The ability to change behavior depending situation of a dynamic environment is an essential ability of animals for survival. However, conventional behavior learning algorithms can learn just in a static environment. So, in this paper, we propose a cognitive architecture that switches its conducting task in real time depending on its environmental situation. The architecture is realized by the combination of a value- based task selection and a parallel environmental situation prediction with an associative memory.
While various models and algorithms for artificial general intelligence have been pro- posed, their theoretical properties are not clear. In this paper, we consider the desirable properties of AGI and describe what kind of models and algorithms can be guaranteed to have such properties from a theoretical point of view.
In order for the artificial brain mounted on" a robot that can control mechanical operation" to act on its own intentions, the artificial brain moves the acting part toward the line of sight (gaze guidance), and acts on the object. 1 read the field of view information (real virtual field of view space) of the area to be acted on, 2 generate "execution story and point of view based on thought/ situation judgment" in the pure virtual field of view space, and overwrites them into a real virtual field of view space. 3 moves the part acting on the point of view of the real virtual field of view space and acts on the object. 4 confirms the result in real-time in the real virtual field of view space. 5 repeats #1 to #4 and realizes a series of actions.