Novelty

It is very easy to make something novel, but what’s difficult is to make it novel and impactful. If you put together some random stuff together, that will be novel, but that doesn’t guarantee its usefulness and impact at all.

There are many ways to think about novelty. One way is in terms of surprise or conditional probability given the past. Given what we have so far, how surprising or new is this? This can be measured in many different ways. But one thing also to note is that not only the dissimilarity from the past but also the similarity to the future is what defines important knowledge. Simon DeDeo‘s work on the French Revolution quantified the influence of speeches using Jensen–Shannon divergence between topic vector of the speech to the past and the future.

Another way to think about novelty is Combinatorial novelty. In many cases, something is novel because it creates new combinations of existing things. Hyejin Youn‘s work on patent categories show that now there are more of such combinatorial innovations than something completely new. Brian Uzzi‘s atypicality (Uzzi2013atypical) or James Evans‘ surprise measure are nice example (Shi2023surprising).

Sometimes, however, the novelty may be about doing old thing, but just way better, replacing old technologies and thinking with new. This idea of replacement may be overlooked if we examine the surprise or combinatorial novelty. A novel contribution may have both old and new components; it may tackle an old problem using new techniques. Or, it may tackle a new problem with old techniques.

Science

Teplitskiy2022novel argues that novel papers are more likely to be accepted in the Peer review process.

In machine learning

In AI/Machine learning, novelty(-seeking) is sometimes a guiding principle for the agent. Boult2021towards attempts to formalize the notion of novelty in this context.

Culture

Askin2017what and O’Toole2022novelty study how novelty and popularity correlate.