Matching artificial agents and users personalities designing agents

Matching artificial agents’ and users’ personalities: designing agents with regulatory-focus and testing the regulatory fit effect Caroline Faur (faur@limsi. fr) Jean-Claude Martin (martin@limsi. fr) Celine Clavel (clavel@limsi. fr) LIMSI-CNRS, rue John Von Neuman, bt 508 91403 Orsay Cedex, France AELEE KIM Cognitive Science, Ph. D. Candidate Methodology in Cognitive Science Professor Byoung-Tak Zhang Seoul National University Fall 2015

Keywords Artificial agents Artificial Companion Personality Social Cognition Regulatory Focus Regulatory Fit

PURPOSE / Challenge Designing agents with personalities to the benefits of users.

INTRODUCTION Human-Computer Interface Artificial Agent Artificial Companion User Personality Social Cognition Regulatory Focus Regulatory Fit

INTRODUCTION Human-Computer Interface Artificial Agent Artificial Companion User Personality Believability Human Social Cognition Theory Regulatory Focus Regulatory Fit

INTRODUCTION Human-Computer Interface Artificial Agent Artificial Companion ”a personalised, multi-modal, helpful, collaborative, conversational, learning, social, emotional, cognitive and persistent computer agent that User knows its owner, interacts with the user over a long period of time and builds a (long-term) relationship to the user” Personality (Sviatlana, Busemann, & Schommer, 2012) Artificial + Companion Social Cognition 인공의, 인위적인, 인조의 + 친구, 동반자, 동료, 반려, 벗 Theory Regulatory Focus Regulatory Fit

INTRODUCTION Human-Computer Interface Artificial Agent Artificial Companion User Personality A coherent patterning of affect, behavior, cognition, and desires (goals) over time and. Cognition space (Revelle & Scherer, 2009). Social Personality : 성격, 사람, 인격, 개성 Theory Help to increase the companion’s Regulatorybelievability Focus Regulatory Fit

INTRODUCTION Human-Computer Interface Artificial Agent Artificial Companion User Personality Social Cognition Social cognition is a sub-topic of social psychology Theory that focuses. Regulatory on how people process, store, and Focus apply information about other people and social situations. Regulatory Fit It focuses on the role that cognitive processes play in our social interactions. Social cognition is the study of how people process social information, especially its encoding, storage, retrieval, and application to social situations.

INTRODUCTION Human-Computer Interface Artificial Agent Artificial Companion User Personality Human Social Cognition Theory Regulatory Focus Regulatory Fit

INTRODUCTION : Regulatory Focus Theory § Originated by Tory E. Higgins from Columbia University, 1997 http: //www. columbia. edu/cu/psychology/higgins/ § Self-regulation strategies § A fundamental Motivational Theory § Promotion VS Prevention types § People’s tendency toward promotion vs prevention focus when they consider what goals to pursue and how to pursue goals. § Regulatory focus can be situational, induced by the context, but theory states that people have a chronic focus, an “habitual” focus used by default.

INTRODUCTION : Regulatory Focus Theory Regulatory 규제의, 조정력을 가진, 단속의 + Focus 초점, 집중하다, 중심, 주안점 Promotion Focus Prevention Focus Gain vs. Nongain Loss vs. Nonloss Approach strategies Avoidance strategies Errors of omission Errors of commission

INTRODUCTION : Regulatory Focus Theory Regulatory 규제의, 조정력을 가진, 단속의 § § + Focus 초점, 집중하다, 중심, 주안점 Promotion Focus Prevention Focus Gain vs. Nongain Loss vs. Nonloss Approach strategies Avoidance strategies Errors of omission Errors of commission Not performing an act or behavior – just didn’t do it Something left out by accident. Transaction is to be left out to register. Partial entry of one transaction. § § § Performing a different act or behavior – not to norm Something wrong is done RS. 1500 recorded as RS. 5100

INTRODUCTION : Regulatory Fit Regulatory-fit : A feeling of rightness about the pursued goal and increases task engagement (Higgins, 2005) Regulatory 규제의, 조정력을 가진, 단속의 + Fit 꼭 맞는, 어울리는, 건강한, 맟추다, 들어맞다 Non. Gains Non. Losses

INTRODUCTION : Regulatory Focus Theory 적극성 조심성

QUESTIONS 1. How can we implement regulatory focus for artificial agents ? Artificial agents에 어떻게 RF를 적용시킬 수 있을까? 2. Is the intended personality perceived as such ? RF를 적용시켰을때 사용자들이 Artificial agents의 퍼스널리티를 알 수 있을까? 3. Can we reproduce a regulatory fit effect between such an agent and users? 사용자와 Artificial agents 간에 Regulatory fit 효과를 재현해 낼 수 있을까?

BACKGROUND Computers As Social Actors (CASA) paradigm (Nass & Moon, 2000) People tend to adopt social attitudes with machines that can elicit social heuristics. Personality Measurement The Five Factors Model (FFM) (Costa & Mc. Crae, 1992) also known as the Big Five: 1. 2. 3. 4. 5. Openness Experience (개방성) Conscientiousness (성실성) Extraversion (외향성) Agreeableness (친화성) Neuroticism (신경성)

BACKGROUND Social Cognitive Models Traits Theories (성격이론) § Useful for the description of the personality. § The socio-cognitive approach to personality underlines the importance of a situation in § But by looking at the global structure of exhibiting personality behaviors. personality, they hide intra-individual differences. (Bandura, 1999). § This approach attempts to understand cognitive and social processes that lead to personality. § For that purpose, it focuses on the interaction between the person and the social context and highlights the intraindividual differences (Mischel, Shoda, & Smith, 2004).

METHODOLOGY : Convey personality via game strategies Board Game Personality Can’t Stop (Designed by Sid Sackson) Stopping a turn, saving the current gains. But loosing in speed Stop-or-Again Game Rule Playing again, taking the risk of loosing the current gains to win more

METHODOLOGY : Convey personality via game strategies Board Game Personality Can’t Stop (Designed by Sid Sackson) Stopping a turn, saving the current gains. But loosing in speed Stop-or-Again Game Rule Playing again, taking the risk of loosing the current gains to win more

METHODOLOGY : Data-driven implementation Regulatory Focus Questionnaire Proverbs Form (RFQ-PF) -> Measuring the strength of the two self-regulatory strategies Participants : 15 = 13 men + 2 women three models : 1. one for the choice of a move during the game 2. two for the ”stop-or-again” decision 1) With and without taking into account personality scores as a feature; 2) The latter should smooth intra - individual differences to produce a ”depersonalized” strategy.

METHODOLOGY : Experimental Design 2 types of strategies + 4 types of agent Random Strategy AI 1. Rand (Random Agent) 2. Avg ( Average Agent) Which chooses randomly its moves and has a 50% probability to stop its turn which follows the ”depersonalized” strategy Agent 3. RF-Pro (Promotion Agent) 4. RF-Pre (Prevention Agent) Which has a promotion score of 7 and a prevention score of 1 Which has a promotion score of 1 and a prevention score of 7.

USER STUDY Hypothesis H 1 : The differences in agents personalities are perceived by the human player (사용자가 에이전트 퍼스널리티의 차이를 인식한다) H 2 : The credibility of the agent is increased by the presence of personality. The RF-agents are perceived as more likeable and more intelligent than the Rand-agent and the Avg-agent. (퍼스널리티가 있는 경우 에이전트에 신뢰성이 증가된다) (RF-agents는 Rand-agent와 Avg-agent 보다 호감도와 지적인 면이 높게 인식된다) H 3 : According to the regulatory-fit theory, human player oriented as promotion find RF-Pro agent more credible than other agents (respectively for RF-Pre). R-fit 이론에 의하면 promotion 경향의 사용자는 RF-Pro 에이전트에 다른 에이전트 보다 더 신뢰성을 보인다. (prevention 경향 사용자에도 같은 가설 적용)

USER STUDY Participants : 20 = 11 men + 9 women (age M = 30. 6 years, SD = 8. 1) Regulatory Focus Questionnaire Proverbs Form (RFQ-PF) -> Measuring the strength of the two self-regulatory strategies 14 Participants : A chronic promotion focus 6 Participants : A chronic prevention focus Played Can’t Stop Game Regulatory Focus Questionnaire Proverbs Form (RFQ-PF) + The Godspeed Questionnaire (likeability, the perceived intelligence of the agent)

USER STUDY : Result Hypothesis H 1 : The differences in agents personalities are perceived by the human player Result § § (사용자가 에이전트 퍼스널리티의 차이를 인식한다) H 2 : The credibility of the agent is increased by the presence of personality. The RF-agents are perceived as more likeable and more intelligent than the Rand-agent and the Avg-agent. § Almost Validated RF-Pro and RF-Pre agents has been respectively perceived as promotion-oriented and preventionoriented Partially Validated § Found a difference in favor of the RF-Pre agent regarding the perceived intelligence. The RF-Pro agent was rated as more intelligent than the Rand Avg agents but the difference was not significant. § Partially Validated § (퍼스널리티가 있는 경우 에이전트에 신뢰성이 증가된다) (RF-agents는 Rand-agent와 Avg-agent 보다 호감도와 지적인 면이 높게 인식된다) H 3 : According to the regulatory-fit theory, human player oriented as promotion find RF-Pro agent more credible than other agents (respectively for RF-Pre). R-fit 이론에 의하면 promotion 경향의 사용자는 RF-Pro 에이전트에 다른 에이전트 보다 더 신뢰성을 보인다. (prevention 경향 사용자에도 같은 가설 적용) Found an interaction between the user’s focus and the type of agent regarding the likeability score : prevention-oriented users found the RF- Pre agent and the Rand agent more likeable than the RF-Pro agent and the Avg agent. § Because RF-Pre and Rand agents were both perceived as prevention-oriented, we could say that regulatory fit happened for prevention- focus users. §

USER STUDY : Result Personality Scores Credibility Scores H 1 H 2

USER STUDY : Result H 3 Promotion Focus Prevention Focus

Key Concept Review Human-Computer Interface Artificial Agent Artificial Companion User Personality Believability Human Social Cognition Theory Regulatory Focus Regulatory Fit

CONCLUSION 1. It is possible to successfully endow artificial agents with regulatory-focus and that this regulatory-focus can be accurately perceived by users. 2. Provided data which point to the possibility of using the concept of regulatory fit with artificial agents.

PERSPECTIVES To better understand the regulatory fit effect with artificial agents : 1. Making more longitudinal studies because only repeated interactions could allow users to form a real model of the agent’s personality 2. Using multi-modality to enhance the interaction, such as verbal and non-verbal behaviors during the game by providing a physical representation of a virtual Agent 3. Complementing self-report measures by users’ behaviors measures, such as engagement for example.

The Linkage between this article and my research Interests

The Linkage between this article and my research Interests Human-Computer Interface Artificial Agent Artificial Companion User Personality Social Cognition Regulatory Focus Regulatory Fit

The Linkage between this article and my research Interests