자료유형  

 학위논문

Control Number  

0017163753

International Standard Book Number  

9798342113182

Dewey Decimal Classification Number  

155.2

Main Entry-Personal Name  

Cao, Xubo.

Publication, Distribution, etc. (Imprint  

[S.l.] : Stanford University., 2024

Publication, Distribution, etc. (Imprint  

Ann Arbor : ProQuest Dissertations & Theses, 2024

Physical Description  

87 p.

General Note  

Source: Dissertations Abstracts International, Volume: 86-04, Section: B.

General Note  

Advisor: Kosinski, Michal.

Dissertation Note  

Thesis (Ph.D.)--Stanford University, 2024.

Summary, Etc.  

요약As artificial intelligence (AI) systems, particularly large language models (LLMs), increasingly integrate into diverse facets of our lives, their interplay with human psychology and behavior becomes an unavoidable area of exploration. At the heart of this lies a dual-edged inquiry: firstly, the potential of LLMs to recognize and decipher human personality traits based on linguistic cues, and secondly, the manifestation, intentional or inadvertent, of 'personality' within the AI systems themselves.This type of research holds significant practical implications for the development of LLMs and for enhancing human-computer interactions. LLMs that can recognize, emulate, and manifest different personality types could significantly improve their user experience. For instance, adjusting chatbots' responses to the personality traits or preferences of their users could lead to a more personalized, engaging, and human-like experience (Qian et al., 2018). LLMs' output, such as stories, would likely be more attractive if they portrayed characters with distinct, consistent, and believable personalities.Moreover, the analysis of large language datasets via machine learning and AI algorithms has become crucial in social science research. As LLMs evolve and refine their capabilities, their impact extends beyond just direct human-AI interactions, reshaping the research methodologies within the social sciences. Particularly prominent is their use to assess and predict psychological constructs, such as personality (Kosinski et al., 2013; Youyou et al., 2015). While personality psychology traditionally relies on self-reports or observer ratings to discern personality dimensions, the development of natural language processing techniques has brought several paradigms shifts in automatic personality assessment (e.g., Farnadi et al., 2016; Park et al., 2015; Pennebaker & King, 1999).Additionally, social science researchers have started to explore how LLMs can be seen as effective proxies for specific human sub-populations (Argyle et al., 2023; Dillion et al., 2023; Horton, 2023). If LLMs can effectively approximate the behavior of individuals with certain combinations of traits, they can offer researchers a unique tool to simulate human responses without actual human involvement. For experimental designs where manipulating variables could be ethically challenging or logistically cumbersome with real participants, LLMs can provide a controlled environment to test hypotheses. While these simulations are far from replacing the depth and unpredictability of genuine human behavior, they may offer an invaluable sandbox for preliminary experiments and proof-of-concept studies.While this topic holds immense interest and potential, there remains a paucity of comprehensive research. In this dissertation, I probe the capabilities of contemporary LLMs, with a focus on the GPT series, as they engage in diverse personality-centric tasks such as personality assessment and manifestation. Further, I delve into the broader implications of these interactions.In the first chapter, I delve into the capacity of LLMs to predict public perceptions of the personalities of public figures (adapted from Cao & Kosinski, 2024). The chapter has two purposes: (1) Predicting and assessing the public's perception of prominent figures carries inherent value. Such assessments find their applications across various disciplines, including Political Science, Management, and Marketing. (2) I investigate whether LLMs retain personality-related information within their vast knowledge base. Essentially, does the LLM "remember" personality traits associated with various figures based on its training data? Can the LLM correctly retrieve the relevant information without further training?In the second chapter, I shift the lens toward the AI's ability to discern personality traits. While assessing a public figure's perceived personality is indeed valuable, its utility remains limited to specific contexts. The broader and arguably more profound application of LLMs lies in their interactions with everyday users. Can LLMs, during a brief interaction, effectively gauge the personality traits of their interlocutor?In the third chapter, I delve into the burgeoning trend of developing AI agents that exhibit distinct personality profiles. This exploration centers on the utilization of prompting techniques to assign specific personality traits to LLMs, and the potential unintended side effects that may arise from these assignments. Specifically, I examine scenarios where manipulating one trait might influence other associated traits-a phenomenon emerging due to the interconnected nature of personality dimensions. Furthermore, I investigate whether such side effects are predicted by existing psychological theories. The discussion extends to the broader implications of these side effects, considering both the technological and ethical dimensions of creating AI personalities.

Subject Added Entry-Topical Term  

Personality traits.

Subject Added Entry-Topical Term  

Large language models.

Subject Added Entry-Topical Term  

Personality psychology.

Added Entry-Corporate Name  

Stanford University.

Host Item Entry  

Dissertations Abstracts International. 86-04B.

Electronic Location and Access  

로그인을 한후 보실 수 있는 자료입니다.

Control Number  

joongbu:655058