10 AI MAGAZINE
“Context-Driven Proactive Decision Support for
Hybrid Teams,” by Manisha Mishra, Pujitha Mannaru,
David Sidoti, Adam Bienkowski, Lingyi Zhang, and
Krishna R. Pattipati, was written specifically for the
maritime domain. In the article, they describe their
key challenge as identifying the context within which
humans interact with a smart Internet of Things.
They define context interdependently as the evolving
multidimensional feature space consisting of a ship’s
mission and its goals, assets, threats and tasks, and the
cognitive states of its commander and human opera-
tors working as a team in uncertain environments
while at sea, including hybrid human-machine teams.
They have created and validated an operational sys-
tem for proactive decision making amid a host of
technical challenges posed by the integration and
allocation of assets and tasks for an Internet of Things
using AI to determine context and achieve superior
performance. In addition, they provide more details,
mathematics, and descriptions about a user test bed
in supplementary material online.
In their article “Identifying Critical Contextual
Design Cues Through a Machine Learning Approach,”
Missy Cummings and Alex Stimpson of Duke University review the safety and productivity benefits
of autonomous technologies with a goal of understanding how autonomous systems can be better
designed to improve the interactions between humans
working with or around autonomous systems. These
safety critical systems generate immense amounts
of data. The authors review and use ML to design
a human-user interface. They evaluate a proposed
pedestrian signaling display mounted on a driverless car through traditional inferential statistics that
looked at broad population characteristics, finding no
significant relationships. Instead, by paying attention
to individual user characteristics using a ML clustering approach, they uncover critical contextual cues
that have led to improved reaction times for one variant of the pedestrian signaling display.
Brian Jalaian, Michael Lee, and Stephen Russell
address how different sources of uncertainty affect
the interpretations of contexts differently when using
different ML methods in “Uncertainty Quantification in Machine Learning.” They start with a review
of basic statistics, observational errors, models and
errors, and optimizations under uncertainty. Then
they review an autonomous mission command architecture, statistical learning and stochastic optimization, and uncertainty in ML. Finally, they address
four sources of uncertainty — noise, parameter
uncertainty, the uncertainty in model specification,
and the uncertainty from extrapolation — providing
readers with a nonparametric Bayesian model that
graphically portrays the uncertainty in the model.
They address the motivation to resolve these uncertainties as critical to the application of ML in the
field, where erroneous forecasts may put lives at risk.
The final article, written by Erik Blasch, Robert
Cruise, Alex Aved, Uttam Majumder, and Todd Rovito,
proposes a method of decisions to data that provides
a path to establish the value of data foraging, collect-
ing data, and sense making, using AI with human
reasoning to assess the context for complex sets of
data. Their model addresses data in various states
(rest, motion, fusion, transition, and use). They
review AI dynamic data-driven applications systems
and IF and how these paradigms align with AI; rea-
soning contexts; types of ML; and applications for
situational understanding that serve to achieve
human-machine awareness. To determine the context
of a dynamic target, they provide an example with
AI and deep multimodal image fusion where the data
are collected in multiperspectives from a command-
guided swarm.
Conclusions
We hope that readers enjoy all six of the articles
contributed on the topic of artificial intelligence,
autonomy, and human-machine teams: interdepend-
ence, context, and explainable AI. We also hope that
readers will join us at a future AAAI symposium on
the topic. The advent of human-machine teams has
created a time of intellectual ferment, extraordinary
technological advances, and the introduction of in-
terdependence to mathematicians, physicists, and AI
theorists and practitioners.
Notes
1. Merriam-Webster, s.v. “context,” www.merriam-webster.
com/dictionary/context.
2. en.oxforddictionaries.com/definition/context.
3. Russian deaths have been denied, and American involvement has been denied.
4. American Electric Power Co., Inc., et al., Petitioners v. Connecticut et al., 564 U.S. 10-174 (2011), www.supremecourt.
gov/opinions/10pdf/10-174.pdf. Justice Ginsburg wrote
the unanimous opinion.
5. For example, Nick Saban, the coach of the University of
Alabama football team, has never lost a game against his
former assistant coaches who had taken coaching jobs at
competing schools (Kirshner 2018); there have been many
failed attempts to clone Silicon Valley (Lucky 2014); and
despite studies on how to recreate the innovation culture
at Bell Labs (Kelly and Caplan 1993) existing in a facility
where the work for several Nobel awards was completed,
the facility has since closed (Martin 2006) and the lab has
been renamed Nokia Bell Labs.
6. For more on these issues, see aaai.org/Symposia/Spring/
sss19symposia.php#ss01.
7. Personal communication with C. Sibley, May 26, 2009.
References
Adelson, E. H. 2000. Lightness Perceptions and Lightness
Illusions. In The New Cognitive Sciences, 2nd ed., edited by
M. Gazzaniga. Cambridge, MA: MIT Press.
Allison, G. 2018. F- 35 to Incorporate Automatic Ground
Collision Avoidance System ‘Five Years Earlier Than Planned.’
UK Defence Journal (February 2). ukdefencejournal.org.uk/
f-35-incorporate-automatic-ground-collision-avoidance-
system-five-years-earlier-planned/
