Science as a (micro-)service
Today I will talk about the science, or more specifically about organization of the science. What is wrong with it and at least one possible way to fix it.
It is not about programming (there will be more programming articles, but they will be here for convenience), but I will use some IT terminology and analogies.
How does science work?
First of all let’s start with the problem. Do you know how modern science is working? Anyone familiar with the subject will immediately ask, which science do I mean? There are different kinds and they work differently.
There is fundamental science. It’s aim is to get the general knowledge and it does not have immediate practical application. The fundamental science is mostly done in scientific institutions and mostly funded by government (or several governments at once). The consensus is that fundamental science must be governmental because nobody else could spend a lot of money without obvious effect (I do not fully agree with that and we will discuss it later). It does not mean, that fundamental science does not give any economical effect. On the contrary, it is huge. All our modern technological advance rises from the success of fundamental science. The problem is that the effect is seldom delayed (by years and even tens of years) and the benefactor is not the one, who invested money, but the humanity as a whole. Fundamental science also plays an important role in educational system and is a pillar that holds other types of science. The most problematic part in fundamental science is to measure the contribution of specific people and institutions in the total result.
The next one is applied science. It differs in terms of aims and methods. It is no longer the general search for knowledge, but the research applied to a specific problem: improving qualities of something, lowering the cost of production, developing of the technology. In this case the aim is much more obvious and measurable. Time frame are smaller (several years at most). The applied science is often funded by corporations. The work is mostly done in small independent groups and the results are sometimes shared (and sometimes are not).
Finally, we can designate another group usually not mentioned in common classifications: commercial research. A short-time research projects with fixed aim and goals. This research is usually commissioned by industrial companies and performed by people, connected to science in some way. Sometimes there are firms that specialize in doing commercial research (like data science). Though without any connection to academia, they could not be effective for a long time.
In this article we will mostly talk about the fundamental science and scientific institutions. Because it is a pillar that holds everything else and it is the most problematic part.
The problem
What are the requirements (beyond just money) to do proper fundamental science:
- Open communication. The more the scientist communicates with other scientists during their research, the better the effect. But the communication is not free. Any web developer would say to you that someone needs to build communication routes and support them. Scientist often are not good in this.
- Infrastructure. In order to do science, you need not only a pencil and paper. You need an office, chairs, tables, computers, services, computation clusters, food nearby, accounting, PR.
- Education. In order to do proper science one needs connection to education facilities. Because proper research always needs more young people. Also training people is one of the most important results of the fundamental science.
As we can see, each research group requires some additional facilities to work. And those facilities are costly.
Let’s look at Picture 1. It shows how science was functioning at the beginning of ХХ century. The green squares show “useful” science. The blue things are local infrastructure and blue arrows show communications paths. They are fat to depict its cost because it was pretty expensive back then. You had to send letters. When I say “expensive” I mean not only actual money, but the time and the effort as well.
Now we can see that the system is not very effective. If we take the surface of green part and divide it by the total surface, you would see that infrastructure takes major part of science costs (I did not precisely measured the surface, but it feels like more or less accurate).
The solution and another problem
The solution to this infrastructure problem is known to any systems architect. Lets bring those science groups together under the common infrastructure. Because its cost is not directly proportional to the size of science group it is serving. It will also partially solve the problem of communication (they work together, it is easier).
Look at Picture 2. It looks much better. Now all science groups have common infrastructure and and work together. The costs are higher than for a single group, but much cheaper than for all groups being separate. This is how scientific institutions came to life. If you base those institutions on existing University infrastructure, you will have integration with education for free. Profit.
It was working like this everywhere for the second part of XX century and the first part of XXI. But there is a flaw in this concept. Not obvious to most, but well known to the same systems architects. What is shown here is a monolith architecture. Easy to create and cost-effective, but very inflexible. If parts of the structure stop functioning or change the direction of research, you can’t do it right away. Because all research group already have designated places in the total structure. Starting something new is also a problem. Because infrastructure is huge and it is very hard to add new parts to it without a lot of preparation. If you had worked in a large research institution, you know the pain. The start of the new project (if it is not pure theoretical) takes years of preparations and tons of bureaucratic work. It is even harder because the work is mostly funded by a government. And government likes to counts its money.
Because of this rigidity, the most active groups start to create their own infrastructure. Their own connections to education, they own budgets etc. They create a “wrapper” around them. Like in Picture 3.
We can see that the cost of infrastructure has risen again, because you have to do it twice (on the laboratory level and on the institution level). And you still have the problem of frigidity. The communication problem also risen again because the institution can’t provide the specialized type of communication needed for the modern research and the groups need to do it themselves again. They start their own conferences, work chats, seminars, etc. If they have specialists to do that, of course (they usually don’t since I have to do it).
It is exactly the same in web services design. You start to create wrappers around individual functional components and the whole system becomes expensive and fragile. You know what they do? They do services.
Science as a service
Let’s get back to Picture 1. Nowadays the infrastructure is much cheaper. You can fully automatize budgeting, offices, issue tracking and even communication. You can work and communicate remotely. And you can even outsource equipment production because nowadays it is done mostly by small private companies, not by scientific institutions themselves.
But modern science is different from XX century science. It is more specialized. Different groups do different things and the results of one group are used by another. It means that you have not the flat structure, where each group “does everything”, but more like a graph structure, where one group depends on the other or even have a choice of several providers of the same services. It is OK since communication and information discovery is cheap nowadays.
The service architecture is harder to establish than the monolith, but it is easier to maintain in the long run. It allows easy replacement and re-specialization of individual parts (or scientific groups in this case). One can open new fields without turning cogs in a huge institutions. You can even mix fundamental and applied science and get resources from several sources.
It is not free of course, one needs to design the communication protocols and conventions. The expertise is always a problem in science and so there should be “ science services” that analyze the performance and capabilities of other science groups. And there should be a consensus about such expertise. And the most important thing: one needs to take risk and start working outside of “traditional” organizations.
From service to micro-service
You probably have noted that service approach solves flexibility problem, but does not fully solve infrastructure problem. Again, this problem is well-known in web development. This is how they developed micro-services. The idea is that you have many very small services with minimal overhead, connected to common data bus and usually to established deployment tooling. The data bus could be quite complex, but since it is done only once for a lot of services mitigates the costs.
The same could be done for science. Consider Picture 4. It is mostly the same as Picture 1. But with much smaller overhead and the common communication tooling. Let’s take money management for example. Money management takes a lot of effort in science. You need to do several months of paper work in order to sign an agreement which will be OK for both you and the customer. It takes even more work to move the money to the scientists themselves. Work contracts, taxes, etc. But we know that this problem is successfully solved in outsource agencies. We have even taxi and food delivery services that work on a common informational and money infrastructure, but otherwise working independently. Is it possible to do that for science? I think it is.
The future begins today
The solution I explained today could be also called “network science” (I took the term from one of Sergey Parkhomenko talks, where he mentioned “network journalism”). Our team has been walking in this direction (lowering infrastructure costs, minimizing expenses, starting our own communication without growing in size) for several years. But now I am happy to say, that we are not only ones there. As it always happens with ideas for future, some other guys started to move in the same direction. For example CASUS science team is too large to be considered “micro”, but they have similar ideas and organizational process. The same goes for our recent acquaintances from Kazakhstan.
We will discuss it more and maybe understand the future better. Feel free to contact me if you have your own ideas or cooperation proposals in this field.
I want to write another article about similar problems and solution in education in near future. The approach I explained here won’t work as is in education, but there could be similar techniques.
References
- Scientific Programming Centre: https://sciprog.center/
- MIPT-NPM lab: https://npm.mipt.ru/en/
- The discussion of modularization in scientific software: https://youtu.be/MNYb-aVDCl8
[UPDATE] About education
The second part of the story (about education) is here: https://medium.com/@altavir/education-as-hexagon-5f0f17f09230