A framework to evaluate and establish trust in generative AI, in learning and development (L&D) applications

今天，大型生成模型可以从用户那里获取简单的指令，并生成代码、文本、图像和视频等人工制品，直到最近，这些人工制品还只能由人类产生。然而，依赖于这些模型的产品的成功需要其用户/利益相关者信任它们。这一点在教育、金融和医疗等高风险领域尤为重要。我们正在打造Elandi，这是一款利用人工智能（特别是生成式人工智能）最新发展的力量，为每个人带来负担得起的个性化L&D的产品：1. Elandi将用户的个人资料/简历与他们的职业目标进行比较（例如，职位描述），并找出差距。使用这个间隙（加上各种用户偏好等），它为生成式AI生成提示。我们目前正在探索的用户体验领域包括课程规划、小型辅导、评估和内容生成。由于我们自2020年以来的成功课程，我们现在拥有一个独特的用户社区（来自100多个国家的1000多名AI人才），可以提供数据和建议。此外，我们还与顶尖的人工智能教授/讲师以及一些世界上最大的人工智能应用公司合作。所有这些都帮助我们在Elandi的产品市场匹配（PMF）方面取得了巨大的进步;从发现到概念设计和ML原型。总体而言，从我们的发现中最重要的教训是，符合目的的内容，改进的用户体验，个性化推荐和专家指导的需求很高。虽然这一进展仍在继续，但我们意识到迫切需要一个全面的人工智能保证框架（例如，审计人工智能，衡量用户对它的信任，量化它的各种风险）。因此，作为我们更广泛抱负的一部分，通过这笔赠款，我们希望启动一个关于这个主题的明确工作流。我们的目标是建立在最新的研究（包括指标和测量）上，提供必要的工具和框架，用于审计L&D领域的AI模型和产品。虽然我们将专注于L&D用例，但我们希望我们的结果能够超越L&D和其他领域。改进的L&D产品将具有深远的社会和商业价值;它将在提高国家和国际生产力方面发挥关键作用。它与政府的战略举措，如“升级”、“终身学习”和“生活技能”相一致。

Today, large generative models can take simple instructions from their users and generate artefacts such as code, text, images and videos that until recently, could only be produced by humans. The success of products that rely on such models, however, requires their users/stakeholders to trust them. This is particularly important in domains such as education, finance, and medicine where the stakes are high.We are building Elandi, a product that uses the power of the latest developments in AI (particularly, generative AI) to bring affordable personalised L&D to everyone:1. Elandi will compare a user's profile/resume with their career goals (e.g., a job description) and identify the gaps.2. Using this gap (plus various user preferences, and more), it generates a prompt for the generative AI.3. The generative AI will then provide the user with the right L&D content.The UX areas we are currently exploring are course planning, bite-sized tutoring, assessment, and content generation.Thanks to our successful courses since 2020, we now have a unique user community (1000s of AI talents from 100+ countries) that can the source of data and advice. Furthermore, we worked with top AI professors/lecturers, and some of the world's largest corporations in real-world AI applications. All these helped us make tremendous progress towards Elandi's product-market fit (PMF); from discovery, to conceptual design and ML prototypes. Overall, the most important learning from our discovery is that fit-for-purpose content, improved UX, personalised recommendation, and expert mentoring are in high demand.While this progress continues, we realised that there is an urgent need for a comprehensive framework for AI assurance (e.g., auditing the AI, measuring users' trust in it, and quantifying its various risks) in the L&D space. Therefore, as part of our broader ambition, through this grant we would like to kickstart an explicit work stream on this topic. Our goal is to build on the latest research (including metrics and measurements) on trustworthy AI, to provide the tools and frameworks necessary for auditing AI models and products in the L&D space. While we will be focused on the L&D use cases, we expect our results to have implications beyond L&D and in other domains..An improved L&D offering will be of profound societal and commercial value; it will play a critical role in improving national and international productivity. It is aligned with strategic government initiatives such as "Levelling up", "Lifelong Learning", and "Skills for Life".