Trust in motivated learning agents
Janusz A. Starzyk, James T. Graham, Adrian Horzyk · 2016
This paper generalizes motivated learning to include selected personality traits of autonomous learning agents. Specifically, we show how a motivated learning agent can develop trust and use it in interaction with other agents in a dynamic unknown environment where it operates and learns. Trust is built over time through interactions with other agents, while considering the interactions' effects on the needs of the motivated learning agent. A positive effect increases trust while a negative one reduces it. A neural network model used in the motivated learning scheme is utilized to compute the trust level for each non-agent character. Simulation of the motivated learning agent is performed to demonstrate the benefits that result from using trust in interaction with other agents. We imply that trust may influence other traits of character like shyness, friendliness, bravery, etc.