Abstract as per original application
(English/Chinese): |
While five finalists in the 2024 Pulitzer Prize for Journalism acknowledged using artificial intelligence (AI) in their work and Bloomberg’s Cyborg system generates thousands of articles within milliseconds, a paradox emerges from this “AI-turn” in journalism: do users perceive AI-synthesized news as more or less biased than human journalism? Competing arguments emerge: users may view AI as objective due to its rule-based operation and freedom from human prejudices, or conversely, as suspect due to algorithmic opacity and its inability to exercise contextual judgment. In politically polarized environments, even balanced reporting can be perceived as biased—a phenomenon known as hostile media perception (HMP)—which leads to news disengagement, selective avoidance, and diminished trust in the broader media environment.
This project examines public perceptions of and responses to AI-synthesized news—AI-powered news curation applications—from socio-psychological and journalism studies perspectives. We examine how source-level (authorship: human vs. AI vs. human-AI hybrid), message-level (editorial transparency disclosure: evidence-based sourcing citations vs. process-based fairness claims), and system-level (platform reach: established commercial application vs. student coursework prototype) factors shape bias perception as stipulated by HMP (Objective 1). Informed by human-computer interaction and self-categorization theory, we explore how this news–perception pathway is moderated by three techno-political dispositions—machine heuristics (mental shortcuts about algorithmic objectivity versus inflexibility), affective polarization (in-group solidarity and out-group animosity), and technological nationalism (belief in one’s nation’s technological superiority)—that condition audience responses to AI-authored content (Objective 2). We also examine how such hostile perception predicts several attitudinal and behavioral outcomes, including news engagement, platform trust, and policy attitudes (Objective 3).
To ground this inquiry in high-stakes environments, this project implements two population-based online experiments in the United States and Hong Kong, using U.S. federal AI regulation (Study 1, U.S. sample) and U.S. semiconductor export controls on (Study 2, U.S. and Hong Kong samples) as test cases. Both topics provide ecological validity through their sustained media coverage, concrete economic impacts, and geopolitical importance.
The findings aim to advance theories in media effects, political psychology, and human-AI interaction (Objective 4), and offer evidence-based insights for media organizations and policymakers (Objective 5). For practitioners, we identify transparency and platform positioning strategies that build credibility. For policymakers, our findings foster democratic media innovation, aligning with the UN Sustainable Development Goals to promote peace, justice, strong institutions, and responsible innovation.
儘管2024 年普立茲新聞獎（Pulitzer Prize）有五名入圍者明確承認在其工作中使用人工智能（AI）、且美國彭博社（Bloomberg）的 Cyborg系統能在短時間內生成數千篇報導，新聞業的「人工智能轉型」引申出一項悖論：受眾究竟會認為 AI 合成新聞比人類採寫的新聞更具偏見，還是更為客觀？學術界對此存在兩種截然不同的觀點：一方面，用戶可能因 AI 依循規則運作且無人類偏見而視其為客觀；另一方面，用戶亦可能因演算法的不透明性及缺乏語境判斷能力而對其產生懷疑。在政治高度極化的環境下，即便是平衡客觀的報導亦可能被視為帶有偏見——傳播學稱其為「敵意媒體感知」（Hostile Media Perception, HMP）現象，進而引發受眾的新聞疏離、選擇性迴避，並削弱對整體媒體環境的信任。
本研究項目從社會心理學與新聞學研究視角出發，深入探討公眾對 AI 合成新聞的感知與反應。本項目的具體研究目標如下：
目標一：基於敵意媒體感知理論，檢視來源層面（作者身份：人類 vs. 人工智能 vs. 人機協同）、訊息層面（編輯透明度披露：基於證據的出處引用 vs. 基於流程的公正聲明）及系統層面（平台觸達：成熟商業應用程式 vs. 學生課程原型）之因素如何塑造受眾的偏見感知。
目標二：結合人機互動（HCI）與自我歸類理論，探索三種技術政治傾向如何調節上述新聞感知路徑，進而影響受眾對 AI 署名內容的反應。
目標三：考察此類敵意媒體感知如何預測受眾的態度與行為結果，包括新聞參與度、平台信任度及政策態度。
為在高風險、高爭議的議題情境中落實本研究，本項目將在美國與香港兩地開展兩項基於人口代表性樣本的線上實驗，分別以「聯邦人工智能監管」（研究一，美國樣本）與「技術出口管制」（研究二，美國與香港樣本）作為測試案例。這兩個議題具備持續的媒體報導、具體的經濟影響及高度的地緣政治重要性，能為本研究提供顯著的生態效度（Ecological Validity）。
本項目的研究發現旨在推進媒體效果、政治心理學及人機互動領域的理論發展（目標四），並為媒體機構與政策制定者提供基於實證的深刻見解（目標五）。對新聞從業者而言，本研究將梳理出能建立公信力的透明度策略與平台定位策略；對政策制定者而言，研究成果將有助於促進民主化的媒體創新，切合聯合國可持續發展目標（SDGs）中關於促進和平、正義與強大機構以及負責任創新的願景。
|
