Life-like Autonomous Service Agent (LLASA)
By
Kenneth Cheng
Email: kcheng@cs.stanford.edu
phone: (415) 723-8226
1. Motivation
Almost all businesses need service agents for various reasons. Some companies hire technical support and customer service representatives to serve their customers in person and through the telephone network. Managers and company executives employ secretaries to make appointment with their clients and answer questions. Tour companies employ tour guides to give tour to their customers. For whatever reason these service agents are hired, the employers expect them to be conversant in the subjects they are dealing with, well-mannered to the customers without exception, and most of all, always available when they are needed. However, meeting these requirements is still a persisting problem to be resolved. The difficulties stem from the following constraints:
1. Lack of time in educating service agents on their service domains
2. Misunderstanding between the agents and the customers in their interaction
3. Lack of fund to hire enough service agents to meet customer’s need
4. Human inclination to act based on their own mood and emotional response to the environment. Thus if a customer appear to be rude, the service agent may react the same why.
The aim of the Life-life Autonomous Service Agent(LLASA) project is to create synthetic service agent that meet the ideal requirements and expectations from both the employers and the customers. The autonomous service agent composes of an animated human character, a graphical user interface for customer input, a database to store data of the expert domain, and an autonomous agent to recognize customers’ mood and infer response to customers. The LLASA can be loaded to the web for public access. The animated character and the GUI will be on the client side accessed by a web browser whereas the database and the agent are resided in a host server.
A computer simulated service agent is preferred to a human agent in various ways:
1. A LLASA is a multi-threaded component that can serve unlimited number of customers as long as the hardware is capable of handling the work load. On the other hand, we may need to employ numerous human service agents and increase work space to accomplish the same task.
2. Once knowledge is downloaded to the knowledge base, it is never forgotten unless in case of system failure.
3. The personality and mood of LLASA is set to be friendly, apologetic and patience in all circumstances. Customers can be unreasonably rude and annoying, but the LLASA will only seek ways to satisfy the customer and make apology when necessary. Satisfying the customer is the fundamental goal to be achieved by the agent.
4. Running LLASA is economical. Once the system is implemented, we can install the system on the web server for public access. The only variable costs involved are maintenance and software upgrade costs.
5. LLASA can bridge the gap of miscommunication between the service agents and the customers. Unlike human agent, LLASA can accurately discern customer’s emotion if the customer expresses his/her emotion correctly by selecting proper mood level through sliders and buttons in the GUI. For example, the customer can express his impatience by moving the slider up. In that case, LLASA will apologize to the customer for his slow service and try to speed up and be more polite. Because customers do not deal directly with an actual agent, they may feel less intimidated in expressing their emotion towards the service.
2. Research Objective
My research objective is to create an autonomous agent that is able to render a specific type of customer service as provided by live service agents. To enhance "life-likeness" of the agent, I will incorporate my research with the principles of "computer politeness" introduced by Nass in his book "Media Equation". He suggested four important elements that contribute to computer politeness:
Quantity: It deals with whether information provided by the agent is too much or insufficient to satisfy the customers
Quality: To ensure information given is correct and complete
Relevancy: To provide enough options in the gui so that customer can comfortably express his requests.
Clarity: It is how information is presented
The elements mentioned above do not fully contribute to agent’s life-likeness. To meet my research objective, I recommend two additional elements:
Sensitivity: This is how and where the autonomous agent shines. In an implicit way, the agent perceives customer’s mood by analyzing options and dialogues that the customer selects. The other way is for the user to adjust his mood by using sliders provided in the gui. The agent will respond and act according to customer’s mood. For example, the agent will greet and chat with the user. If the response from the customer is positive, the agent may attempt to talk a little more before it answers customer’s request. If the user says "I am in a hurry" instead of "I am fine. Thank you", the agent will respond to customer’s requests in a more direct way.
Believability: In fact, agent’s sensitivity is part of believability. However, my emphasis to this element is on the user interface design and agent’s dialogues. I would like to make the interface more visually believable by creating an animated service agent in his/her 3-D office. The customer should feel like he is talking with a live-agent. In addition, the agent will talk and render data in an adaptive manner relative to the customer’s moods. I will investigate on the feasibility to record and deliver the agent’s dialogues over the web so that the agent can actually talk to the customer. More details will be given in the design section.
I will use Winnie as a prototype in this research project. Winnie is a virtual receptionist. She is able to answer questions from customers regarding research projects as well as employee information in the Virtual Theater team. Winnie is well-behaved, patient, polite, and helpful. The implementation of Winnie will be given in the design section of this proposal.