There are numerous applications of Artificial Intelligence or AI since it is supposed to be a self learning system. Whatever you teach the system, it is supposed to understand and adapt to it, while learning some things additionally, which are not actually coded. There are many companies working on their own version of AI but Elon Musk founded OpenAI is now in the news for training a robot hand that can solve a Rubik’s Cube. While it might not seem like something big at first, but it is indeed a notable achievement.
Dactyl is a robotic arm developed by OpenAI, which was used to demonstrate increased dexterity in robotics. The company trained a pair of neural networks to solve a Rubik’s Cube using Dactyl. However, to understand what’s happening, we need to take a look at how general AI works. We currently use task repetition for training a neural network, which is basically letting it practise something in a virtual environment for years at a hastened pace.
Automatic Domain Randomization (ADR) to the rescue
Remember OpenAI Five? If not, then know that it is an AI system that was able to best some of the best DOTA 2 human players. It was trained in the game to learn its rules and formulate strategies by playing thousands of matches. While this method is suitable for such software based endeavours, you can’t really envision a training a robot hand to solve a Rubik’s Cube similarly for years.
Thus, OpenAI came up with some new simulations to train it. While simulations have already been used to accomplish such a task, they were not really precise and didn’t resemble real-world physics that well. Thus, the company developed a new method called Automatic Domain Randomization (ADR) that is capable of endlessly generating progressively more difficult environments in simulation. As per OpenAI, “This (ADR) frees us from having an accurate model of the real world, and enables the transfer of neural networks learned in simulation to be applied to the real world.”
Fear the robot uprising?
As dramatic and technical as it sounds, neural network aren’t anywhere near human speed and dexterity levels, especially when it comes to solving a Rubik’s Cube. Although, the demonstration by OpenAI is a good example of how simulations can be implemented and to lay the groundwork for general-purpose robots. “We believe that human-level dexterity is on the path towards building general-purpose robots and we are excited to push forward in this direction,” says OpenAI.
Image credits: OpenAI
Stay tuned to Silicon Canals for more European technology news.