The artificial intelligence is like a small child, his way of discovering the world is what is the scope and tools that offer to open to the world. Google knows that with DeepMind, artificial intelligence platform, go step by step, and today are adopting a learning method based education used in children.
Although DeepMind is already a power play and won the Go, plus it was already preparing for his next challenge in StarCraft II, the platform should start with the basics, which is able to identify objects and their properties in order to take Other types of decisions based on experiences.
A children’s play
Young people pull, push, break, and generally experiment with the physical world from a very early age, this is the best way to learn and Google knows, as the managers of DeepMind along with researchers from the University of California, Berkeley, they are adopting this method of learning now adapted to virtual environments.
The aim is that DeepMind learn about the properties of physical objects to interact with them, something that is known as ‘learning deep reinforcement’, which allows to solve tasks without specific instructions, something very similar to animals when they act on instinct, or Well, like humans when they face a problem they do not know.
You may also like to read another article on iMindSoft: How many hours should kids play with video games, according to science
The researchers created two situations so that DeepMind’s artificial intelligence could experiment and learn from their mistakes. The first setting made will face five blocks of the same size but different weight, where the objective was that the platform would identify what was the heaviest, where he learned that the only way to know was interacting with each and every one of the blocks.
The second experiment consisted of towers of up to five blocks, where the platform had to calculate how many blocks there were in each one, where it again learned that the key was to interact, in this case separating each of the blocks and counting them, in order to determine the answer Correct. In both cases, offering rewards in case of negative feedback for successes and failures, so that if DeepMind knew was right or had to try again.
Such experiments allow artificial intelligence to discover new ways to act based on the ingenuity with which will be able to find solutions when there is no clear instructions, or simply do not exist, and where they do not need to know the physical properties of the object or The physical laws, which will undoubtedly represent an important advance in the development of artificial intelligence.