Spotlight article “The story of statistics in geotechnical engineering”

Georisk is pleased to present a Spotlight paper “The story of statistics in geotechnical engineering” by Kok-Kwang Phoon. Georisk launched the “Spotlight” series in 2013. The purpose of this series is to invite distinguished scholars to review an important topic, to highlight research gaps, and to suggest fruitful research directions. Professor Kok-Kwang Phoon is Distinguished Professor at the National University of Singapore, and one of the pioneers on statistical and other data-driven methods in geotechnical engineering.

The aim and scope of Georisk was revised in 2018 to cater to the rapid pace of research in data analytics and data-driven methods: “Many opportunities to leverage on the rapid advancement in Bayesian analysis, machine learning, artificial intelligence, and other data-driven methods also exist, which can greatly enhance our decision-making abilities”. In an era in which data is recognised as the “new oil”, it makes sense for us to lean towards decision making strategies that are more responsive to data (Phoon 2020). Georisk is publishing an increasing number of papers on compilation and novel applications of databases to better support decision making (e.g. Bilgin, Arens, and Dettloff 2019; Feng and Vardanega 2019; Phoon and Tang 2019a, 2019b; Uzielli and Mayne 2019). It is anticipated that these papers will shift the research agenda and influence the practice of decision making in the face of uncertainty in the not too distant future. Other recent related developments include the establishment of ISSMGE TC309 “Machine learning and big data” and the launch of the database sharing initiative 304 dB by ISSMGE TC304 “Engineering practice of risk assessment and management”.

The publication of this Spotlight paper is timely. It covers the estimation of useful statistics from the original classical univariate setting to a more realistic multivariate setting encountered in a typical site investigation programme. Generic geotechnical databases are surveyed first to set the stage. The characteristics of geotechnical data are succinctly described by Prof. Phoon as MUSIC: Multivariate, Uncertain and Unique, Sparse, Incomplete, and potentially Corrupted. The paper then traces the history of how uncertainties are managed as the prevailing paradigm influencing the way we ask and answer questions shifts with time (Kuhn 1970). A deterministic world view necessarily requires characterisation using crisp numbers (single points on the real line). Under this paradigm, uncertainty is handled by adopting conservative values that are crisp numbers. It is not possible to talk about “correct” conservative values – they are all ad-hoc in nature. This paradigm remains highly influential in geotechnical engineering practice. The probabilistic paradigm underlying reliability-based design allows uncertainties to be characterised explicitly as random variables or random fields. It is possible to validate probabilistic models more formally using model identification/selection methods. Nonetheless, this does not imply that the “true” uncertainty is probabilistic in nature. It could be imprecise, fuzzy, or others. The current machine learning paradigm looks at data as an asset in itself for supporting decision making. It does not presuppose the uncertainty (or other “imperfect” MUSIC data characteristics) as intrinsically bad and thus requiring reduction or mitigation. The salient philosophical point presented in this paper is that the notion of uncertainty being “bad” is related to the deterministic paradigm that privileges crisp numbers as “ideal” numbers. Prof. Phoon advocates the pragmatic approach of allowing data to speak for themselves in terms of demonstrating their concrete value to decision making, rather than privileging some data characteristics as intrinsically good or bad because of their fit or lack of fit to a prevailing paradigm/model. The new generation of researchers has the opportunity to step out of the boundaries imposed by existing paradigms to uncover novel values, be it using white box, black box, or grey box data-driven methods. His vision of how the story of statistics will unfold is exciting, including his exhortation to initiate a flagship “AlphaGeo” project that will ignite our imagination in the same way the Google “Alpha Go” project did to AI researchers. Prof. Phoon concluded with an intriguing question on the role of human judgment in an immersive cyber-physical reality. The insights of this Spotlight paper will be far reaching.

ORCID

Limin Zhang http://orcid.org/0000-0001-7208-5515