Page 40 - GSATJanuary2020

1E +7 (A) Computing Power Trend 100 Human Performance
95
Transistors 1E +6 % Accuracy 85
90
80 (E) Machine Learning Image
1E +5
1E +4 75 Recognition Capability Trend
70
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018
Year 100 Human Performance Year
% Global Population 50 (B) Internet Users Trend % Accuracy 95 Language Understanding Trend
60
40
90
85
30
80 (F) Machine Learning Natural
20
75
10
0
2004
2006
2008
2000
2002
Year 2010 2012 2014 2016 2018 70 2000 2002 2004 2006 2008 Year 2010 2012 2014 2016 2018
20 (C) Internet Connected Devices Trend 100 Human Performance
Devices (Billions) 10 5 % Accuracy 90
95
15
85
80 (G) Machine Learning
70
0
2004
2006
2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 75 2000 Speech Recognition Trend 2008 2010 2012 2014 2016 2018
2002
Year 0.5 Year
MOOC’s (Thousands) 10 (D) Massive Open Online Courses Trend % of References 0.4 (H) Machine Learning
12
8
0.3
6
0.2
4
0.1
2
Geoscience Literature Trends
0
0
2000
2004
2002
2006
2008
Year 2010 2012 2014 2016 2018 2000 2002 2004 2006 2008 Year 2010 2012 2014 2016 2018
Figure 1. (A) Number of transistors in an integrated circuit (Rupp, 2018). (B) Worldwide Internet users (World Bank, 2018). (C) Number of global Internet-
connected devices (Mercer, 2017). (D) Number of massive open online courses (MOOC’s) offered (Shah, 2018). (E) Machine learning image recognition
capability (Russakovsky et al., 2015). (F) Machine learning natural language from text understanding (Zilly et al., 2016). (G) Machine learning speech recogni-
tion (Saon et al., 2017). (H) Percentage of geoscience publications with machine learning in the abstract, title, or keyword.
future students who become exposed to the data or black box algorithms used in an anal- passion for their work. Technological dexter-
geosciences through on-campus classes. ysis may be of questionable quality. Until ity will certainly bring additional value to
General education geoscience courses, taught skilled geoscientists can “crack” the codes our field, but we believe it will be the combi-
by passionate faculty and often supplemented and truly understand algorithm mechanics nation of deep fundamental geoscience
with field trips, are an important tool to recruit and limitations, this problem will remain. In knowledge and digital fluency that will be
new geoscience students. We think it is response, a growing number of journals are the foundation for the next era of geoscience
important that universities not lose their reinforcing new best practices, such as pub- innovation and discovery.
emphasis on field and lab work or their com- lishing codes and raw data. We assert that it
mitment to undergraduate research. These is the role of the geoscience community to ACKNOWLEDGMENTS
unique, high-impact learning experiences can establish standard techniques and other best We thank numerous colleagues for helpful discus-
only happen in person and are essential to practices to solidify the correct use of popu- sions that informed the ideas in this paper. This manu-
mentoring students in our discipline. lar new technologies. script was improved due to the helpful reviews by the
GSA Today editor and an anonymous reviewer.
Another point of caution is the potential
erosion of key geoscience skills from an CONCLUSIONS REFERENCES CITED
over-reliance on digital technology. This has The geoscientists that are most likely to Alaudah, Y., and Al Regib, G., 2016, Weakly-super-
been recognized as a potential risk in petro- thrive in this new technological environment vised labeling of seismic volumes using reference
exemplars: Image Processing (ICIP), Sept. 2016
leum geology for more than a decade are those willing to be agile and remain in a IEEE International Conference, p. 4373–4377.
(Yeilding, 2005), as subsurface interpreters state of continuous learning. Maintaining Deming, D.J., Lovenheim, M., and Patterson, R.,
began to rely heavily on workstation-pro- static methods of teaching and practice will 2018, The competitive effects of online education,
duced maps that often provide geologically be insufficient. Paradoxically, despite the in Hoxby, C.M., and Stange, K., eds., Productivity
unrealistic solutions. If virtual field trips and need for digital comprehension and capabil- in Higher Education: Chicago, University of Chi-
cago Press, 392 p.
digital map-making become students’ pri- ity, the most important functional knowledge De Paor, D.G., 2016, Virtual rocks: GSA Today,
mary exposure to geologic mapping, the and skills that any geoscientist should pos- v. 26, no. 8, p. 4–11, https://doi.org/10.1130/
problem may grow even worse. sess will likely remain the same. These GSATG257A.1.
ML applications in geoscience present include a deep understanding of fundamen- Di, H., Shafiq, M., and Al Regib, G., 2018, Multi-at-
some unique challenges. Insufficient train- tal Earth processes, an ability to creatively tribute k-means clustering for salt-boundary de-
lineation from three-dimensional seismic data:
ing data and poor experimental designs can integrate data from various sources, the clar- Geophysical Journal International, v. 215, no. 3,
lead to erroneous conclusions. Open-source ity to communicate difficult concepts, and a p. 1999–2007, https://doi.org/10.1093/gji/ggy376.
40 GSA Today | January 2020