Deep Learning to Take AI to Next Level

By Ernest Worthman, IEEE Senior Member

January 11, 2018

What will be the biggest bump for AI is something called “deep learning.” AI will go from the typical if X = Y, then C, else D, of typical smart devices to artificially intelligent systems (AI 2.0) that will actually learn, suggest and automate processes by analyzing patterns and behaviors. Deep learning will become an integral element of AI and the major factor that will take AI to the next level.

Perhaps the latest example of that is an AI-based computer system, called Libratus. It has mastered a game once considered too difficult for AI. This poker bot has played thousands of games of heads-up, no-limit Texas hold’em against a cadre of top professional players at Rivers Casino in Pittsburgh. Moreover, it beat them all!

Why this is a big deal is because poker isn’t just analytics – it requires both reasoning and intelligence, even emotions, of the type that has proven difficult to impossible for machines to imitate. Because an opponent’s hand remains hidden, it is extremely complicated to use logic in developing the perfect strategy, regardless of the complexity of the algorithm, or depth of the database, of several possible approaches your opponent can take. And, this is a classic example of how nonlinear elements such as body language and human-type actions/reactions are part of the advanced AI equation.

The “gotcha” here is that in no-limit Texas hold’em, there is no single “correct” play, as in chess, for example. Therefore, it is virtually impossible to run all possibilities, regardless of resources. Instead, the AI must use “game theory” to calculate optimal plays presented by the uncertainties, weight them and pick the most likely opponent scenario.

While this is only a game, the implications are huge. It shows that AI does have the potential to “think” and use logic in the fuzzy realm.