revAIsor: Revise AI Solutions for Optimal Results

revAIsor：修改人工智能解决方案以获得最佳结果

• 批准号: 10075158
• 金额: $ 6.37万
• 依托单位: HELED LTD
• 依托单位国家: 英国
• 项目类别: Grant for R&D
• 财政年份: 2023
• 资助国家: 英国
• 起止时间: 2023 至 无数据
• 项目状态: 已结题
• 来源: https://gtr.ukri.org/projects?ref=10075158
• 关键词: revAIsor; Revise; AI; Solutions; Optimal

The revAIsor project aims to tackle the growing concern of trust and transparency in AI systems by developing an AI compliance platform that incorporates auditing using advanced tools. The goal is to create a platform that adopts a compliance-by-design approach, ensuring that AI systems are built with compliance in mind, and a life cycle management process, enabling the continuous monitoring and auditing of AI systems.As AI systems are being used more frequently in critical applications, the need for AI assurance is becoming increasingly important. While AI has the potential to provide significant benefits, it is crucial to ensure that these systems are trustworthy, reliable, and compliant with regulations.Recent studies have shown that AI models can contain biases and inaccuracies, leading to discriminatory outcomes. Moreover, GPT language models, including GPT-3, have been shown to generate biased and inaccurate responses in specific contexts. While these models have impressive capabilities, they are not infallible and require careful monitoring and evaluation to ensure they are providing accurate and unbiased results. Additionally, we will explore using synthetic data to test for simple bias and explainability, enhancing the reliability and robustness of the AI models.To address these issues, AI compliance by design is critical. This means ensuring that AI systems are designed with compliance and ethical considerations in mind rather than trying to retrofit compliance requirements onto existing systems.The revAIsor platform will be built using blockchain and web3 technology, ensuring that audit records are tamper-proof and secure. The platform will use decentralised storage and encryption protocols to store and protect sensitive audit data, making it virtually impossible to hack or manipulate.The platform will offer several features, including data and model validation, explainability analysis, bias and fairness testing, and compliance monitoring. These features will enable auditors to quickly and easily evaluate the quality and reliability of AI systems.By providing auditors with advanced features for data and model validation, bias and fairness testing, compliance monitoring, and explainability analysis, the revAIsor platform will play a critical role in ensuring the trustworthiness and compliance of AI systems. Auditing using advanced tools, such as the revAIsor platform, will enable auditors to evaluate the quality and reliability of AI systems more effectively, ultimately promoting transparency and trust in the development and use of AI technology.

RevAISor项目旨在通过开发一个AI合规平台来解决人工智能系统中日益增长的信任和透明度问题，该平台使用先进的工具进行审计。我们的目标是创建一个采用合规设计方法的平台，确保人工智能系统在构建时考虑合规性，并建立生命周期管理流程，实现对人工智能系统的持续监控和审计。随着人工智能系统在关键应用中的使用越来越频繁，对人工智能保证的需求变得越来越重要。虽然人工智能有潜力提供显著的好处，但确保这些系统值得信赖、可靠并符合法规至关重要。最近的研究表明，人工智能模型可能包含偏见和不准确，导致歧视性结果。此外，GPT语言模型，包括GPT-3，已被证明在特定的上下文中产生偏见和不准确的反应。虽然这些模型具有令人印象深刻的功能，但它们并不是绝对可靠的，需要仔细的监测和评估，以确保它们提供准确和公正的结果。此外，我们将探索使用合成数据来测试简单的偏差和可解释性，提高AI模型的可靠性和鲁棒性。为了解决这些问题，AI设计合规性至关重要。这意味着确保人工智能系统的设计考虑到合规性和道德因素，而不是试图将合规性要求改造到现有系统上。revAISor平台将使用区块链和web 3技术构建，确保审计记录防篡改和安全。该平台将使用分散式存储和加密协议来存储和保护敏感的审计数据，使其几乎不可能被黑客攻击或操纵。该平台将提供多项功能，包括数据和模型验证、可解释性分析、偏见和公平性测试以及合规性监控。这些功能将使审计人员能够快速轻松地评估人工智能系统的质量和可靠性，通过为审计人员提供数据和模型验证、偏见和公平性测试、合规性监控和可解释性分析等高级功能，revAISor平台将在确保人工智能系统的可信度和合规性方面发挥关键作用。使用revAISor平台等先进工具进行审计，将使审计人员能够更有效地评估人工智能系统的质量和可靠性，最终促进人工智能技术开发和使用的透明度和信任。