Started Aug 17, 2021 11:40AM UTC   •   Closed Jan 11, 2022 12:00AM UTC

How will the percentage of highly cited U.S. AI publications supported by a DoD grant change over the next three years?

This question is a metric for an issue campaign on the future of the DoD-Silicon Valley relationship. To learn more about this issue campaign and the relevance of this question, see the campaign's subpage and a related blog post. To learn more about our new rolling question formats, see this demo video or blog post.

Data and resolution details. This question resolves based on publication and grant data from Dimensions. A publication is an AI publication if it would be categorized under the artificial intelligence, machine learning, computer vision, computation and language, multi-agent systems, or robotics categories on arXiv, as determined by a CSET-developed classifier. "Highly cited" publications are those in the top 10 percent by citation count. A publication is a "U.S. publication" if any of its authors are affiliated with an institution based in the United States.

The percentage being forecasted is the number of highly cited U.S. AI publications supported by a DoD grant as a percentage of the total number of highly cited U.S. AI publications. 

***

The historical data underlying the graph is here.

* * *

To suggest a change or clarification to this question, please select Request Clarification from the green gear-shaped dropdown button to the right of the question.


Question clarification
Issued on 08/23/21 04:13pm
A user asked whether "grants" for this question include cooperative agreements. https://comptroller.defense.gov/Portals/45/documents/fmr/archive/12arch/arch1205_12_05.pdf The answer appears to be yes. No distinction is made in the Dimensions data between grants and cooperative agreements, but in Sam.gov (where the data comes from), cooperative agreements are classified under "grants," and some grant abstracts in the Dimensions data refer to themselves as cooperative agreements.
Issued on 12/02/21 02:52pm
The historical data for this question tracks citations for papers published in a given year. For example, the 2017 data point is for papers published in 2017, not all papers published by 2017. That said, for the historical data points, all citations are counted, even citations by papers published in later years. That means "historical" data points -- such as 2017 -- could actually change, as some papers might receive disproportionately many or few citations in later years. As it turns out, however, relative citation ranks in a paper's first year are reasonably predictive of their citations ranks in later years. The percentage being forecasted for a particular year, however, is the percentage at the end of that year. For example, the 2021 percentage will be the percentage as of 1-1-2022.
This question has ended, but is awaiting resolution by an admin.

Final Crowd Forecast
Time Period 90% chance to be above Crowd Forecast 90% chance to be below
2021 3.0 5.34 10.07
2022 2.46 5.43 11.13
2023 1.98 5.35 11.83
2024 1.48 5.76 10.73
Tip: Mention someone by typing @username