Research progress – both on the scale of the individual researcher and the scale of a research community – relies on balancing two competing objectives: (1) exploiting existing knowledge, and (2) continuing to learn new things. I frame these objectives as analogous to two types of research or intellectual “investments”: short- and long-term.
Short-term intellectual investments exploit one’s current knowledge and skill set to advance a project or achieve a goal. Examples include working on a project that is a direct extension of one’s previous research papers; applying off-the-shelf tools for data analysis; and writing a small module for a software package in a language that one is already comfortable with.
These investments can be “cashed out” quickly in the form of project completion or a research paper. These investments typically involve direct application of one’s existing skill set to a problem. This means that the work involved tends to be relatively rote, and there tends to be limited opportunity for deep intellectual growth along the way. However, this type of work leads to quick, tangible progress.
The types of goals that can be achieved via short-term investments are limited by the scope of the previous long-term investments one has made in their skills and knowledge. We describe these next.
On the other hand, long-term intellectual investments don’t yield any immediate payoff (and it may always remain difficult to attribute any particular success to these investments), but they widen the universe of possible goals one can achieve. For the (STEM-oriented) researcher, examples of long-term intellectual investments include learning about a new branch of mathematics; learning a new programming language or software engineering framework; and reading textbooks from seemingly unrelated disciplines.
Long-term intellectual investments require dedicating a longer period of time to developing more fundamental (and possibly less directly applicable) sets of skills. It usually won’t be possible to immediately “cash out” these new skill sets in the form of publications or tangible progress. However, there is a hope that they will pay off in the long run by broadening one’s perspective of a field of knowledge. Having a wide knowledge base also allows for more easily making connections between fields, which is often a fruitful way to generate research ideas.
The downside of long-term intellectual investments is that they are usually high-risk. At the outset of one of these investments, it’s typically unclear what the tangible benefits will be, and many of these endeavors may be useless in the long-run.
The distinction between short- and long-term intellectual investments becomes especially apparent as a graduate student (the occupation that arguably allows the most time for long-term intellectual investments). Feeling the pressures of career advancement and research progress, many grad students understandably want to publish research papers as early and as quickly as possible in grad school. To do this, they capitalize on short-term intellectual investments by using the skills that they already had prior to starting grad school. However, this strategy closes the door to developing fundamentally new skill sets and perspectives.
While the above descriptions may seem to imply a superiority of long-term intellectual investments, both types are crucial for meaningful research progress. By my own observation, the most impactful researchers have a balanced “portfolio” of both types of investments. This is also analogous to the exploration/exploitation trade-off in sequential decision making problems. On one side, researchers have a set of low-risk, short-term investments which will clearly result in publications and research progress (however incremental). On the other side, they have a set of high-risk, long-term investments that have the potential for larger impact down the road. Anecdotally, it appears that the optimal research strategy consists of a blend of the two.
Let’s consider some examples of research projects at the scale of research communities. Here are four examples, each classified as a combination of low- or high-risk and short- or long-term. Note that the categorizations I propose here are highly debatable.
The table below shows four research projects sorted by risk level and time scale. Importantly, I want to emphasize that the definition of “risk level” used here is roughly the level of uncertainty, assessed *before starting the project*, that it will succeed. It does not refer to the level of danger of the eventual product or outcome of the research.
|Short-term||Testing new deep learning architectures||Developing mRNA vaccines for COVID-19|
|Long-term||Human Genome Project||Mars colonization|
Let’s look at each of these in turn.
Proposing new deep learning architectures is a short-term, low-risk project. Given the large number of plug-and-play programming tools for deep learning (TensorFlow, PyTorch, etc.), the time required to implement and train a new deep learning model is relatively small. Moreover, even if the model doesn’t perform as well (or better than) other existing models, what’s lost is just a short amount of time and some compute power (although this may not be negligible, depending on the model), making the project relatively low-risk.
The development of mRNA vaccines during the COVID-19 pandemic could be considered a short-term, high-risk project. When the pandemic began, there was a need for a vaccine as soon as possible. However, while mRNA vaccines had been studied for other diseases prior to COVID-19, none of these vaccines had been approved for use in humans. Thus, at the start of the pandemic, there was substantial uncertainty around whether mRNA vaccines would be a fruitful path toward immunity for COVID-19. Thankfully, the mRNA vaccines developed by BioNTech, Moderna, and others turned out to be incredibly effective. If the race to an mRNA vaccine had failed, it could have prolonged the horizon toward successfully immunizing people, making this project high-risk.
The Human Genome Project could be considered a low-risk, long-term project. By the time of the project’s inception in 1990, it was clear that there would be immense value in knowing the sequence of the human genome. Furthermore, much of the technology was already in place to be able to carry out this sequencing (although the efficiency and cost of these technologies would greatly improve over the span of the project). What was needed was a large amount of time and money (about 13 years and $3 billion) to complete the project. Thus, even though completion of the project required an enormous amount of effort, researchers were relatively confident that the expected payoff (in terms of knowledge gain and applications of the work to medicine and biology) would be high.
Finally, the human colonization of Mars could be considered a long-term, high-risk project. Given that no single human has set foot on Mars, it is reasonable to predict that the creation of a full human settlement there is still a ways away. Carrying out such a project – whose goal is to provide a safe haven from a future Earth that may be uninhabitable due to climate change, overpopulation, or otherwise – would require an enormous amount of human and financial resources. Moreover, pursuing this path to human sustainability diverts resources away from tackling environmental issues on Earth (which are much more well-studied), adding to the risk of the project.
Importantly, all of these projects could provide lot of value to the world (and three of them already have).