A software solution for trustworthy AI

越来越多的公司使用机器学习模型做出的预测来实现自动决策系统。来自自动系统的决策看起来是公正的，但它们反映了用于训练机器学习模型的数据中的问题，以及设计模型的工程师所做的选择。有缺陷的决策可能会伤害已经脆弱的消费者，反过来，对那些不加检查的公司产生负面的商业影响。自动决策缺乏透明度对我们社会的信任程度产生负面影响。黑箱方法在不解释驱动因素的情况下做出决策，削弱了人们对系统及其预测的信任。Etiq认识到，在处理客户数据的行业，特别是金融行业，公司都在努力应对这些风险。因此，我们正在开发一个ML/AI风险管理平台，在生产前和生产阶段提供错误检测、根本原因分析和修复建议。我们的平台将通过提供每个模型预测的驱动因素列表以及置信区间来进一步提高这种信任。它将使组织能够快速轻松地对机器学习模型和数据管道进行测试，然后识别数据和人工智能错误，意外偏见以及改进领域。通过这种方式，公司将能够监控模型性能并诊断问题，以帮助尽早解决问题，修复问题，从而开发可信的机器学习模型和数据管道。

Companies increasingly implement automatic decision-making systems using predictions made by machine learning models. Decisions from automatic systems can seem impartial, but they reflect problems in the data used to train the machine learning models and the choices made by engineers designing the models. Flawed decisions may harm already vulnerable consumers and, in turn, have a negative business impact on the companies that left their decisions unchecked.The lack of transparency in automatic decisions has a negative impact on the level of trust in our society. The black box approach, delivering decisions without explaining the driving factors, erodes trust in the system and its predictions. Etiq recognizes that companies in sectors dealing with customer data, especially the financial sector, struggle with these risks. As such we're developing an ML/AI risk management platform providing error detection, root cause analysis and repair recommendations across pre-production and production stages. Our platform will boost that trust further by providing a list of driving factors with each model prediction as well as a confidence interval. It will allow organisations to quickly and easily run tests on machine learning models and data pipelines, then identify data and AI errors, unintended biases, areas of improvement. This way companies will be able to monitor model performance as well as diagnose the issue to help root problems early, repair issues and thus develop trusted machine learning models and data pipelines.