Emergence transparency for enterprise agent-based models

基于企业代理的模型的出现透明度

• 批准号: 10066332
• 金额: $ 4.56万
• 依托单位: AEROGILITY LIMITED
• 依托单位国家: 英国
• 项目类别: Collaborative 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=10066332
• 关键词: Emergence; transparency; enterprise; agent; based

To decide long-term strategic policies, there is a need in large organisations to predict how these policies may play out in varied hypothetical futures. AI techniques can greatly help with this by accounting for diverse interconnected characteristics of an organisation in forecasting, but those using AI software need to be able to trust the outputs to apply them in their decision-making. _Transparency_, the ability to see an explanation of how an AI's output was arrived at, is vital to this trust, as without knowing the reasoning an AI performed was justified and complete, it is unwise to base a critical decision upon its conclusions. Artificial intelligence can be classified into data-driven and model-based approaches, where in data-driven techniques such as machine learning the AI's reasoning comes from a model derived automatically from processing data while in model-based AI that model is created from human expert knowledge and concepts. These classes of technique are applicable to different problems, and transparency is one inherent advantage of model-based AI. Within model-based AI, there are two kinds of transparency: _step transparency_, being able to explain each step in reasoning towards a prediction, and _emergence transparency_, being able to explain the predicted outcomes given a vast number of interdependent steps. While existing systems offer step transparency through traceability, visualisation, etc., emergence transparency is a largely unaddressed problem. For Aerogility, the lack of emergence transparency in commercial AI technology limits the business problems to which it can apply, both in aerospace, where most of our current clients operate, and in new sectors. In this project, we will conduct a feasibility study on research approaches which may, in complement, provide emergence transparency and construct a consortium to provide the full set of skills required to implement and evaluate the solution.

为了决定长期战略政策，大型组织需要预测这些政策在各种假设的未来可能会如何发挥作用。人工智能技术可以在预测中考虑到组织的各种相互关联的特征，从而极大地帮助实现这一点，但那些使用人工智能软件的人需要能够信任输出，以便在决策中应用它们。_透明度_，即看到人工智能输出是如何得出的解释的能力，对这种信任至关重要，因为在不知道人工智能进行的推理是合理和完整的情况下，基于其结论做出关键决定是不明智的。人工智能可以分为数据驱动的方法和基于模型的方法，在机器学习等数据驱动的技术中，人工智能的推理来自于处理数据自动得出的模型，而在基于模型的人工智能中，该模型是根据人类的专家知识和概念创建的。这些技术类别适用于不同的问题，而透明度是基于模型的人工智能的固有优势之一。在基于模型的人工智能中，有两种透明度：一种是步骤透明，能够解释推理中的每一步预测；另一种是浮现透明，能够解释给定大量相互依赖的步骤的预测结果。虽然现有的系统通过可追溯性、可视化等提供步骤透明度，但出现透明度在很大程度上是一个尚未解决的问题。对于Aerogility来说，商业人工智能技术缺乏出现透明度，限制了它可以应用于的商业问题，无论是在我们目前大多数客户运营的航空航天领域，还是在新的行业。在这个项目中，我们将对研究方法进行可行性研究，这些方法可以提供应急透明度，并建立一个联合体，提供实施和评估解决方案所需的全套技能。