AI is Eating the (Research) World

Use of Machine Learning and Artificial Intelligence in all research disciplines has been experiencing and explosive growth over the past decade

A couple of days ago I came across this incredible graph that shows the rise of AI and ML in the context of scientific and academic research:

The graph seems to come from an article in Nature, which, as of this writing, I have not been able to track down. It shows that in Computer Science, for instance, AI/ML has been pretty steadily hovering in the high single digit percentile between 1993 and 2013, roughly between the end of the last “AI Winter” and the start of the current Deep Learning revolution. It also shows that the interest in AI has been growing exponentially in this field since 2013, well before the current Generative AI craze.

In other disciplines the use of AI/ML techniques has been very modest at best and only gradually ramping up until 2013, when finally the DL revolution started making headways there as well. Physical sciences seem to be leading the way, probably as a combination of being data rich and having many practitioners with advanced computational and technical skills required for the successful use of ML and AI.

My main takeaways form this graph are the following:

1. AI is not a hype or a bubble. The AI revolution has been steadily permeating all aspects of our lives, including how we do research. The current Gen AI revolution is just the latest development, albeit it will probably have some significant qualitative ramifications for certain fields. (See below.)

2. As has been argued many times before, most significantly recently with the advent of ChatGPT and other Gen AI tools, AI is a General Purpose Technology. As such, it is applicable across a wide spectrum of fields and disciplines.

3. IMHO, use of ML and AI still lags far behind where it should be in non-CS disciplines. I feel there are still huge unexplored opportunities there. I myself have been involved in several such projects, from identifying proteins in single cells, to working on the stability of RNA molecules, and even venturing into Computational Political Science with work on using Twitter (now X) data for measuring political polarization. Based on my experiences with these projects, I believe we are still only scratching the surface of what can be done.

4. As mentioned above, one of the main restrictions on the wider use of ML in research has been the availability of good large datasets for many projects. However, the advent of foundational models could change that dramatically. These models act in a smilier fashion to human experts in certain subfields, and they could help unlock new insights even in the fields where there is either very little data, or where datasets pose formidable challenge for more traditional analytical approaches.