We have added a new physical layer to the core - the task layer. It allows you to implement a universal algorithmic system that controls the knowledge of artificial intelligence. The basis of the task layer includes a modified Petri net and the local decision space associated with it, which is a subset of abstract and factographic layers’ elements.
The local decision space serves as a repository of the problem’s context. We can work with an arbitrary number of tasks and execute the decision algorithms for solving each of them at the same time. At any time, it is possible to switch between tasks, pause a decision making, cancel / delete a task, force completion. The task layer is closely related to the previous physical layers (abstract, factographic, logical) of the hierarchical multi-agent system and uses similar knowledge structures. At the moment, the functions of the task layer can be demonstrated with a dialogue between a user and the application service provider (ASP) of IT services.
Here is an example of a dialog:
(1) The task’s activation, the AI starts to think about its solution.
(2) A short answer to the question, the AI understands that in this context, to abroad and temporary SIM-card to far abroad are the exact same thing.
(3) The answer to the AI question and a new task’ s activation, the AI starts to solve the task received from the user and postpones the previous task till the completion of the new one. The task immediately shows info about the program’ s installation folder and its name, the AI automatically uploads the necessary data into the memory.
4) Date Processing.
(5) There are two ways to change the parameters of the task before it is confirmed:
(6) After all parameters are confirmed and the application is closed, the application is sent to the rightservice.
(7), (8) After the completion of one application, the AI proceeds to the previously created and open task.
(9) Processing the response to AI questions and confirming application parameters.
(10) Confirmation of the application, the application is submitted to the right service.
This example shows that the technology allows you to implement such dialogs in all areas: restaurant business, retail, banking, healthcare, tourism, insurance, etc. Dialogues can be very flexible because there is no rigid script to its structure. The solution of the problem can be implemented in an arbitrary order ( that does not violate the rules of the transition between subtasks), which gives variability and uniqueness to the dialogue. We will continue the development of the task layer by adding the ability to implement fuzzy transitions, as well as visualization of the active tasks and the process of solving them. In addition, we plan to use this technology in some later products.