AI: Not Only Big Data

tomerefrat
Dec 20, 2017
2 min read

The necessary and fruitful connection between machine learning and huge data bases has been clearly demonstrated in the booming field of Artificial Intelligence. The tech giants employ skillfully big data in order to improve and create sophisticated products using machine learning. Such projects are the autonomous cars, voice AI based programs and image processing.

However, the progress in Machine Learning is not only based on big data and IOT, a growing in importance branch is Reinforcement Learning. Included among the “10 breakthrough technologies in 2017”, by the MIT Technology Review, this field generates AI based on an internal process of trial and error embedded into the system. Following a defined set of rules integrated into the computer, the system starts acting accordingly, receiving feedbacks as “rewards” or errors along the process. Hence, the more the machine “exists” and interacts with its environment, the better, “cleverer” and more sophisticated it gets.

This has been demonstrated with AlphaZero, an AI program created by DeepMind using a reinforcement learning structure. After only several hours, AlphaZero managed to master well beyond any other game playing computer (or human) Chess as well as the Japanese game “Go”.

Not only in the game area Reinforcement Learning is being explored. Currently it is also investigated in discovering energy materials as this is an ongoing changing landscape. Using reinforcement learning algorithms enable a quicker research process of molecules and their environment without defining each step separately. Also in the field of protein folding analysis, applying Reinforcement Learning is a breakthrough. The characteristic of protein structures appearing in nature is diverse and huge in quantity. This makes it's comprehensive investigation basically impossible. By using Reinforcement Intelligence this task becomes manageable and feasible.

Hence, aside of supervised and unsupervised Machine Learning there is a lot to expect from AI through Reinforcement Learning.