Life-like Agents for the Internet: Present and Future
Vienna Workshop "Agent Culture", 24-25 August 2001
Brigitte Krenn (ÖFAI, Vienna)
Erich Gstrein (sysis interactive simulations ag, Vienna)
Martine Grice (Saarland University, Saarbrücken)
There are already a wide range of roles which can be filled byembodied conversational agents (ECAs) in internet applications. Theycan support customers in e-warehouses and e-showrooms; they can offerguidance in decision making processes, e.g. acting as an interface tojob or partner-matching algorithms; they can simply presentinformation, as do virtual newscasters; or they can enact the role ofa character in a game as do avatars in computer games. Although userscan be supported and informed without the use of embodied agents,such agents introduce a social dimension to the applications.
The main reason for bringing life-like characters onto theinternet is of course an economic one: Users tend to buildrelationships with such characters and can therefore be more easilypersuaded to buy specific goods and services or to commit themselvesto particular sites on which the provider sells advertising. However,there are some distinct advantages for the user: the use of ECAs canoften make information gathering and decision making both simpler andmore fun. In the context of the internet, ECAs can be used to supportintercultural learning and can be used as a testbed for socialinteraction.
We believe that, in applications where the social dimension iscrucial, agents must be identifiable as belonging to a particularsocial and cultural group. This is particularly important as theiranimation becomes more sophisticated, their texts are generatedautomatically, and as speech synthesis used for the voice improves inquality, so that personality traits and emotional states are moreeasily recognised. Thus for the future generation of internetcharacters,socio-cultural modeling of (verbal and nonverbal)behaviour will become increasingly important.
Even though there is a vast literature on socio-cultural aspectsin psychology, sociology, or foreign language learning, much work hasyet to be done to apply these findings in the design of ECAs. We willdiscuss another way of addressing the issue of socio-culturalmodelling: An effective way of accessing information as to what userswithin a given culture consider to be acceptable behaviour is throughthe use of online questionnaires that are integrated into anapplication. The availability of embodied agents also enables us topresent the user with the questions in a more appealing way. Weexpect that this will positively influence the willingness of theuser to answer these questions. Moreover, combining questions to theuser with examples of particular behavioural patterns displayed by anagent enables us to more closely tie stimulus and response, whichshould improve the quality and reliability of the user responses.
In our contribution, we refer to Flirtboat, a multi-user netapplication developed by _sysis interactive simulations_. In thisapplication agent representatives of the users interact to find apartner by arranging dates. We give a description of the currentFlirtboat, which is a precursor to one of the net environmentspopulated with agents interacting in a social setting which will bedeveloped in the NECA (Net Environment for Embodied EmotionalConversational Agents) project. In the current implementation, theanimation of the agents is rudimentary and no natural language orspeech processing is available. Instead, text is selected from adatabase of fixed chunks and presented in written form. Theserestrictions were necessary to enable _sysis_ to make the applicationwidely available on the internet, given the state of the art ininternet technology and given bandwidth considerations. Nontheless,the agents are autonomous characters with personality and emotions.They simulate physiological needs such as hunger, and respond tosituational requirements within the application such as the number ofdates made. The user can only influence the agents by giving advice,in particular by answering questions generated by the system.
We then discuss what we have learned from launches of Flirtboat ontwo Austrian internet sites and another launch of an English versionof the game on a British site. We also discuss how the concept ofECAs can be sucessfully transferred to net environments likeFlirtboat, and what is still required from future research in orderto develop internet applications with conversational agentsinteracting in a socially and culturally appropriate way.