Making a clever intelligent agent: The theory behind the implementation

Abstract

The study of how humans establish mutual understanding is intertwined with the design of artificial conversation systems [1,2,3,4,5]. The focus of this paper is perspectivetaking in and artificial imitation of communication. Regardless of whether an engineer takes psychological theory into consideration when building an agent, an underlying philosophy of perspective-taking is evident when observing the agent's performance. Furthermore, theories of perspectivetaking offer designers an advantage in two ways: 1) These agents could better imitate human behavior. 2) These agents could use common tendencies in human behavior as an advantage in communicating with humans.

Synthesis and Analysis in Artificial Intelligence: The Role of Theory in Agent Implementation

“…As our first author has mentioned previously [17], we suggest that a successful communicative agent needs to be more than intelligent. It must also be clever enough to use what we know about how humans communicate to its advantage.…”

“…Chatbots that are designed exactly the same but set in different environments can have very different effects on their human interlocutors [17]. People tend to assume that they are understood by their listeners and that they are on the same page or following along with their speakers, they are fooled easily by chatbots that are set in constrained environments.…”

A truly human interface: interacting face-to-face with someone whose words are determined by a computer program

“…Chat bots demonstrate a poor capacity to reason about conversation, cannot consistently identify and repair misunderstandings, and generally talk at an entirely superficial level ( Perlis et al, 1998 ; Shahri and Perlis, 2008 ). According to Raine (2009) , many chat bots work “based on an assumption that the basic components of a communication are on a phrase-by-phrase basis and that the most immediate input will be the most relevant stimulus for the upcoming output” (p. 399), an operative model that can lead conversation to irreparably fall apart when the perspectives of parties to a conversation diverge in terms of the meaning or intention each party assigns to an utterance. Human communication is fundamentally temporal and sequential, with many past and possible future utterances feeding into the meaning of a given utterance ( Linell, 2009 ).…”