All AI’s chains of thought and lines reasoning can be tracked, documented and explained
The structure of knowledge of AI allows to transfer it from one core to another without any leakages of data. Moreover, it is possible to input knowledge in the system as a plain text.
The system can be trained with any amount of raw data. All training requires one iteration. Also there are methods to control and edit new knowledge.
Knowledge is arranged in a multi-level hierarchical structure. Learning mechanisms allow to check the consistency of new and previously obtained knowledge.
AI’s knowledge can be extended without losing the previously obtained knowledge. In addition, there are mechanisms of forgetting which do not lead to the loss of knowledge integrity.
Knowledge can be gradually accumulated in the system. It is possible to store contradictions, vague knowledge, etc.
The core processes and stores information in a form independent of the natural language.
It is possible to assemble a solution suitable for certain tasks from the components of the core. This solution will use all the power of the core and will not include models that are unnecessary for the task.
The core can operate isolated from the Internet and cloud servers.
The structure of knowledge and work algorithms is developed without any dependencies on the operating system or hardware. All you have to do is just to transfer the core of the AI from one operating system to another.
Also, there are no combinatorially complex calculations in the work of agents, which allows you to run the core on a "weak hardware".
This is a feature of architecture that allows you to store knowledge in various sections. It gives the flexibility to represent any objects and phenomena in the intellectual core.
Each core layer is optimized for certain types of knowledge and stores them in the most compact form. It allows to significantly save resources.
At the moment, the core of AI and all its knowledge take about 7 megabytes.