We are in the midst of the Artificial Intelligence Revolution (AIR), the next major epoch in the history of technological innovation. Artificial intelligence (AI) is globally gaining momentum not only in scientific research, but also in business, finance, consumer, art, healthcare, esports, pop culture, and geopolitics. As AI becomes increasingly pervasive, it is important to examine at a macro level whether AI gains from disorder. Antifragile is a term and concept put forth by Nassim Nicholas Taleb, a former quantitative trader and self-proclaimed “flâneur” turned author of New York Times bestseller of “The Black Swan: The Impact of the Highly Improbable.” Taleb describes antifragile as the “exact opposite of fragile” which is “beyond resilience or robustness” in “Antifragile: Things That Gain From Disorder.” According to Taleb, antifragile things not only “gain from chaos,” but also “need it in order to survive and flourish.” Is AI antifragile? The answer may not be as intuitive as it seems.

The recent advances in AI are largely due to the improvement in pattern recognition abilities via deep learning, a subset of machine learning, which is a method of AI that does not require explicit programming. The learning is achieved by feeding data sets through two or more layers of nonlinear processing. The higher the volume and faster the throughput processing of data, the faster the computer learns.

Faster processing is achieved mostly through the parallel processing capabilities of the GPU (Graphics Processing Units), versus the serial processing of CPUs (Central Processing Unit).  Interestingly, computer gaming has helped accelerate the advancements in deep learning, and therefore also play a role in the current AI boom. GPUs, originally mostly used for rendering of graphics for computer games, are now an integral part of deep learning architecture. To illustrate, imagine there are three ice-cream carts with customers lined up at each one, and only one scooper. In serial processing, the scooper aims to finish serving all of the carts at the same time and does so by bouncing between carts scooping out a few cones at a time before servicing the next. In parallel processing, there are multiple scoopers, instead of just one. A savvy customer will divide up the order among the carts at the same time to achieve faster results.