Abstract

• Paper is talking about creating agents that can sense, perceive, reason about, learn from, and respond to affect, behavior, and cognition of other agents (human or artificial).
• Paper is discussing set of underlying technical challenges and open questions for researchers across computing communities to advance social-AI

Introduction

• Homosapiens evolved with core facts that have been the emergence of language culture and societies, they heavily relied on using social intelligence to be able to have long term relationships
• AI Agents therefore, must have social intelligence competencies in order to function seamlessly alongside humans and other ai agents
• The core technical challenges in trying to understand a conversation between human in day to day life goes by this-
  • Ambiguity exists in social construct definitions and can be interpreted differently by different scientists, which can be both wrong and right with different intentions
  • Nuanced signals in conversations can cause sudden turning point of whole meaning of a context, so it is important to be able to detect them
  • The entire perspective of a conversation can be infinite
  • The individuals learn from different social experiences differently and adapt to social contexts which influence their own goals and identities
• To build ai social agents we have to provide the agents with ability to sense, perceive, reason, learn and respond to affective behavioral and cognitive constructs of other agents.

• A third party observer or an ai agent outside a conversation will interpret things by-
  • Having Vector Dimensions of Interactions
    • Hierarchical social units - Networks, Communities, Situations, Events, Behaviors, Signals
    • Interaction structures - Monad, dyad, multiparty
    • Temporal scale - moment, short term, life long
  • Social Setting
  • Social Attributes - Age, Race, etc.
  • Actors and their roles
• Several studies and experiments showed that Virtual text agents like HAYLIE improved the confidence of people going through difficult times by following different approaches such has insert and replacing general degree of empathic words with better empathetic words of text based context and improving empathy of the response to the user, with data from various affective signals from wearables in a study from MIT SANO supported well being, and so on.

Social Intelligence Concepts

• Paper poses a question, what makes entities social? and follows, a person is a natural kind with physical properties but a friend or stranger are social constructs unlinked to those physical properties
• Language plays a key role in forming social constructs, such as referring a close friend as close friend
• Social constructs are very subjective as it depends from infinite perspectives

Social Intelligence

• It is a context dependent intelligence
• It differs from other intelligence, such as abstract intelligence (idea- focused), mechanical intelligence (object-focused) and social intelligence (people-focused)
• Human perceive the intelligence through verbal and nonverbal sensory channels
• Prevalent model developed with construct validation confirmed that 5 competencies enables humans to navigate social situations:
  • social perception - Reading tension from body language
  • social knowledge - Factual and Norms Knowledge
  • social memory - ability to store and recall social knowledge about the self and others
  • social reasoning - interpret social stimuli and make inferences based on stimuli and commonsense understanding
  • social creativity - theory of mind - ability to counterfactually reason about social situations