By Andrew Gunder, DCMME Graduate AssistantImage result for dueling AI

I know what you’re thinking. Dueling AI sounds a lot like a game of Rock ‘Em Sock ‘Em robots, or maybe you’ve conjured up images of the Siri and Alexa arguing with one another. Dueling AI refers to the capability that one AI engine can be used to train another AI engine.

Like humans, AI engines don’t start out smart right out of the gate. They must be educated over time. Take a chess playing robot for example, it may not win its first match or even its second or third, but after studying previous outcomes it begins to recognize patterns and tailor its strategy accordingly. Chess essentially has an infinite number of outcomes after several moves, so it is next to impossible for the AI to learn each move, it must learn instead.

Dueling AI isn’t just limited to robotics. Imagine the ramification on other industries or other possibilities. Suppose we are trying to determine between an authentic image and a digitally altered image. One AI attempts to create a realistic image, and another AI attempts to decide whether the image is real or artificial.

If AI is capable of learning and training another AI, you’re probably wondering what’s preventing it from achieving world domination. Fortunately for us, AI only becomes as smart as the data that it is fed. It possible to “poison” AI by feeding bad data, thus showcasing that as smart as AI can be it is not without limitations.


Can you think of some other tasks AI can train another AI on?

What industries can be impacted the most by dueling AI?

Do you foresee ways that AI might be able to overcome its limitations?