Science funding agencies (NASA, DOE, and NSF), the science community, and the US taxpayer have all benefited enormously from the several-decade series of National Academies Decadal Surveys. These Surveys are one of the primary means whereby these agencies may align multi-year strategic priorities and funding to guide the scientific community. They comprise highly regarded subject matter experts whose goal is to develop a set of science and program priorities that are recommended for major investments in the subsequent 10+ years. They do this using both their own professional knowledge and by synthesizing details from many thousands of existing and solicited documents.
Congress, the relevant funding agencies, and the scientific community have placed great respect and value on these recommendations. Consequently, any significant changes to the process of determining these recommendations should be scrutinized carefully. That said, we believe that there is currently sufficient justification for the National Academies to consider some changes. We advocate that they supplement the established survey process with predictions of promising science priorities identified by application of current Artificial Intelligence (AI) techniques These techniques are being applied elsewhere in long-range planning and prioritization.
We present a proposal to apply AI to aid the Decadal Survey panel in prioritizing science objectives. We emphasize that while AI can assist a mass review of papers, the decision-making remains with humans. In our paper below we summarize the case for using AI in this manner and suggest small inexpensive demonstration trials, including an AI/ML assessment of the white papers submitted to Astro2020 and backcasting to evaluate AI in making predictions for the 2010 Decadal Survey.
[1]
Geoffrey E. Hinton,et al.
Dynamic Routing Between Capsules
,
2017,
NIPS.
[2]
Jonathan Krause,et al.
Using deep learning and Google Street View to estimate the demographic makeup of neighborhoods across the United States
,
2017,
Proceedings of the National Academy of Sciences.
[3]
Olga Kononova,et al.
Unsupervised word embeddings capture latent knowledge from materials science literature
,
2019,
Nature.
[4]
J. D. Huba.
Comments on Best Practices for a Future Open Code Policy for NASA Space Science
,
2018
.
[5]
Geoffrey E. Hinton,et al.
Matrix capsules with EM routing
,
2018,
ICLR.
[6]
Mario Krenn,et al.
Predicting research trends with semantic and neural networks with an application in quantum physics
,
2019,
Proceedings of the National Academy of Sciences.
[7]
Lutz Bornmann,et al.
Growth rates of modern science: A bibliometric analysis based on the number of publications and cited references
,
2014,
J. Assoc. Inf. Sci. Technol..
[8]
Beta Writer.
Lithium-Ion Batteries
,
2019
.
[9]
Yiming Yang,et al.
XLNet: Generalized Autoregressive Pretraining for Language Understanding
,
2019,
NeurIPS.