Google succeeded to produce an artificial intelligence (AI) that can generate its own artificial intelligence. When you think about the technology companies designing artificial intelligence, it can be said that Google is successful. The famous DeepMind of Google is the best example of this. Artificial intelligence does not just keep up with complicated tasks; it is much more successful than any other person living in the complex Go game.
CEO of Google, Sundar Pichai has made some interesting explanations about AutoML, another neural network process which writes layered complex code to “learn” about its surroundings. Each of these layers, which can be seen as parts of the whole AI, normally has to be written by people, which is a time consuming process. Google has been working on the idea that existing AI would write its own code layers. It seems that this process is going much faster and more efficiently than any human technician can. In other words, Google’s artificial intelligence is becoming its own creator.
According to researchers working on this project, a control neural network can suggests child level architecture and this new model can then be trained for a specific task. This network is constantly monitored by the controlling AI whatever the task, and the feedback is used by AI to develop this “child”. This process is repeated thousands of times, creating new architects, testing and checking AI from feedback.
The AutoML procedure is currently only applied to image detection and language modeling. AI-generated programs created by humans and AI-generated programs created by itself show some coding similarities. However, they also have important differences. The code written by AI contains small sub-programs which seem to work for human code writers.
As a result, with an artificial intelligence that can write codes, almost everyone can develop AI and convert it to the type that can develop itself by just using a few simple commands.