We are in the early stages of understanding how the COVID outbreak will reshape society. The impact of the virus will be wide-ranging, and artificial intelligence will not be an exception. Fears around travel may lead to low participation at major research conferences like the International Conference on Machine Learning and NeurIPS, even after lockdowns ease. Immigration and visa limits may make it difficult for students and researchers to seamlessly move among companies and universities around the world. All of these changes will directly affect the field of machine learning.
One dynamic I'm tracking is the economic impact of the pandemic. COVID is placing a massive strain on the economy and forcing researchers, companies, universities, and funding organizations to take a hard look at their priorities. They are being forced to consider questions like: “How should we allocate our work on techniques that can be quickly applied and more speculative research?”, “Should we put resources into building new research partnerships or deepen existing ones?”, and “Should we reduce spending on research and development entirely?”
As individuals and organizations readjust their research agendas to meet new financial realities, it could play out in a few different ways. As an initial way of investigating the possible paths forward, we’re forecasting a set of metrics related to the future of machine learning research.
Scenario 1: A New AI “Winter”
The most pessimistic view is that recent investment in machine learning was enabled by an economic climate allowing for spending on speculative research and new product development. Robust investment also underwrote a frothy market in hiring machine learning talent and supported new research collaboration.
Now that COVID is projected to create a substantial contraction in the global economy, we might anticipate the arrival of a new AI “winter” in which activity in the field declines significantly. Some indicators that we might be headed in this direction include:
|Quantity of AI research decreases||Quantity of AI publications |
|Peer to peer interaction decreases||New coauthor pairings [Forthcoming]|
|Hiring in AI / ML decreases||Job postings for ML roles |
Scenario 2: Shuffling the Research Agenda
A somewhat less pessimistic scenario is one in which COVID shuffles the list of research priorities in the field. As resources become more limited and companies, universities, and funding organizations are forced to invest in different areas, the rate of progress made in certain subdomains within machine learning will change.
An adjusted research agenda might leave overall publishing activity relatively unchanged, but result in major changes in the types of problems pursued by researchers. Some indicators that might show this reshuffling include:
|Decline in the level of investment and research activity around reinforcement learning||Percentage of AI publications on reinforcement learning [Forthcoming]|
|Decline in the level of investment and research activity around ML fairness||Percentage of AI publications about fairness [Forthcoming]|
|Decline in research focus on the problems of “AI ethics”||Corporate press releases about AI ethics |
Scenario 3: A New AI “Spring”?
An alternative view is that all this pessimism is unwarranted. COVID may spur demand for machine learning, even as the economy remains under pressure. Demand might rise for using machine learning to assist in tracing the virus and identifying potential cures. Businesses might automate more of their operations as a way of “virus-proofing” against future outbreaks.
These dynamics may support a renewed AI “spring,” in which investments to machine learning research continue at pre-COVID levels, hiring for machine learning talent increases, and the applications of the technology expand. We might track whether we are headed to this scenario in a few ways:
|Quantity of AI research increases||Quantity of AI publications |
|Hiring in AI / ML increases||Job postings for ML roles |
|AI Industry grows||Money raised by AI companies [Forthcoming]|
The implications of COVID for the research landscape mean policymakers and scientific funding agencies need to get a handle on where funding gaps might appear in the near future. Anticipating the future trajectory of the field is critical to ensure that funds are allocated to have the greatest impact. Supporting basic research, even as private funders withdraw, may be crucial to ensuring continued national competitiveness in the field.
We’d love to get your input and any suggestions for other metrics that might build on this hypothesis! You can read and respond to our forecasting questions at cset-foretell.com.
Author: Tim Hwang, CSET