The world’s leading publication for data science, AI, and ML professionals.

Paid time off and the nature of science

Flexible work arrangements aren't just a nice HR practice. They expose our brains to messy real-world patterns, which helps us find…

The view from this week's office.
The view from this week’s office.

I’m writing this in a beach house on the North Carolina coast. I just spent a week with my wife and kids playing on the beach, canoeing inland waterways, learning about rescued sea turtles, and mini-golfing.

This has also been the most productive work week I’ve had in about two months.

The last two months have been fine – I’ve gotten a lot done – but I’ve been hitting my head against a particular analytic problem. I classify problems into two types: mysteries and puzzles. With a puzzle, you kind of know what you’re trying to build and you just have to figure out how all the pieces fit together. With a mystery, you don’t really know what you’re looking for, so you’re a lot slower in figuring it out. This week, my mystery turned into a puzzle, and I’ve already put a lot of pieces together. I did this while working remotely on Tuesday through Thursday between call-in meetings, and also a little on Friday morning while my kids were watching SpongeBob.

This isn’t one of those if-you-can-just-decompress-you-can-see-things-clearly sorts of deals. Both my job and the company that provides that job allow for a lot of decompression. By doing, over the course of this week, things that I don’t normally do, my brain has needed to process things it doesn’t normally think about. Some of those things have proven useful.

I’ve always been strongly in favor of progressive HR practices like unlimited paid time off and work from home policies. I’ve supported these policies mostly on the principle that I get hired and paid to deliver results, not to sit in a specific room for a certain amount of time each week. If I can deliver the results I’ve promised, I should be able to arrange my life however I feel I need to (assuming, of course, that my doing so does not interfere with my coworkers’ ability to deliver their own results, etc.) It means a lot when an employer is willing to display that kind of trust in me, and it just generally improves my quality of life to not have to worry about a quota of days I can be away from the office.

Not that I believe unlimited PTO or remote work are unreservedly good. I think both work best when then represent a change from a normal work routine. When remote work is the routine it can be really draining, especially if the majority of people on the team aren’t working remotely and therefore haven’t developed the habits or infrastructure of accommodate remote workers. I think when flexible work arrangements remain the exception for an employee rather than the rule, the demonstrate employer trust and can be very healthy and empowering for the employee.

All of the above reasoning is still important to me, but over the last few days, I’ve started to view my vacation and work-from-home time as pretty essential to my ability to be a good data scientist: being away from my normal routines actually helps me perform those routines better. I’m certainly not the first person to look at things this way. For example, look at how mathematician Henri Poincaré described his discovery of what came to be known as automorphic forms:

I left Caen, where I was then living, to take part in a geological conference arranged by the School of Mines. The incidents of the journey made me forget my mathematical work. When we arrived at Coutances, we got into a break to go for a drive, and, just as I put my foot on the step, the idea came to me, though nothing in my former thoughts seemed to have prepared me for it, that the transformations I had used…were identical with those of non-Euclidian geometry. I made no verification, and had no time to do so, since I took up the conversation again as soon as I had sat down in the break, but I felt absolute certainty at once. When I got back to Caen I verified the result at my leisure to satisfy my conscience.

I then began to study arithmetical questions without any great apparent result, and without suspecting that they could have the least connexion with my previous researches. Disgusted at my want of success, I went away to spend a few days at the seaside, and thought of entirely different things. One day, as I was walking on the cliff, the idea came to me, again with the same characteristics of conciseness, suddenness, and immediate certainty…

The stuff I figured out this week certainly isn’t as impressive as Poincare’s discoveries, but they’re important to me and, I believe, will be important to the business that employs me and to the customers who employ that business. I didn’t walk along the shore thinking about my work – in fact, work was the furthest thing from my mind, until it suddenly wasn’t.

Something in this idea seemed familiar to me, which led me to search for an essay I hadn’t read for a very long time: "Thick Description" by anthropologist Clifford Geertz. Geertz rose to prominence in 70s for promoting an new, interesting, and useful way of studying human culture, and is viewed ambivalently by anthropologists today because his approach didn’t end up being as new, interesting, or useful as his early supporters’ hype made it out to be. Detractors aside, I appreciate Geertz’s views on the difficulty of understanding complex and opaque systems (which is more or less what data scientists do). I’ve omitted the words "culture," "anthropology," and similar terms from the following excerpt, because I think it applies to any analytic task.

Coherence cannot be the major test of validity for a description. Systems must have a minimal degree of coherence, else we would not call them systems; and, by observation, they normally have a great deal more. But there is nothing so coherent as a paranoid’s delusion or a swindler’s story. The force of our interpretations cannot rest, as they are now so often made to do, on the tightness with which they hold together, or the assurance with which they are argued. Nothing has done more, I think, to discredit analysis than the construction of impeccable depictions of formal order in whose actual existence nobody can quite believe.

If interpretation is constructing a reading of what happens, then to divorce it from what happens – from what, in this time or that place, specific people say, what they do, what is done to them, from the whole vast business of the world – is to divorce it from its applications and render it vacant. A good interpretation of anything – a poem, a person, a history, a ritual, an institution, a society – takes us into the heart of that of which it is the interpretation. When it does not do that, but leads us instead somewhere else – into an admiration of its own elegance, of its author’s cleverness, or of the beauties of Euclidean order – it may have its intrinsic charms; but it is something else than what the task at hand…calls for.

I’ve said before (for example: [here](http://housesofstones.github.io/2013/07/09/anthropology-and-data-science-need-each-other/) and here) that a data scientist’s job is so similar to that of an ethnographer that Data Science and anthropology really go hand in hand. As Geertz says, an ethnographer can easily construct a model of a society that is elegant and beautifully coherent and completely divorced from reality. While data scientists often (although certainly not always) have validation sets that allow them to sanity-check their work, there’s an always-present temptation to pick a mathematically elegant solution simply because of its elegance.

I’ve spent the last two months trying to get multiple elegant solutions to fit the facts and they just wouldn’t. This week, I thought up a far-from-elegant solution. It was so inelegant that I had to do some extra work just to make sure it would scale. But it works. At first I thought it was kind of ugly, especially compared to the former approaches I tried. But it occurred to me that the reality I’m trying to model with this approach is itself really ugly – it’s a mishmash of signals from multiple sources, each filtered through multiple intermediaries, and reported with varying amounts of fidelity.

I’d spent so much time over the last few months working with models of reality instead of actual reality that my mind was falling into the rut of seeing life as more simple – and more clean— than it really is. That was limiting my ability to tackle a messy problem. That’s not a just a problem for people who work with computers or people who use math. I think it’s a problem for scientists in general, because models are a scientist’s stock-in-trade. When the models become our reality, we cease to do good Science. We need time away from our models.

That might seem like a pretty high-flown argument for unlimited PTO, but for your average data practitioner – and I think I’m pretty average – paid time off and remote work arrangements are the only ways to get prolonged exposure to the messiness of the real world. Data practitioners benefit disproportionately from flexible work arrangements because the stuff we’re trying to understand it real but the models and even the data representing that real stuff is necessarily artificial. We need to see new patterns in real life to keep our edge in finding new patterns on computers.


Related Articles