Dr. Yuna Wong Profile picture
Defense analyst at IDA, President of Women's Wargaming Network, adjunct professor at Georgetown. Standard caveats. Yuna Huh Wong on LinkedIn.

Jul 30, 2020, 9 tweets

I’m running the working group at this year’s Connections Wargaming Conference on Enhancing Wargaming through AI/ML. (AI = artificial intelligence, ML = machine learning). Here are some thoughts about the topic. [1/9]

Why this topic to begin with? DoD’s focus on introducing AI into operating forces & concepts + DoD’s renewed emphasis on #wargaming makes this a natural subject of interest. Currently there is very little ML in defense wargames. [2/9]

Two distinct problems are within this topic of AI and wargaming: 1) representing AI in wargames, and 2) using AI in defense wargames. Entire articles could be written about #1 but our WG will deal with #2. [3/9]

What are potential benefits of using AI in wargames? One: if you expect U.S. forces to face off against adversaries with AI-enabled decision making in future conflicts, start exposing them to actual AI-enabled decision making now. [4/9]

The first time U.S. forces encounter an AI-enabled adversary should NOT be in real life. We need real AI in wargames because humans simulating AI can only go so far. There is even value in realizing you have no idea why an adversary AI is doing something. [5/9]

Two: training against an AI can considerably increase human skill. We’ve seen the rise of training AIs dedicated to improving human performance in games like go. I find this a very interesting type of human-machine collaboration. [6/9]

Three: AI adversaries could give us entirely new ideas. Deep Mind’s tantalizing success with GAN to train AlphaGo Zero (with no human knowledge) to beat AlphaGo (trained on human games) makes us wonder what human players wouldn’t think of. [7/9]

Four: algorithms trained to maximize different rewards could yield different but consistent “adversaries” for humans to train against. (Human red cells vary by who staffs them.) Repeated training against a range of adversaries could help teach adaptability. [8/9]

Yet considerable challenges remain before we can utilize ML in wargames. Games need to be rigidly adjudicated, simple enough for the machine, AND still useful to professionals. Lots of work required. [9/9]

Share this Scrolly Tale with your friends.

A Scrolly Tale is a new way to read Twitter threads with a more visually immersive experience.
Discover more beautiful Scrolly Tales like this.

Keep scrolling