8

I am currently a PhD student in computational social science, and I feel like my work doesn't really contribute positively to humanity. Basically I feel like I am running analyses and producing plots to "prove a point" or (optimistically) discover new insights about the world. But I'm not actually sure that the point is true, because if I run the analysis in a slightly different way, it proves another point entirely. It seems to very much depend on the specific dataset you use and the specific way you conduct your analyses.

I want to switch to a field where this doesn't happen. Ideally in order to have a publication you would have to produce a piece of software that does something useful for the world, and it would be impossible to fake its usefulness. I am affiliated with the computer science department at my university, so I was wondering what fields of computer science are the most "legitimate" in this way, and also friendly to people with very little background in them.

user60158
  • 81
  • 1
  • I can understand why you don't want to talk to your advisor about this. .Have you talked to the colleagues in your CS department about it? – Nobody Aug 09 '16 at 04:52
  • Most fields of applied CS (Machine learning, NLP, and Robotics I know of the top of my head), are vulerable to cherry picking of results -- that is to say you can test your method on say 10 or 20 data sets, then only report the results for the datasets you do well on. But in general all science in vulnerable to cherry picking in this way. Is that part of what you meant?

    Also most of the same fields are vulnerable to data contamination of the training set with the test set -- except on new challenges, with single submission. (Eg current SemEval comps for NLP).

     – Frames Catherine White Aug 09 '16 at 04:53
  • @scaaahu Unfortunately I don't feel like there's anyone I can talk to. I've mostly stopped talking to students in my department, since my close friends are mostly outside the university, and I work from home a lot. Among the faculty, the one person I know other than my advisor also works on computational social science, and might be even worse than he is in this regard. – user60158 Aug 09 '16 at 05:29
  • 2
    If you want to see your software do something directly and immediately useful in the world, you should be working in industry, not academia. Academic computer science is more about the theoretical underpinnings, not the use-this-year nuts and bolts. – Patricia Shanahan Aug 09 '16 at 06:32
  • @PatriciaShanahan I disagree. Industry is about what is profitable, generally in the short term. There is a lot of applied academic research that is useful, but not profitable. For example: computer vision applied to analysis of uncommon tumours, processing of astrophysical data... – Davidmh Aug 09 '16 at 10:24
  • @Davidmh: Useful, yes, but not directly and immediately so. Research in CS rarely results in a usable product. It results in techniques that can be implemented into a product, or possibly also in prototypes that showcase how these techniques could work in a real-world product. But the result is rarely a software that is directly usable (as in, optimized for requirements of real usage scenarios such as ease of use, robustness, ...), has all the flavours of features required (rather than just some exemplary ones that show that the concept is feasible), and supports the entire workflow ... – O. R. Mapper Aug 09 '16 at 11:14
  • ... (rather than just the part that lies within the subject of the research). To build upon one of your examples: A computer vision technique for analysing uncommon tumors is useful, but not directly usable. For becoming practically usable, it probably needs to be embedded in a larger system that archives and organizes the scanned data, that features an intuitive user interface to access the scanning functions, a settings menu for modifying the detail parameters of the analysis, some logging to detect how the analysis was conducted, possibly some ... – O. R. Mapper Aug 09 '16 at 11:14
  • ... network connectivity to exchange the scanned data with the rest of the infrastructure, and so on. Not to mention peripheral things, such as a user manual, a somewhat reliable means of getting support, compatibility with different systems rather than just with the environment it was developed in, and appropriate installation tools. – O. R. Mapper Aug 09 '16 at 11:14
  • 2
    @Davidmh To be profitable in the short term, there have to be people willing to pay money for systems incorporating the software. Generally, people pay money for systems because they find the systems useful, also in the short term. There may be exceptions, but generally if you want immediate usefulness, follow the money. – Patricia Shanahan Aug 09 '16 at 11:37
  • I think you're taking a shortsighted and impatient view of the field. Discovering something that is "useful" for the world is a long process, usually not begun and ended in a single paper. A paper that achieves something "useful" will almost always be based on other attempts that turned out to not quite work, and on theory whose practical significance wasn't immediately apparent. If people didn't publish those, the "useful" work would never be able to happen at all. – Nate Eldredge Aug 09 '16 at 18:13

3 Answers3

1

I work in the field of computational social science and during my PhD, there were many instances in which I had similar thoughts.

You provide me a hint in your second paragraph - you are affiliated with the CS department in your university. Unfortunately, a common fallacy of most CS departments which have programs in computational social science is to focus on the "computational" part and not really on the "social science" part. It is not an uncommon workflow to use machine learning techniques on a dataset collected using other computational techniques and then report results. This is because of many reasons, the least of which is that folks need to know computation in great detail - this is at many times possible in computer science departments and not in social sciences (although that is changing in many departments that I know of)

This is what is probably causing the disconnect in your mind. I strongly suggest either expanding your reading to the parts of social science that you work on or are interested in and talking to more social scientists. You need to think at the intersection of both computational and social sciences and not as a computer scientist. Social science contributes many understandings of individual and collective behavior (often in networks) For instance, there is a big difference in paper writing and reporting of results in WWW, KDD versus CSCW and CHI although all these venues promote and support computational social science.

Ultimately, at the end of the day, if you feel like you need to change your field you probably should! You should do what excites you the most but I just wanted to point out possible avenues of thinking about computational social science since it is quite novel.

Shion
  • 9,582
  • 1
  • 42
  • 61
1

There are many fields of applied CS where your computational analysis/design/calculation produces a result that is then tested experimentally. Some examples are computational biology, computational physics and computational chemistry.

Bitwise
  • 10,813
  • 1
  • 41
  • 64
0

You can find examples in safety, integrity and security related applications. The first ones jumping to my mind: real-time computing, cryptography, computer system security, compression.

Laurent Duval
  • 4,995
  • 15
  • 35
  • But in all these fields, purely theoretical results are still of great interest, aren't they? – Nate Eldredge Aug 09 '16 at 18:06
  • @Nate Eldredge Yes they are, even more when theory meets practice. I am thinking for instance about turbo-codes, or bounds in lossless compression – Laurent Duval Aug 09 '16 at 18:38