad that Bad Bobo Cop doesn't even tell good cop how to sample from the Right Distribution of answers before the interview begins.

You'll have to work on the details on how this is implemented (I'm talking to you, Niniane, and your Pasty Friends too), but it may just work and produce better hires in the long run, while encouraging a gentle spirit of cooper-tition.

There's a lot more to be said about the power dynamics and issues of diversity that arise when you start talking about how static hierarchies propagate institutional oppression, but I need to eat breakfast, and you might need to digest some of these tasty? nuggets.

Enjoy, but eat not in excess.

~L