Introduction

The natural conversation between the artificial agents can be viewed as a cognitive process, demands a consistent information processing infrastructure similar to the human mind. Normally after realising a new stream of textual signal, the information processing engine of a human brain incorporates with the following characteristics:

• Explorative
• Adaptive
• Associative
• Incremental
• Temporal

In this context in a naive approach, we propose that the complete information of individual agent may distinguished as understanding about its own world and the understanding about the world of its conversing partner. Altogether we termed the extracted cells and their associated connections as the MindMap. Following picture depicts the situation during a textual conversation between Alice and Bob.



Architecture

The goal of this ongoing work is to optimize the MindMap and keeps it in a healthy (consistent) state by applicable machine learning processes. For that we propose the following model, where the input signals are processed and structured by three core units.

• Pattern Formation Unit: The input text streams are properly tagged by several meaningful cells and their associated connections using a specific Natural Language Processing (NLP) approach.
• Short Term Memory (STM) Unit: Structured cells and their connections are temporarily stored and filtered.
• Long Term Memory (LTM) Unit: Highly activated cells are identified and extracted from the STM. This unit works as the final permanent storage of the processed information.

Application

Computational Trust: The MindMap contains the relevant and meaningful extracted input signals for each agents. Therefore in our understanding the measures of common information between two artificial agents exactly depicts the amount of trustworthiness between them. Finally, we propose different trust computational scenarios as per the status of the MindMap.

"MindMap" is mentioned on: Jayanta Poray