I-Corps: A Software Platform to Customize, Inspect and Improve Artificial Intelligence (AI) Systems

这个I-Corps项目更广泛的影响/商业潜力是开发一个软件平台，使人工智能（AI）模型更加可靠。人工智能正迅速成为日常业务和组织的一部分。然而，使用人工智能系统的关键问题是它们缺乏可靠性和可解释性，以及它们在内部工作方面缺乏透明度，其中输出推断和预测是不可解释的。这使得开发人工智能模型并检查和减轻其故障模式的过程既耗时又具有挑战性。该提议的技术旨在使用另一个人工智能系统自动开发、检查和改进人工智能模型，该系统在优化过程中使用人类反馈。理解和减轻人工智能模型的可靠性问题可以减轻其在实践中部署的风险。此外，这些努力可能会使非专家对人工智能系统的可靠使用民主化，并增加人类对这些系统的信任。I-Corps项目基于自动化和统一软件平台的开发，该平台提供多模态可解释性和可靠性分析和监控工具，用于设计、培训、检查和改进人工智能（AI）系统。拟议的技术旨在利用用户的独特数据自动发现和解决人工智能模型中隐藏的可靠性问题。它简化了识别和减轻潜在可靠性风险和可解释性挑战的复杂过程，这可能有助于确保人工智能模型提供可信和准确的结果。此外，用户可以比较数百种人工智能模型，并为其特定应用选择效率和可靠性最高的模型。它还交互式地将用户反馈纳入其优化中，以提高人工智能模型的可靠性和可解释性，同时可靠性变得透明和可管理，使用户能够更有信心地做出明智的决策。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

The broader impact/commercial potential of this I-Corps project is the development of a software platform to make Artificial Intelligence (AI) models more reliable. Artificial intelligence is rapidly becoming a part of everyday businesses and organizations. However, key concerns in using AI systems are their lack of reliability and explainability, and their lack of transparency with respect to internal workings where output inferences and predictions are not interpretable. This makes the process of developing AI models and inspecting and mitigating their failure modes time-consuming and challenging. The proposed technology is designed to automate developing, inspecting and improving AI models using another AI system that uses human feedback in its optimization. Understanding and mitigating reliability issues of AI models may mitigate the risks of their deployment in practice. In addition, these efforts may democratize the reliable use of AI systems by non-experts and increase human trust in these systems. This I-Corps project is based on the development of an automated and unified software platform that provides multi-modal interpretability and reliability analysis and monitoring tools to design, train, inspect, and improve Artificial Intelligence (AI) systems. The proposed technology is designed to automatically uncover and address hidden reliability issues within AI models employing the user’s unique data. It simplifies the complex process of identifying and mitigating potential reliability risks and explainability challenges, which may help to ensure AI models deliver trustworthy and accurate results. In addition, users may compare hundreds of AI models and select the ones with the maximum efficiency and reliability for their specific applications. It also interactively incorporates user feedback in its optimization to improve reliability and explainability of AI models while reliability becomes transparent and manageable, empowering users to make informed decisions with increased confidence.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.