There are already a lot of developments in the field of artificial intelligence and, so-called, deep machine learning. However, all self-learning neural networks available today have one essential point: they are not completely independent. That is, they need to initially set certain parameters by which they will work. At the same time, not a single electronic brain is capable of learning from scratch, as any person does in his life. And it is such a self-learning AI that experts from MIT and Honda are planning to create.
According to the editorial office of Engadget, the new AI will be called the Curious Minded Machine, and the project should take about three years. The ultimate goal of the developers, according to one of the authors of the project Soshi Iba from the research department of Honda, is to create new types of AI that will be able to learn continuously, regardless of the original data.
As an example, the authors cite the fact that the system can observe the work of people in a certain area. And then, by analyzing this data and studying the area in more detail, AI can suggest better and more efficient ways to accomplish these tasks.
In this project, Honda will act as an investor and project leader, contributing data to its AI expertise. MIT, on the other hand, will create a system responsible for predicting various events and the consequences of their actions, as well as processing information.
In addition to MIT and Honda, experts from the University of Washington and the University of Pennsylvania will also be involved in the project. The former will develop a curiosity algorithm based on data on how children learn about the world. And their colleagues from Pennsylvania will create a "simulated biological interaction" - an analogue of the learning processes that are inherent in humans, adapted to use AI.
Vladimir Kuznetsov
