In Episode 26 of my weekly update I’m sharing stories and insights around developments in AI including Facebook AI Research’s Pluribus AI defeating a group of poker champions, an algorithm that has uncovered brand new insights based on 3 million abstracts of science papers dating back to 1922, and AI created from a sheet of glass with no need for electricity or the internet.
In Episode 20, I shared insight around Google’s DeepMind, the creators of AlphaGo, the first computer program to defeat a Go world champion (a landmark triumph watched by over 200 million people worldwide). In the latest achievements of AI trained to play and beat humans at their favorite games, Facebook AI Research has developed Pluribus, which has beaten some of the world’s top Texas Hold’em poker players. Pluribus’ core strategy involved the AI playing against copies of itself (self-play), a common technique for training AI. Pluribus’ predecessor Liberatus used a large number of servers/GPUs (15 million core hours) to develop its strategies and 1,400 CPU cores during live game play, whereas Pluribus completed the blueprint of its strategy in eight days using merely 12,400 core hours and only 28 cores during live game play.
Moving on from AI’s newly discovered and growing influence over the way we view some of our favorite games, we now see how AI can help increase the scope of previously learned knowledge. Over 3 million science paper abstracts from over 1000 journals dating back to 1922 were collected and fed into the Word2vec algorithm. The results? Truly astounding: discoveries of new materials and suggestions of as-yet-unknown materials. Without training the AI in the subject matter of materials science, the algorithm was able to analyze relationships between words in the science papers and deliver previously inconceivable insights. This breakthrough is an important reminder that the impact of AI is not simply limited to defining future innovation, but that it has the capability to reach back into the past to reconfigure traditional industries as we know them.
Imagine facial recognition, but without the need for electricity or the internet. This is one of the implementations being explored with artificially intelligent glass. As I have shared previously, AI requires considerable computer resources and power for high-end processing. But, researchers have now developed new methods using ‘smart glass’ that requires no circuitry. Perhaps the applications may not be as extensive as it doesn’t need computers. However, smartphone security and other functionality are being proposed as possible uses. Just when you thought peak levels of AI innovation had been achieved, I wanted to do my part to ensure your mindset remains open to the changes bring brought about as a result of exponential technologies.