隨著生成式人工智慧（Generative AI, GenAI）技術的快速發展，新聞媒體的商業模 式與資訊主權正面臨前所未有的挑戰 ， 許多 GenAI 模型仰賴大量線上內容進行訓練，可 能間接削弱付費新聞內容的價值，導致媒體營收受損，進而改變媒體與科技業者間的互 動關係，本研究以具代表性的數位付費牆新聞媒體為研究對象，探討在不同合作與防禦 情境下，其對 GenAI 的策略選擇與利潤表現。 本研究結合霍特林模型（Hotelling Model）、賽局理論與多代理強化學習方法，建立兩 家新聞媒體與 GenAI 業者間的互動經濟模型，並使用 Nash Q-learning 與 Friend-or-Foe Q-learning 演算法，模擬媒體在不同市場參數下的最佳定價策略與阻擋強度。 透過四種合作情境的模擬實驗，研究結果顯示：媒體策略受品牌敏感度、內容品質、 GenAI 生成能力與合作補貼等因素顯著影響；合作媒體可穩定利潤但將喪失內容主導權， 而非合作媒體則需承擔防禦成本與利潤不對稱風險。 本研究為媒體面對 GenAI 帶來的內容再利用與市場再分配挑戰，提供策略模擬與量 化依據，亦驗證 Nash Q-learning 於複雜策略互動場景中的應用潛力，對新聞產業、 AI 平 台發展與數位治理議題具有實務與學術貢獻。;With the rapid advancement of generative artificial intelligence (GenAI), the business models and information sovereignty of news media are facing unprecedented challenges. Many GenAI models rely on vast amounts of online content for training, potentially reducing the value of paid news and threatening media revenues. This evolution reshapes the strategic dynamics between media companies and technology platforms. Focusing on digital paywall-based news outlets, this study explores media strategies and profit outcomes under various cooperation and defense scenarios involving GenAI providers. By integrating the Hotelling model, game theory, and multi-agent reinforcement learning, this research constructs an economic model that simulates interactions between two competing news media firms and a GenAI platform. It applies Nash Q-learning and Friend-or-Foe Q- learning to derive optimal pricing and blocking strategies across different market conditions. Simulations of four cooperation scenarios reveal that key factors—such as brand sensitivity, content quality, GenAI capabilities, and revenue-sharing incentives—significantly influence media decisions. Cooperation can reduce defense costs and stabilize profits, but may weaken content control; non-cooperation requires higher defensive investment and carries asymmetric profit risks. This study offers a quantitative framework for evaluating strategic responses to GenAI- induced content reuse and market redistribution. It also demonstrates the potential of Nash Q- learning in modeling complex strategic interactions, providing theoretical and practical insights for media strategy, AI governance, and digital policy.
