Technology Innovation and Great Power Competition,TIGPC, Gordian knot Center, DIME-FIL, department of defense, dod, intlpol 340, joe felter, ms&e296, raj shah, stanford, Steve blank, AI, ML, AI/ML, china, Wargames
Team Wargames - 2022 Technology, Innovation & Great Power Competition
1. Lessons Learned
Original Statement
The U.S. needs a way, given
a representative simulation, to
rapidly explore a strategy for
possible novel uses of
existing platforms and
weapons.
David Song
B.S. CS
Yujing Zhang
M.A. East Asian Studies
Shashank Rammoorthy
M.S. CS
Jonathan Sepulveda
M.S. Materials Science
Project Wargames
30+
interviews
Current Statement
Strategic wargames stand to
benefit from a stronger integration
of AI+ML but are struggling to find
adoption and usage. How can this
be addressed?
1
2. Lessons Learned
Defining a Problem: Research
Weeks 2 3 4 5 6 7 8 9
Scavenging the Solution Space: Testing
Future
Exploring the Problem
2
6. Lessons Learned 6
So we went ‘back to the drawing board’...
Picking one case study of interest, we
focused on the Cross-Straits scenario…
● We visited talks on Taiwan’s
Economic Security to understand
variables often overlooked
● The sad reality is, wargame players
often have very little time to
anticipate or prepare for wargames,
and are not concerned with non-
combatant variables
● The most important element of
wargames ought to be to teach
people to think creatively
7. Lessons Learned
We realized wargaming was not easy
Then we started learning and asking…
● What are CoAs?
● Differences between wargaming vs.
M&S?
● Why does the DoD want innovation
within such traditional processes?
● What hasn’t been attempted
already?
7
Excerpt from Battle! Practical Wargaming by Charles Grant
8. Lessons Learned
Problems we found with wargaming
8
We immediately found out that there are…
● Many moving parts & elements:
researchers → designers →
wargamers → Red and Blue teams
● Little data is shared amongst
designers and team members
● There are multiple types of
wargames
● Most games remain TTXs
● Too difficult to build de novo a
robust, AI-capable Red adversary
That left us thinking:
● Virtual wargames are often used for
training, but seldom as formal
proceedings; why?
● Should we seek to replace
wargamers altogether with AI+ML
‘magic’?
9. Lessons Learned
What some of our interviewees told us…
9
Sebastian Bae
CNA Corp. | Wargamer
Yuna Wong
IDA | Defense Analyst
Jacquelyn Schneider
Hoover, CISAC | Researcher
Pete Pellegrino
U.S. NWC | Game Design Lead
"There are a lot of reasons for
why AI/ML hasn't been used in
wargaming in the DoD…technical
illiteracy within the DoD…policies
are inflexible, and systems are
difficult to change"
10. Lessons Learned
What some of our interviewees told us…
10
Sebastian Bae
CNA Corp. | Wargamer
Yuna Wong
IDA | Defense Analyst
Jacquelyn Schneider
Hoover, CISAC | Researcher
Pete Pellegrino
U.S. NWC | Game Design Lead
"We should never use new
wargames, but rather well-studied
wargames and already-studied
ones in predicting our outcomes &
enhancing our adjudications”
11. Lessons Learned
What some of our interviewees told us…
11
Sebastian Bae
CNA Corp. | Wargamer
Yuna Wong
IDA | Defense Analyst
Jacquelyn Schneider
Hoover, CISAC | Researcher
Pete Pellegrino
U.S. NWC | Game Design Lead
"Wargames need data to eat from.
You need a volume of data that
AI+ML can learn from and be helpful;
where is that data coming from a
wargame? Wargames don’t generate
simulation data in copious amounts”
12. Lessons Learned
What some of our interviewees told us…
12
Sebastian Bae
CNA Corp. | Wargamer
Yuna Wong
IDA | Defense Analyst
Jacquelyn Schneider
Hoover, CISAC | Researcher
Pete Pellegrino
U.S. NWC | Game Design Lead
"We put too much emphasis on the
Red...players are generally not from
Russia or China, and so you struggle to
extrapolate specific country foreign
policy decisions versus general human
reactions to international crises.”
Many different objections to
AI+ML!
13. Lessons Learned
We got out of the building, went to an NPS Conference…
13
● Existing AI+ML capabilities have to be
better leveraged
● AI+ML does not need to make drastic
technical advances to be feasible for
wargaming uses
● There’s no readily accessible central
repository of past wargame data
outcomes to help future teams
● Ideally, forecast decision spaces as
opposed to having AI+ML play moves
15. Lessons Learned
Defining a Problem: Research
Weeks 2 3 4 5 6 7 8 9
Scavenging the Solution Space: Testing
Future
15
Pivot: Let’s help humans
play wargames better with AI
tools instead of replacing
humans completely with AI
16. Lessons Learned
Defining a Problem: Research
Weeks 2 3 4 5 6 7 8 9
Scavenging the Solution Space: Testing
Future
16
Identifying
Recommendations
17. Lessons Learned 17
1. Digitize wargames to increase iterations
"Accelerating and making
wargames more efficient will
come down to a series of
technologies that could very
well be unclassified in
development”
Major defense startup
company
18. Lessons Learned 18
2. Create AI tools for digitized wargames
“Not many people in the U.S.
have access to the decision-
making teams you find in Red
teams [the PLA and the CCP].”
Elizabeth Bartels | RAND
19. Lessons Learned 19
3. Implement organizational changes
We read-up on traditional approaches to
building wargames, noting its strengths..
● We found more support that cross-talk
between wargamers & designers and
players is important
● Academics do not differ that sharply
from the assessment of wargame players
insofar as AI+ML integration is concerned
● Information asymmetry predominates
within wargames–not an attribute of UCC
commanders
● Need more collaborations among
wargamers, designers, combat &
command
20. Lessons Learned
Defining a Problem: Research
Weeks 2 3 4 5 6 7 8 9
Scavenging the Solution Space: Testing
Future
20
Next steps
1. Validate our recommendations
2. Shadow a wargame
3. Continue working with our sponsor