Performance of PAWS
Figure 6. PAWS Algorithm Defender Utilities Against Security Experts Playing as Poachers in Computer Game.
A survey administered at the end of the unit
addressed unit objectives. Approximately 79 percent
of learners were at least somewhat willing to adopt
PAWS (mean = 5.9 on a scale of 1 [strongly disagree]
– 7 [strongly agree], SD = 1.1). Additionally, open-ended responses also largely supported the purpose
of the PAWS software. Roughly half of respondents
(n = 15) commented that PAWS could optimize
patrols and would make the job of patrolling easier.
When asked about software limitations, respondents
recommended increasing complexity of models,
including approximately one-third (n = 10) of
respondents suggesting that dynamic animal distribution models be added. This latter point highlights
the challenges faced by AI researchers in accurately
representing all details of a given real-world scenario.
Regarding satisfaction with the unit, 86 percent
rating the learning experience as at least somewhat
useful (mean = 5.7 on a scale of 1 [completely use-less] – 7 [extremely useful], SD = 0.8), and more than
96 percent of respondents rated it as at least somewhat important (mean = 6.0 on a scale of 1 [extremely unimportant] – 7 [extremely important], SD = 1.0).
Additionally, more than 86 percent of respondents
reported that they were at least somewhat likely to
recommend it to peers (mean = 6.04 on a scale of 1
[strongly disagree] – 7 [strongly agree], SD = 1.04).
44 AI MAGAZINE
This article describes our unique approach that used
a real-world problem-to-project scaffolding framework to teach game-theoretic concepts to several
audiences. Learners included students at an urban
public high school, university undergraduate students, and law enforcement officers and park rangers
who protect wildlife in Indonesia. Our instructional
units began by presenting real-world problems in
wildlife and other security domains that painted the
broad picture for why learning security game concepts is important. Throughout the learning units,
techniques from project-based learning along with
instructional support were used to progressively
introduce complex AI concepts and help learners tie
learning activities to the larger real-world security
goals. Games were a key learning tool in our
approach. Members of all three audiences played the
role of two different actors as part of our games: (1)
playing as rangers, they generated defender strategies to protect against poachers’ attacks; (2) as
poachers, they attempted to outsmart the defender
strategies generated by their peers to earn the highest possible rewards. This approach not only gave
learners valuable hands-on experience with complex
AI concepts, but also in the development of real-world applications for security.
Participant feedback was consistently positive,
with the majority of participants from all three audiences rating the learning experiences as useful, and