Posted by: jonkatz | January 12, 2010

What is the “right” amount of funding for theory?

Aaron Sterling’s recent post about ICS noted how well-funded theoretical computer science appears to be in China*, and contrasted the “malaise” toward theory in the US to the high regard in which theory seems to be held in China. (Being at an institution that, as a whole, does not seem to value theory certainly makes me sympathize with Aaron’s remarks. Having said that, I especially enjoyed the third comment on his post. [Go ahead and read it, I’ll wait.] Not that I entirely agree with it — I just found it amusing.)

That post and the ensuing discussion raise the larger question: what is the “right” level of funding for theoretical computer science research? More broadly, how would one go about determining how much money “should” be allocated to any of the different sub-areas of CS? Yes, theory is important. But surely other areas of CS are important, too? (Maybe one or two areas are even more important?) We are used to grabbing as much money as we can for theory — both because theory has been historically underfunded (and under-appreciated?) and also because that’s the way the game is played by all academics — but consider this question: say you were in charge of allocating NSF’s budget for computer science research. How would you determine the “right” allocation? Of course, the question only becomes more complicated when you consider the allocation of funds across all the sciences.

I don’t have any immediate answers to these questions. I just find it interesting that, with all the (past) focus on increasing funding for theory (e.g., theorymatters.org, which hasn’t been active in a while), these questions haven’t been discussed. It’s easy to say “theory should be given more funding”, but harder to answer the question “in what areas should funding decrease in order for that to happen?”.

(*As an aside, I suspect that the total level of theory funding in the US is orders of magnitude more than the total level of theory funding in China. Tsinghua University has a high concentration of top-notch theorists, but how many theorists are there in all of China compared with all of the US?)

Advertisements

Responses

  1. Is theory more underfunded than other areas? I’m new to the funding treadmill (still trying to figure out where the “on” button is) and most of my knowledge comes from theorists who repeat the mantra “there isn’t enough money”. How little is the money? How would you even quantify how much funding theory deserves compared to other areas?

  2. Is theory more underfunded than other areas?

    I don’t have the answer off the top of my head, but I believe the theorymatters website has some historical data regarding this issue.

  3. One way to gauge how much money theory or any other field should obtain from the government is to distribute it proportionally to how many theory faculty are hired at universities, assuming that the hiring follows demand of the students for theory classes.

  4. If the goal of research funding is to benefit society, then perhaps we should to partition theory further. Certain areas have had a significant impact in practice, while others not so much. To further complicate matters, often there is a large lag time in TCS between discovery and application. This means that often we are left to read the tea leaves during the funding process. For example think back to the early 70’s: is this PKC idea just a pipe dream or will it eventually pan out into an actual useful cryptosystem?

  5. If you could choose between curing cancer and proving that P is not NP, what would you chose?

    I’d submit that any “normal” human being would choose to cure cancer. In fact: the two objectives are so incomparable that even a 1% chance of curing cancer within the next decade is preferable to a 99% chance of proving that P is not NP within the same time period.

    The implication is that if one has to allocate a certain amount of money each year to two programs – say, one aimed at cancer research, and the other at complexity theory – one should allocate 0% to complexity theory.

  6. Cancer research is funded at a rate of about $10 billion a year from government sources and about an equal combined amount from industry, states, and non-profit organizations. Compared to that, one could safely say the amount of funding for TCS does round down to 0%.

  7. Here are some yearly academic R&D spending numbers for the US in 2008 according to http://www.nsf.gov/statistics/infbrief/nsf09318/ , in billions of dollars per year:

    Medical science 17.3
    Non-med life sci 13.9
    Physical Sciences 3.9
    Elec Eng 1.7
    Computer Sci 1.5
    Mech Eng 1.2
    Math 0.6

    The total (including some other areas) is 51.9 billion dollars per year.

  8. In 10-20 ranked computer science schools in US, I think funding in theory is really scarce. I am a theory grad student and I always have to worry about getting funding to go to conferences, whenever my paper gets accepted. That is really frustrating and demotivating.

  9. A noticeable and growing portion of electricity in the US is expended on computation. We have Google buying up old aluminum smelters to get the electricity to power its server farms. Efficiency in computation is increasingly of benefit to society as a whole. Seems like a good reason to support a field focused on the question even independent of the benefits that the field has provided in the past.

  10. I’m surprised that nobody asked whether there is intrinsic value in knowledge. Why do we spend money on astronomy, space exploration, experimental physics (which usually has 0 application)? Personally, I believe human understanding has its own value. We should be comparing TOC to other theoretical fields, not only to applied computer science and cancer research. (Admittedly, TOC straddles this divide, so it does make sense to compare it to applied research as well. But most of these comments are neglecting to ask about the value of the knowledge it contributes, in addition to the value of the derived application.)

  11. “The implication is that if one has to allocate a certain amount of money each year to two programs – say, one aimed at cancer research, and the other at complexity theory – one should allocate 0% to complexity theory.”

    I don’t believe this is the implication of your proceeding question. If it were, you’d have to choose your most pressing question or need and dedicate *all* extra money, beyond the minimum we need to support ourselves, to solving that problem. Obviously there is value in diversifying our expenditure. Haven’t you ever played any Rio Grande games? 🙂

  12. Experimental physics has zero applications? One of the most noted experimental physicists of the second half of the XX century was involved in the invention of the radar, the cobalt radiation chamber for curing cancer and night vision googles. The laser, the electronic microscope and the microwave are also inventions with experiment physics pedigree.

    On the subject of time-lagged applications, cancer and the P=?NP question, perhaps a certain key advance in the P vs NP question would allow us to sequence genes and predict protein structure faster thus enabling the cure for cancer.

    Say, if we believe there is a 1% chance of such a breakthrough being made in the next ten years, this _alone_ would justify an expenditure of $100 million a year on TCS.

  13. We should also keep in mind that maths and TCS is cheap, compared to say, biology or experimental physics: we have no expensive equipment. Even when there are experiments, it is usually done on cheap computers.

  14. “I’m surprised that nobody asked whether there is intrinsic value in knowledge… Personally, I believe human understanding has its own value…We should be comparing TOC to other theoretical fields, not only to applied computer science and cancer research. “

    But our resources are limited, so somehow we have to compare what we gain from the intrinsic value of knowledge to what we gain from more applied endeavors – like saving people’s lives. Spending $X on theoretical physics may uncover the secrets of the universe a decade faster than otherwise, and spending this same amount of money on alzheimers might give an extra year of normal life to 1000 people.

    Somehow you have to make this tradeoff. As I said before, it seems obvious to me that the two gains are wildly incomparable, but I cannot prove it to you – it is a question of values.

    “perhaps a certain key advance in the P vs NP question would allow us to sequence genes and predict protein structure faster thus enabling the cure for cancer.”

    Its not impossible. But if you started out with the goal of curing diseases, its completely unclear that you should allocate resources to the P vs NP question .

    In general, I am skeptical about whether the spillover from theory to practice justifies investment in theory. Its true that ideas that have been developed by theorists have sometimes ended being used by practitioners. This does not, however, justify the mountain of completely useless work produced by theorists. If you started with the goal of developing all kinds of useful stuff, its unclear that a good way to do this is to give money to people who do useless stuff knowing that one or two of their ideas will end up being useful.

  15. sorry for posting under two names.
    -pierre/alex

  16. I don’t think the implication that theorists are oblivous to applications holds. Not even for most of them.

    Certain subfields of TCS tend to ignore applications, but this is not true across the board. Think AGT: Papadimitrou was inspired by actual applications when he started proselitizing about mechanism design back in 1996.

    Or to bring it closer to home, think PKC. This area was pioneered by theoreticians who very much had an eye on applications.

    I can think of countless other examples such as the work of Kleinberg, Mitzenmacher, Muthukrishnan, Papadimitriou, Karp, Lipton, the the three RSA guys, etc.

  17. Pierre/Alex:

    Just keep in mind that TCS != complexity theory. The relationship is

    complexity theory \subsetneq TCS

    currently we have

    TCS=crypto+algorithms+computational complexity+theory B+quantum+approximation algos+parameterized complexity+distributed algos+parallel algos+mechanism design+…

  18. I think that compared to its actual impact, CS theory is way over-funded. Over the last few years, CS theory has moved more and more towards complexity theory, and away from any kind of relationship to any practical application. The only way to discourage this trend is to reduce the funding for this kind of work even further.

  19. Over the last few years, CS theory has moved more and more towards complexity theory, and away from any kind of relationship to any practical application.

    I think it depends where you look. For example, over the last decade and a half MIT has moved away from CC towards algorithms and of the big three conferences SODA is the one that is alive and thriving, while STOC and FOCS stagnate.

  20. Prof,

    I agree. Instead of asking “what is the right level of funding for TCS,” I think its better to ask “what is the right level of funding for crypto,” “what is the right level of funding for quantum computing,” and so on. Different subareas have very different potential real world impact.

  21. Some problems may just be out of reach of the research agendas of present, and throwing more money at them may not necessarily bring a solution.

    I was shocked to find out the extent of money being misspent and wasted in these so called “research” activities for alzheimer’s / cancer from friends working at places such as NIH.

    Diversification is a prudent strategy in dealing with uncertainty. You don’t know what field’s fruits you’ll be harvesting next. What initially seemed mere intellectual curiosity brought you the hardness of approximation, and similar curiosity may let you base cryptography on wider-believed assumptions.

    One cannot justify less funding for complexity because of some bs papers because in every field there is some bs research. I can tell that one of the reasons that I moved to theory after almost a decade of doing “practical” stuff is because of the amount of bs I’ve seen done in Software Engineering, AI, and even VLSI-CAD.

  22. I’m writing a comment from India. Funding any sort of research in India is really hard. The Government obviously likes to fund research it sees as directly benefiting a large number of people. You can imagine how TCS isn’t exactly a favorite. Surprisingly, most TCS (and this includes supporting students, conference travel and so on) is funded by the Software Industry, who in turn tend to use theoreticians as consultants !! It would also come as no surprise that if you looked at the quality of research groups it’s again the industry that’s far ahead of academia. MSR-India, IBM-IRL and TIFR have better TCS groups than academia.

  23. I think that compared to its actual impact, CS theory is way over-funded.

    In general, I am skeptical about whether the spillover from theory to practice justifies investment in theory.
    Actually, CS theory produced the ideas that led to the pagerank algorithm, it created RSA that allows you to do banking online, and it produced ideas that sequenced the human genome, to name a few. Of course you would say that we should fund these ideas instead of funding all of CS theory. You should also invest only in stocks of companies that will increase 1000-fold in the next few years, and not in all of the others.

    If you started with the goal of developing all kinds of useful stuff, its unclear that a good way to do this is to give money to people who do useless stuff knowing that one or two of their ideas will end up being useful

    In the same way it is unclear that a good way to govern a country is by using a democratic system. It is not an ideal system, but we do not know of a better one. Keep in mind that we would not have RSA if it were not for completely and utterly useless number theory research. I think a committee deciding what is useful and what is not is much worse than a peer-review system determining what is important and what is not. It has it’s inefficiencies, but every big system has its inefficiencies.

    Prof,

    I agree. Instead of asking “what is the right level of funding for TCS,” I think its better to ask “what is the right level of funding for crypto,” “what is the right level of funding for quantum computing,” and so on. Different subareas have very different potential real world impact.

    Different people also have different potential real-world impact. In fact, different grad students have very different potential real world impact, and even different problems a grad student may think of have very different potential real world impact. Maybe every morning, a grad student should submit a proposal of what to do that day, and then if the funding powers determine that this idea-of-the-day is potentially useful to the world, he gets paid that day.

    Sarcasm aside, I think that the balance between subareas is in fact taken into account by the funding agencies, either explicitly (Quantum is in a different branch, as is crypto) or implicitly when choosing the number of people on the committee from each area.


Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out / Change )

Twitter picture

You are commenting using your Twitter account. Log Out / Change )

Facebook photo

You are commenting using your Facebook account. Log Out / Change )

Google+ photo

You are commenting using your Google+ account. Log Out / Change )

Connecting to %s

Categories

%d bloggers like this: