Current processor is designed for processing structured data using pre-defined paradigm, i.e. the input, output, and storage are accurate and the processing rules are determined. However, most of natural data are unstructured, such as video, image, and voice. Although many efforts have taken to process unstructured data using predefined algorithm or a neural network, they still have weak expression ability for complicated reasoning tasks which may have many steps and complicated logical relationships. In such cases, the rules and experiences are applied as a ‘program’ to process the data. Differs from the current processor’s program, these programs called ‘experience rules’ are inaccurate and are obtained from unstructured data previously, or are taught following a teaching mechanism. In this work, a newly established processor architecture is introduced using the experience rules as the program. The processor is designed for obtaining program from unstructured data following basic learning rules. Then the experience rules are applied for processing new unstructured data. The processor is designed for animating the thinking procedure of human beings, which paves the way to the basic principles of the artificial general intelligence (AGI) processing machine.
Fig.1. Comparison between proposed intelligence processor and the traditional processor.