Every week, my team at Invector Labs publishes a newsletter to track the most recent developments in AI research and technology. You can find this week’s issue below. You can sign up for it below. Please do so, our guys worked really hard on this:
From the Editor
Chess has long been considered a solved problem in the context of artificial intelligence(AI). Since IBM’s DeepBlue defeated Garry Kasparov in 1998, AI-based chess engines have been getting better and better to the point of being able to play games near to perfection. However, while AI-based chess have excels at complicated tactical maneuvers it is still lacking the strategic creativity of the human mind. Its hard for an AI-powered chess engine to sacrifice a pawn in order to gain a better position or to go into an endgame with material disadvantage but better position. This is partially due to the fact that most chess engines derive knowledge from evaluating massive libraries of openings, middle game positions and end games.
Recently, DeepMind’s AlphaZero took a different approach to AI-based chess building an engine based on reinforcement learning that, essentially, mastered the game by playing millions of games against itself. AlphaZero has not only become the most powerful chess engine in the world but one of the few ones that exhibits a creative, attack-prone style that resembles many of the greatest players in history like Kasparov, Fischer or Capablanca. This week, DeepMind unveiled some evaluation results of AlphaZero clearly showing how reinforcement learning is a potential path to mimic human creativity. As Garry Kasparov wisely said in a recent article, “AlphaZero shows us that machines can be the experts, not merely expert tools”.
Now let’s take a look at the core developments in AI research and technology this week:
AI researchers from introduced a technique from improving the performance of simple machine learning models using knowledge from more sophisticated models.
Microsoft AI researchers proposed a new method for training machine learning models in a way that imitates human teaching.
ElementAI proposed a new probabislistic model for better to improve the adaptability of AI agents in uncertain environments.
Cool Tech Releases
PyTorch 1.0 is now generally available and is showing impressive growth metrics.
DeepMind showcased how their AlphaZero system was able to master Chess, Go and Shogi(Japanese chess).
OpenAI introduced CoinRun, a new environment for training reinforcement learning agents.
AI in the Real World
Alphabet’s subsidiary Waymo unveiled a new autonomous taxi service called “Waymo One”
Chess legend Garry Kasparov published a very thoughtful article about the breakthrough that DeepMind’s AlphaZero represents for the chess world.
Nvidia uses AI to learn elements from videos and recreate entire virtual world.