mathbabe

I’m at math camp, about to start the first day (4 hours of teaching a day, 3 hours of problem session) with my three junior staff (last year I only had one!). I expect I’ll be blogging quite a bit in the next few days about math camp stuff but today I wanted to respond to this blog post, entitled “The Fallacy of the Data Scientist Shortage”. I found this on Data Science Central which I had never known about but looks to be a good resource.

The author, Neil Radan, makes the point that, although we seem to have a shortage of data scientists, mostly what they do can be done by non-specialists. Just as you waste your time during a plane trip on things like security, waiting to board, and taxiing, the average data scientist spends most of her time cleaning data and moving it around.

If I understand this post correctly, they are saying that, because data scientists don’t spend that much time doing creative stuff, they can be replaced by someone who is good with data.

Hmm… let’s first go back to the idea that data scientists spend most of their time cleaning and moving data. This is true, but what do we conclude from it? It’s something like saying concert cellists spend most of their time practicing scales and rosining their bows, and don’t do all that much actual performing. Or, you could compare it to math professors who spend most of their time meeting (or avoiding) students and not much time proving new theorems.

My point is that this fact of time management is maybe a universal rule. Or even better, it may be a universal rule for creative endeavors. If you’re a truck driver then you can fairly said you worked the whole time you drove across the country, at a pretty consistent pace. But if you’re doing something that requires thought and puzzling then the nature of things is that it isn’t an 8-hour-a-day activity.

It’s more like, as a data scientist, you work hard to see the data in a certain way, which takes lots of time depending on how much data you have, then you make a decision based on what you’ve seen, then you set up the next test.

And I don’t think this can be done by someone who is strictly good at moving around data but isn’t trained as a modeler or statistician or the like. Because the hard part isn’t the data munging, it’s the part where you decide what test to perform that will give you the maximum information, and also the part where you look at the results and decipher them – decide whether they are what you expected, and if not, what could explain what you’re seeing.

I do think that data scientists can and should be paired with people who are experts at data moving and cleaning, because then the whole process is more efficient. Maybe data scientists can be brought in as 2-hour-per-day consultants or something, and the rest of the time there can be some engineers working on their tests. That might work.

Best Case

The scientific models and, when possible, the data have been made available to the wider scientific community for vetting. Incorrect or non-robust results are questioned and thrown out by that community, interesting and surprising new results are re-tested on larger data sets under iterative and different conditions to test for universality.

The result is that a person, with the help of their doctor and thorough exams and information-gathering session, and with their informed consent to use this data for their benefit, will have a better idea of what to watch out for in terms of health risks, how to prevent certain diseases that they may be vulnerable to, and how the tried-and-true medicines would affect them.

For example, in spite of the fact that Vioxx gives some people heart attacks, it also really helps other people with joint pain that aspirin or ibuprofen can’t touch. But which people? In the future we may know the answer to this through segmentation models, which group people by their attributes (which could come under the category of daily life conditions, such as how much someone exercises, or under the category of genetic profile).

For example, we recently learned that exercise is not always good for everyone. But instead of using that unlikely possibility as an excuse not to do any exercise, we could be able to look at a given profile and tell a person if they are in the clear and what kind of exercises would be most beneficial to their health.

It wouldn’t solve every problem; people would still die, after all. But it could help people live happier and healthier lives. It depends on the open exchange of ideas among scientists as well as strong regulation about who owns personal data and how it can be used.

Worst Case

The scientific community continues its practice of essentially private data collection and models. Scientific journals become more and more places where, backed by pharmaceutical companies and insurance companies, paid Ph.D.’s boast about their latest breakthrough with no cultural standard of evidence.

Indeed there is progress in segmentation models for disease and medicine, but the data, models, and results are owned exclusively by corporations, specifically insurance companies. This leads to a death spiral in modeling, where the very people who are vulnerable to disease and need medicine or treatment the most are priced out of the insurance system and no longer have access to anything resembling reasonable medical care, even for chronic diseases such as diabetes.

And you won’t need to give your consent for those insurance companies to use your data – they will have already bought all the data that they need to know about you from data collectors, which have been gleaning information about you from your online presence since birth. These companies will know everything about you; they control and sell your data for extra profit. To them, you represent a potential customer and a potential cost, a risk/return profile like any other investment.

I go back and forth on whether there should be an undergrad major or Ph.D. program on data science. On the one hand, I am convinced it’s a burgeoning field which will need all the smart people it can get in the next few years or decades. On the other hand, I’m just not sure how capable academics really are at teaching the required skills. Let me explain.

It’s not that professors aren’t super smart and great at what they do. But the truth is, they typically don’t have access to the kind of data that’s now available to data scientists working in Google or Facebook or other tech companies (see this recent New York Times article on the subject). Even where I work, which is a medium sized start-up, I have access to data which many academics would kill for. This means I get to play with an incredibly rich resource, assuming I have built up the toolset to do so.

So while academics are creating (unrealistic) models of “influence” based on weird assumptions about how information gets propagated through networks, nerds at Facebook and Google and Foursquare just get to see it happen in real time. There’s an enormous advantage to having the data at your fingertips – you get good results fast. But then since it’s all proprietary you can’t publish it (a topic for another post).

Another thing: since academics typically don’t have this kind of big data, they also don’t have to create tools or methods for taming huge data. Sometimes I hear statisticians say that data science is just statistics, but they are typically missing the point of this “taming” aspect of data science. Namely, if we use state-of-the-art proven statistical methods on 15 terabytes of data and it takes 50 years to come up with an answer, then guess what, it doesn’t work.

At the same time, data science isn’t purely algorithmic time considerations either, and a computer scientist without a good statistical background would be equally wrong if they said that data science is just machine learning.

For that matter, data science also isn’t purely speculative research – there’s a bottomline business aspect to it, and the intention is (usually) to make profit. But there’s no way someone with a business degree that doesn’t know how to model can be a data scientist either.

End result: To teach data science for reals, you’d need to form a inter-disciplinary department across business, computer science, applied math, and statistics. Even so, I’m not sure how well strictly academic departments can really teach the nitty gritty of data science if they do collaborate across departments because they just don’t have good enough data (and by the way, this is a huge “if” – it seems politically impossible in some of the universities I’ve talked to).

On the other hand, I think it’s a good idea to try, because it is a great opportunity to teach at least some basic stuff and to instill a code of ethics in young data scientists.

The way things work now, the tech industry takes in former mathematicians, physicists, computer scientists, and statisticians and puts them on projects creating models of human behavior (I’ll include finance in that category) that are infinitely scalable and sometimes nearly infinitely scaled. Nobody is ever taught to stop and think about how their models are going to be used and how to think about the long-term effects of their models.

In spite of all the data problems and political obstacles, I feel that for the sake of this conversation, i.e. of personal responsibility of a modeler, we should go ahead and make a program, because it’s important and it isn’t gonna happen in your typical finance firm or tech startup.