Rambles around computer science

Diverting trains of thought, wasting precious time

Wed, 03 Mar 2021

Career thoughts on academia, industry and points in between

In early 2018 I was at a crossroads in my career: academia, industrial research, or somehow go it alone? I was a postdoctoral researcher at the University of Cambridge, but feeling some pressures to move on. This post is a (somewhat edited) time capsule containing some notes-to-self I made about that decision.

What happened is “history”, of course: I became an academic at the University of Kent. However, you wouldn't necessarily have predicted that from reading these notes. I'm still not wedded to the academic path; most of the questions here are still live in my mind. What I am wedded to is my personal line of research; I'm not just going to “go and work for Google” (shudder). Being a computer scientist, the options and trade-offs I face are somewhat different from those in various other disciplines.

(I should qualify: here academia means mostly “UK academia”, which has some particular problems right now... but is, sadly, not too divergent from the rest of the world.)

I have not made that much effort to edit the notes, so they are sometimes staccato, sometimes repetitive, and don't reflect where my thinking has moved on (especially from experience at Kent; in hindsight they contain a lot of Cambridge privilege). They start here.

Industrial research obviously forms a spectrum: some has more of a development flavour, whereas in some extreme cases it resembles academic research. I would want to fall near the latter end, although not necessarily maximally so.

Problems with academia. There are many. Let me start somewhere, with the following.

Government interference, managerialism. I have become jaded and frustrated by these aspects of the academic sector. They are arguably worse than corporate nonsense in industrial research, because they are so needless. Endless political “initiatives” and fad-chasing, in academia, seem less excusable than when arising from a profit motive. See also “overheads” below.

Research councils are unreliable; it's better not to depend on them. However, as an academic one has the option of using them or not; industry funding is also an option, modulo perhaps high overheads.

Unlike most sciences, I don't need big funding from research councils. My skills are in demand; perhaps I could make enough money from consulting work? My corner of CS is one where a small group, or even a focused individual, can do great work; I don't need to pay for a huge lab or huge group.

I have always felt like an outsider in the academic world. I have a habit of rejecting established research programmes, and another habit of attacking cold problems. My work meets with bewilderment fairly often. Many of the people who seem to “get it” come from industry—thinking of my blog, Twitter and industry-talk audiences. Industry conferences are interesting—they make me feel sometimes a peer, sometimes a learner, sometimes an educator. It is an educational environment, but worlds away from undergraduate computer science; food for thought.

Skill set, part 1. I see myself as both a creative individual and an individual creator. Unlike many [natural] “scientists” I do not see myself as a collaborative agent in the service of some a grand research programme. To stick with academia seems to mean sticking with a system where this is frowned upon or misunderstood, partly from how my kind of CS research is different from “science” “research” in the main. (Of course creative direction doesn't preclude “collaborating”, but it has to be a collaboration on the basis of shared vision, not just nuts and bolts.)

Skill set, part two. Context-switching is not my forte. I don't know for sure, but I feel I could easily be crushed by even a moderate teaching or admin load.

Textbook teaching is a turn-off. In Cambridge I already spend a lot of teaching effort on material that I don't even know, that I've never needed to know, that is not where my expertise lies, and that has dubious educational value (never mind training value). In this instance I'm thinking mostly about the tedious way in which compilers are taught here [in Cambridge] and elsewhere. That perhaps comes down to the fact that my strength is not mathematics, yet in CS, at least in Cambridge, the educational material is desperate to make things mathematical wherever it can. I care first for education rather than training. But this is neither good training nor good education for many of the people with relevant skills/interests. (It is interesting that education may be tailored to a skill/interest profile without becoming “training”—or so I contend.)

Textbook rejoinder: I should take care to remember that the textbook stuff did teach me some things. As a beginning programmer, the mechanics of programming were fine, but I couldn't come up with how to solve certain problems: how to write a game or other reactive system (using an event loop), or how to do heuristic rather than brute-force search (I struggled with my “Countdown numbers game” program; I remember marvelling at the pathfinding abilities of later Sierra SCI games); or how to write a parser that worked systematically rather than by ad-hoc rewritings (I remember wondering how Sierra AGI games did this... noticing a theme here). Only by imbibing some textbook-ish wisdom (sometimes from lectures) could I solve these problems.

Academic, non-practical teaching of practical topics: I've wasted too much of my life staring at students' handwritten code on paper not being sure whether it works or not. A very innovative academic course might avoid this using automated testing. But the cost of redeveloping these from scratch is prohibitive if done on a per-institution or per-academic basis... once again, justifying MOOCs (more on MOOCs below). One could argue that a lectureship would allow me to fix that, by designing the courses differently. However, no doubt my room for variation would be limited, and it all comes out of the time budget.

Teaching repetition, teaching overstretch. Lecturers produce hastily-written slide decks over and over again. They throw away their predecessors' and start over, but throw in their own new fresh set of mistakes when they do so (thanks, Philip). As a supervisor I'm tired of reading hasty, sloppy, unclear lecture materials and of seeing older materials thrown away once a new lecturer starts. The set-up of departments asking lecturers to teach courses, which the lecturer then “owns”, encourages this. Might it be better if the institution owns the course? It's hard to say. Might it be better if lecturers own courses but maintain them whether or not the institution asks them to keep giving it? Seems optimistic. Continuity seems to be the important thing: not just of running the course, but in the creative vision behind it. The current system takes liberal opportunity to break continuity.

High overheads. The academic sector has high and increasing overheads on the funding it receives, not unrelated (call me cynical) to its growing managerialism. If someone in industry has an interest in funding some exploratory research with a university, paying 130\% overheads (or whatever) can quickly erase their perceived value-for-money. It doesn't help that (Cambridge) institutional policies on overheads fail to distinguish big/rich companies from small/startup ones.

Availability of eager young helpers. This is where the academic world does well. You get access to some bright young things who can contribute prodigious efforts towards your cause for little or no money, because (sad but true) they're still in a mode of paying money for the privilege of learning. Still, to date I've yet to really make student projects work for my research; I've seen others do it (e.g. Hrutvik's CakeML project), but it seems better in Cambridge than elsewhere simply because strong students are in greater supply. Industrial research does well too in this regard, usually via established programmes of internships, but that works mostly with more advanced students and requires big money.

Libraries, journal subscriptions etc. Any form of going it alone, as some mix of consultant and “gentleman scientist”, would suffer from lacking these, unless I could also keep (some fraction of) a foot in an academic institution. Or maybe physical “local resident” access to an enlightened university's library is enough.

Future of academic CS, part one: does traditional academic learning have a healthy future in general? Does it have a healthy future in CS-style learning in particular? I'm being vague because I start to doubt the value of much of CS teaching—although I should preface that “in Cambridge”. In certain other institutions, it looks more vocational, which is potentially fine, although it's not really my calling and I don't like to see it masquerade as “education”. My working hypothesis is that textbook academic-style CS content can be delivered via MOOCs fairly effectively, so there's no point investing myself in this style of teaching unless I have the appetite to innovate in that space (I don't, yet).

Future of academic CS, part two: two other aspects of “CS education” are less easily dispensed with. Firstly, practical skills; they take practice. Secondly, deeper and longer-term perspectives, which in theory are what academics are good at. These are the sorts of things that (I hypothesise) make my writings and talks appealing to audiences like Strange Loop, Hacker News, etc.. These people mostly have experience under their belt and enjoy material that helps widen their viewpoint. It would be hard to teach undergraduates, meaningfully, the sort of content that I write in those articles. This could all be an artifact of how we have to teach CS “basics” at degree level. Perhaps (I'm guessing) in other subjects one can deliver a more educational degree because students know the basics and are ready to think about the deeper things.

Many things I wouldn't miss about leaving academia. Two are constant pressure to overextend myself, and constant dissatisfaction from doing a mediocre job. (These have “Matt Welsh resonance”.) Another is the expectation of poor work/life balance. I'm okay with the Collegiate Cambridge model of building your life around your work, i.e. deliberately eroding any separation. In fact I think that dedicated creative work benefits from this. But it doesn't mean that one should be trying to work unhealthy hours. Creativity and clear thought don't benefit from that. Modern academic careers are so pressured that it seems hard to avoid this.

The PhD mill, and my moral objection to it: there's a sort of expectation that as an academic I would acquire funding and “acquire” people, typically PhD students (and worse, RAs) to do “my work” (as opposed to just “their work, that I supervise”). Overall I'm not sure I can get behind this modus operandi. As a PhD student I did my own work. That only created trouble for me later, since it's not the normal thing, but still I feel no need to apologise for doing it. Meanwhile, fetishising lineage in research is part of what turns mostly-spent research programmes into self-perpetuating monsters, and promotes monoculture. Even the word “mentorship” makes me feel cynical.

There are some good things about academia. One reason for wanting a steady academic job has been stated to me as a “cushion for the lean years”. But there are other ways to build a cushion; how much cushion is really necessary? Do I expect any lean years? One should expect some periods that area lean in terms of grants/funding. But if I were a consultant, should I expect years that are lean in terms of clients? One would just have to put up with less interesting work, presumably.

The state within a state. Academia offers institutional support. Sometimes, purely public services (thinking public libraries) and/or wider enterprise (thinking co-working spaces etc) provide similar support. Is that enough? An intellectually stimulating environment is something academia can provide, but probably only the top institutions do really well.

R-e-s-p-e-c-t. Respectability of the academic path is another factor; but I reject that, in principle at least (though I'm not completely immune).

Growing a group: an ability to build a group around my agenda is probably a potential good thing about academia, despite the aforementioned distaste. A small group could perhaps be manageable without abstracting too much time away from activity that is nominally research. But is having 2–3 PhD students, plus teaching and admin, really productive and/or enjoyable relative to what I could do by myself (say, on research time derived from consulting income)?

Scaling to more than one person: academia lets one grow small and even medium-sized teams, although not without also becoming something of a manager. A small team is still many times better than a team of one. There is also some built-in direction: a PhD supervisor has some leadership-style influence. Of course I instinctively dislike that the moment it becomes “control”. In other contexts the same dynamic might conceivably fly (or not) for my work... I can think only of open-source as a team-building tactic. Getting an equivalent extent of team-building that way seems hard. Open source has its own problems: even fewer funding avenues, potential for pulling in different directions (but this happens in academia too), project politics, unreliable (and/or physically remote) contributors, unpredictable timeframes, mob rule of online forums, etc..

What do I think about US-style CS degrees and academia? They admit some amount of liberal arts-style breadth, and take longer to teach material [than in Cambridge], including (sometimes) proper practical classes. But their research story is probably still infected by many of the things I don't like: mathematical orthodoxy, more generally the tyranny of many incumbent research programmes that I don't believe in, the gameability of modern research culture, ditto for career incentives, and the faddish government/funding initiatives (is that less bad there than here? unclear).

Some more points for going it alone. I could pursue long-term research in perhaps a more focused, lower-stress working environment—but perhaps lonely and isolating if it's only me there... ideally want a balance. I could be free to tick some personal “living as I choose” boxes in doing so, regarding lifestyle, physical environment and working environment. After years in West Cambridge on a woefully mis-designed or undesigned site, I miss what makes the centre of Cambridge such a good space: space-efficient small-town planning. The University is moving further from this model, partly for understandable reasons, yet its managerial rather than academic leadership means it is failing to develop an acceptable alternative. I like the idea of working somewhere more rural, with easier access to nature. I could vote with my feet in favour of economic localism and other values that large organisations are painfully slow to catch on to.

Reading the page on "rat race" on Wikipedia, it fits academia well. I am reminded that although the academic world attracts a formidable density of very bright people, bright people need not be far-sighted or big-thinking people. In fact I suspect the correlation between these is fairly weak. We shouldn't suppose that adherence to “the system” is a consequence of intelligent analysis and rational optimisation; bright people can be remarkably strongly bound to social norms and mores, even when they have the intellectual capacity to question them.

An institution in mine own image? I am somewhat inspired by the Recurse Center. And I am a believer, in principle, in “if you don't see the institution you want, create it”. Creating “my own institution” really just means doing things in ways that work for me but also in a way that might provide “a home for other people”, i.e. to fulfil the moral duty of an elder. This has some appeal. If I did create a space for others, it would be intentionally small and definitely not mainstream, so it would be fine if it didn't appeal to many people. It would mainly be about enabling people to do research, as well as enabling myself; but maybe it'd have an educational angle too. I am a fan of Cambridge-esque college-style residential establishments, as a way both to build communities and to limit costs (but this may not fly without big/expensive perks, and needs updating anyway). I'd be pleased if such an institution could contribute to a low-wage-economy area (which would keep property costs down; thinking the antithesis of the San Francisco) as long as it was not too far from civilisation. My educational interests are more in substantial life-long learning, than in an undergraduate hothouse or “stamping machine”. Ditto research. How could it work?

[/highered] permanent link contact


Powered by blosxom

validate this page