Google scientists have developed a new artificial intelligence (AI) system that uses basic reasoning to execute complicated tasks such as navigating the London Underground system all by itself. Not just that, the new AI is also capable of answering questions about the content of snippets of a narration, or even analyze a family tree to work out different family relationships.
The next phase in the program, the scientists predict, could enable virtual assistants to instantaneously dive into the right corners of the internet to find answers to important questions and follow instruction with unprecedented precision.
The new AI system makes use of a “differential neural computer”, or DNC, that can carry out real-time analysis of the memories it had gained in the past to produce answers to complicated questions. The answers are then matched with other alternatives which include the correct response as well.
In the London Underground environment, the AI successfully answered all the questions about the best routes between two stations, as well as which station you would end up at if you took a specific route.
Even though the tasks given to the system seem fairly easy, it happens to be one of the first programs to combine memory with the deep learning approach. This ensures that the program can deliver an output independently like a sentient being rather than like a computer program fed with a set of instructions.
Acknowledging the complexities in Google’s new AI, Herbert Jaegar, a computer scientist at the University of Bremen“I think this can be described as rational reasoning. They [the tasks] involve planning and structuring information into chunks and re-combining them,” the Guardian reports.