好的，以下是内容的中文翻译： **绝对没问题！以下是对视频文字稿的总结，重点关注马克和格雷格讨论的关键点：** **总结：万事达卡的AI之旅以及金融服务的未来** 在这一集“走进金库”节目中，马克采访了万事达卡首席AI和数据官格雷格·奥尔里希，讨论万事达卡在人工智能（AI），尤其是生成式AI（GenAI）方面的策略，以及它对金融服务生态系统的影响。 格雷格分享了他的背景，从专注于衡量干预措施影响的非营利工作开始。 然后他转到应用预测技术公司（APT），后来被万事达卡收购，之后担任战略职务，最后领导AI和数据计划。 他强调，万事达卡已经使用AI几十年了，尤其是在欺诈检测、个性化和预测方面。 随着GenAI的兴起，新的机遇已经出现。 然而，技术的选择取决于用例。 传统的AI和机器学习仍然适用于结构化数据和预测模型，而GenAI则适用于知识管理、内容创作和非结构化数据。 格雷格概述了万事达卡的AI战略，该战略围绕四个关键领域展开： 1. **更安全（Safer）**：加强整个生态系统中的欺诈管理和检测。 2. **更智能（Smarter）**：优化交易路由，并为商家和发卡机构提供见解。 3. **更个性化（More Personal）**：通过银行和商家等合作伙伴为消费者提供个性化的优惠。 4. **更强大（Stronger）**：提高万事达卡员工的内部运营和生产力。 他提供了早期GenAI部署的例子： * **决策智能（Decision Intelligence）**：使用GenAI为现有欺诈检测模型添加功能，通过分析商家行为模式来提高准确性。 * **购物新闻（Shopping News）**：一个提供个性化购物推荐的聊天机器人。 * **数字助手（Digital Assistant）**：一个简化万事达卡产品入门流程的工具，使客户更容易集成和使用它们。 格雷格强调了AI应用中数据安全和可信赖性的重要性。 万事达卡优先与那些拥有相同价值观和对数据保护的承诺的公司建立合作关系。 对于希望与万事达卡合作的新兴技术公司来说，与万事达卡的战略重点（更安全、更智能、更个性化、更强大）保持一致至关重要，拥有GenAI的可靠用例也同样重要。 格雷格描述了万事达卡用于AI决策的中心辐射（hub-and-spoke）模型。 中央AI和数据团队与业务部门合作，共享知识、资源和最佳实践。 这种模型平衡了集中的专业知识和分散的创新。 所有新的AI计划都有关键绩效指标（KPI），这些指标都会被持续跟踪。 他还指出，定性数据，例如员工满意度，对于代码助手而言很重要，因为这是一种衡量解决方案工作效果的方式。 格雷格还强调了外部学习的重要性，包括参加行业活动、建立联系以及寻求不同的观点。 虽然人们对AI的潜力感到兴奋，但格雷格也承认存在对准确性、幻觉和功效的担忧，尤其是在面向客户的应用中。 在金融服务等受监管的行业中，采用谨慎的方法，并保持人工参与是很常见的。 展望未来，格雷格对以下几个趋势感到兴奋： * **多模态（Multi-modality）**：结合文本、图像、语音和视频来创建更全面的解决方案。 * **推理模型（Reasoning Models）**：AI能够更好地理解自身的局限性，并知道何时承认自己没有答案。 * **信任与责任（Trust and Responsibility）**：将透明度和道德规范融入AI的开发和部署中。 * **数据作为差异化因素（Data as a Differentiator）**：利用独特的数据资产来创建卓越的解决方案和见解。

Absolutely! Here's a summarization of the video transcript, focusing on the key points discussed between Mark and Greg: **Summary: MasterCard's AI Journey and the Future of Financial Services** In this episode of "In the Vault," Mark interviews Greg Olrich, Chief AI and Data Officer at MasterCard, to discuss MasterCard's approach to artificial intelligence (AI), particularly generative AI (GenAI), and its impact on the financial services ecosystem. Greg shares his background, starting from nonprofit work focused on measuring the impact of interventions. He then transitioned to Applied Predictive Technologies (APT), which MasterCard acquired, and later took on strategy roles before leading AI and data initiatives. He highlights that MasterCard has been using AI for decades, particularly in fraud detection, personalization, and forecasting. With the rise of GenAI, new opportunities have emerged. However, the choice of technology depends on the use case. Traditional AI and machine learning remain effective for structured data and forecasting models, while GenAI is suitable for knowledge management, content creation, and unstructured data. Greg outlines MasterCard's AI strategy, which is framed around four key areas: 1. **Safer**: Enhancing fraud management and detection across the ecosystem. 2. **Smarter**: Optimizing transaction routing and providing insights to merchants and issuers. 3. **More Personal**: Enabling personalized offers for consumers through partners like banks and merchants. 4. **Stronger**: Improving internal operations and productivity for MasterCard employees. He provides examples of early GenAI deployments: * **Decision Intelligence**: Using GenAI to add features to existing fraud detection models, improving accuracy by analyzing merchant behavior patterns. * **Shopping News**: A chatbot that provides personalized shopping recommendations. * **Digital Assistant**: A tool that streamlines the onboarding process for MasterCard products, making it easier for customers to integrate and consume them. Greg emphasizes the importance of data security and trustworthiness in AI applications. MasterCard prioritizes partnerships with companies that share its values and commitment to data protection. For emerging technology companies looking to partner with MasterCard, alignment with its strategic priorities (safer, smarter, more personal, stronger) is essential, as is having a solid use case for GenAI. Greg describes MasterCard's hub-and-spoke model for AI decision-making. The central AI and data team collaborates with business units, sharing knowledge, resources, and best practices. This model balances centralized expertise with decentralized innovation. All new AI initiatives have key performance indicators (KPIs) that are consistently tracked. He also notes that qualitative data, such as employee satisfaction, is important for coding assistants, because it is a way to measure how the solution is working. Greg also highlights the importance of external learning, including attending industry events, networking, and seeking diverse perspectives. While there's excitement about AI's potential, Greg acknowledges concerns around accuracy, hallucinations, and efficacy, especially for customer-facing applications. A cautious approach, with humans in the loop, is common in regulated industries like financial services. Looking ahead, Greg is excited about several trends: * **Multi-modality**: Combining text, images, voice, and video to create more comprehensive solutions. * **Reasoning Models**: AI that better understands its limitations and knows when to admit it doesn't have the answer. * **Trust and Responsibility**: Incorporating transparency and ethics into AI development and deployment. * **Data as a Differentiator**: Leveraging unique data assets to create superior solutions and insights.