RUNNING HEAD: Artificial Intelligence to Improve US Military Tactics
Development of Artificial Intelligence to Improve US Military Tactics
Faris Syed & Joshua De Vera
Progress Report VIII
4/14/2020
Objectives:
Within the previous two weeks, our group has continued to refine the artificial
intelligence algorithm with more test data and working towards a more accurate project.
Likewise, we’ve worked on cleaning up the code and adding restrictive features to prevent
incorrect inputs and improve useability.
1
Methods and Materials
Similar to how we’ve previously tested data, the collection of specific battles given
particular circumstances to fit our constants has been used to modify our coefficients such that
our program functions similarly to the results which occurred throughout history. This is our
final stage of testing given the current state of the program and the coefficients will remain fairly
constant throughout the remainder of the project.
Figure 1. The following image is the final coefficient page which we will be using for the
remainder of the project. These changes are based on the final battles which we recently tested at
which we feel we have enough data to create a reliable program.
Likewise, we wanted to continue cleaning up our code to make it as efficient and
user-friendly as possible with minimal errors while executing the code. Similar to the Country
class, we introduced regular expressions, or regex, into our code in order to limit the inputs the
user can enter such that the program doesn’t have a syntax failure. One instance of this was in
our getArmyResources() method which initially checks whether the input is a number before
executing the remainder of the code. We implemented this similar idea to the other methods
which get resources from the user and computer.
2
Figure 2. The following method in our class represents one of the many introductions of regex
into our code to help the program execute without syntax errors by limiting the data which enters
the methods.
In a similar way as seen in Figure 2, we utilized regex when determining the user’s
moves so they can only input maneuvers that we have implemented into the program and
prohibit them from using numbers or special characters. All of these have been implemented into
our current program and allows the main function to run smoother despite mistyped inputs.
3
Data and Analysis
With our new coefficients, we’ve tested over 75 battles. Based on those inputs, we’ve had
the program reproduce similar results based on those historical battles and once put into a
random battle, the win rate gradually increased with each battle tested. The final average win rate
of the artificial intelligence program was about 82% given all the final game coefficients.
Figure 3. The following graph represents the win rate of the artificial intelligence program after
increasing the number of test battles which also increased the accuracy and reliability of the
algorithm. Based on the first 70 battles tested, we reached almost a 76% average win rate, which
surpassed our goal of at least 75%.
At the regional science fair, judges along with other people were able to give us a lot of
feedback on our analysis which we’ve analyzed since then. The most notable input was the lack
of modern data as the most recent data we’ve collected was from almost 20 years ago
(Afghanistan War conflicts). Given the complexities of modern warfare, such as alliances and
less data open to the public , we weren’t able to achieve reliability in terms of current conflicts as
4
of yet since our current program focuses heavily on one-on-one confrontations. However, the
judges provided insight on the possibility of quickly implementing the use of alliances using
neural networks and enough data given the countries we currently have. Given our time frame,
we don’t feel like we can create and analyze a neural network before the research symposium,
but it is something we are interested in expanding on in the future. Another suggestion we
received was turning the program we have into a game to help train potential military leaders,
but also to reach out to a broader audience with the game aspect. This is something we may
consider doing since it can be quickly implemented and tested for useability very quickly.
5
Future Plans
As we start to finish the project, we would like to try to improve the project with
the constructive criticism we received from judges at the regional science fair which
include possibly changing the functionality of the program to a game instead of a single
battle simulator to connect to broader audiences or possibly branching out with our
artificial intelligence algorithm (from a linear model to a system similar to neural
networks). We’d also like to finish cleaning up the entire program and try to remove as
many possible syntax errors in our code.
6
References
Battle, A. (2016). Tactics Tutorial. Retrieved April 1st, 2020, from
http://www.theartofbattle.com/tactics-tutorial/
Freedberg, S.(2019). Simulating A Super Brain: Artificial Intelligence In Wargames. Retrieved
March 26, 2020, from
https://breakingdefense.com/2019/04/simulating-a-super-brain-artificial-intelligence-in-wargame
s/
Klein, B. (2011). Neural Networks. Retrieved March 14, 2020, from
https://www.python-course.eu/neural_networks.php
