Topic Title: TmallGenie with Curiosity

 

 

Technical Area:

human-computer interaction

natural language processing

 

Background

Intelligent companions powered by artificial intelligence (AI companions for short) provide a new promising way for everyday human-computer interaction. By referring to an AI as a companion, we emphasize on a closer, long-term relationship between the user and the AI companion. An AI companion needs to be socially intelligent. It is expected to play an equal role in human-computer interaction, acting as if another individual who is equipped with social skills to conduct real life communication. However, most existing scenarios are activated passively, when the user delivers specific instructions (e.g., tell me the weather and show me the next appointment).  Moreover, in most scenarios there are only limited conversational interactions, such as short suggestions and recommendations, which make the AI companions not engaging and trustworthy enough to elicit emotional response from the user and hinder them from building close long-term relationships. In recent years, curious AI has emerged as a powerful learning paradigm to deal with situations when rewards are not immediately available. As a human nature, curiosity may also be vital for building truly smart AI. When equipped with artificial curiosity, machines can develop spontaneous and efficient learning strategies to achieve new capabilities. Curious machines can trigger more meaningful and engaging human-computer interactions to benefit users of all ages while making TmallGenie a perfect AI companion.

 

Target

  1. We are going to examine the current functions of TmallGenie and identify those functions that can be largely improved with curiosity, such as recommendation.
  2. We are going to explore curious AI techniques to equip TmallGenie with human-like curiosity, such as novelty-seeking and proactive question-asking
  3. TmallGenie with human-like curiosity will be evaluated in terms of task performances as well as user satisfactions.

 

Related Research Topics

 

Personal behavior modeling. We aim to conduct research on the proactiveness of AI companions by harnessing curious AI. Through the new theories and design in curious AI techniques and the seamless integration with social and emotional theories, the curious AI companion will be able to build a personal behavior model of the user, discover abnormal and suspicious situations, and generate personalized recommendations.