如何建立一家以人工智能为中心的反舞弊公司？

标签：浏览量：0 2022-10-14

Artificial intelligence (AI) is changing how businesses run — especially in anti-fraud and compliance. Here we describe how to build an AI-focused, anti-fraud technology company and what mindsets, strategies and tactics work well for organizations and entrepreneurs.

人工智能正在改变企业的经营方式，尤其是在反舞弊和合规方面。在这里，我们描述如何建立一家以人工智能为中心的反舞弊技术公司，以及哪些思维模式、战略和战术适合组织和企业家。

人工智能与反舞弊的融合

Convergence of AI and anti-fraud

The ACFE’s Occupational Fraud 2022: A Report to the Nations notes the median financial loss per case was $117,000, with over one in five cases having losses more than $1 million. Half of those cases occurred because of a lack of internal controls or an override of existing controls. Clearly, there’s an ROI case to be made for improving internal controls and increasing compliance monitoring.

ACFE的《2022年职业舞弊：致各国报告》指出，每个案例的财务损失中值为11.7万美金，超过五分之一的案例的损失超过100万美金。这些案例中有一半是由于缺乏内部控制或对现有控制的凌驾。显然，改善内部控制和增强合规性监控有一个具有投资回报率的案例。

And in its most recent guidance, the U.S. Department of Justice (DOJ) expressed its support for machine learning and collaboration across companies in a secure, data-sharing-type consortium.

美国司法部（DOJ）在其*新的指导意见中表示要大力支持机器学习和跨公司在安全、数据共享型联盟中的协作。

1、初创企业：专注于人工智能意味着什么

Startups: What it means to be AI-focused

The goals of building a startup technology company versus an AI-focused one are notably different.

建立一家初创技术公司与一家专注于人工智能的公司的目标明显不同。

Instead of trying to get a product out the door, an AI-focused company is attempting to make its predictive model(s) accurate. Instead of having traditional product features as milestones, an AI-focused company has measurable model results. The output is a prediction (e.g., 25% likely to be a potentially improper payment) versus a mathematical calculation, or a rules-based test or query.

一家专注于人工智能的公司没有试图推出产品，而是试图使其预测模型准确。专注于人工智能的公司不会将传统的产品功能作为里程碑，而是关注具有可测量的模型结果。输出的是预测（例如，25%可能是潜在的不当支付）或是数学计算或是基于规则的测试或查询。

2、如何打造一支专注于人工智能的明星团队

How to build an AI-focused, star team

So, here’s the good news — as a fraud examiner, you’ve already completed the first checkbox and are on your way to building an AI-driven team. Looking at Table 2, you’ll note that building an AI-focused team doesn’t start with software engineers. It starts with you, the anti-fraud professional with the expertise to ask the right risk questions. And you don’t need to have all these individuals on your team from day one. Bring on the data analyst first, then seek to add the data scientist and so forth as your number of recoveries and demand increase. One warning, however, as you move down the chart: Market demand (and hence, the cost) of these individuals goes up.

所以，好消息是，作为一名舞弊审查师，您已经完成了第一个复选框，并且正在构建一个人工智能驱动的团队。查看表2，您会注意到，构建一个专注于人工智能的团队并不是从软件工程师开始的。它从您开始，您是反舞弊专业人士，具有提出正确风险问题的专业知识。而且不需要从一开始就让所有这些人都加入你的团队。首先，聘请数据分析师，然后随着恢复次数和需求的增加，寻求添加数据科学家等等。（其他相关职位，详见下表）

3、有哪些要考虑的人工智能工具？

AI tools to consider？

Data scientists can’t do much without the right tools. Giving them the resources to do great work helps inspire creativity.

没有适当的工具，数据科学家也做不了什么。给他们提供良好工作的资源有助于激发创造力。(下表是按不同类型列出的一些可使用工具)

4、机器学习算法

Machine-learning algorithms

There are five general categories of machine-learning (ML) algorithms: supervised, unsupervised, reinforcement, transfer and deep learning. The following are brief introductions to the categories.

机器学习（ML）算法有五大类：有监督、无监督、强化、转移和深度学习。以下是对类别的简要介绍。

Supervised learning is ideal when data is available, but the algorithm is unknown or missing. Supervised ML methods include random forest trees, decision trees, regression and neural networks. They’re quite often used to find patterns or “profiles” of potentially improper transactions and risk-driving variables.

当数据可用，但算法未知或缺失时，监督学习是理想的。监督机器学习方法包括随机森林树、决策树、回归分析和神经网络。它们经常被用来寻找潜在不正当交易和风险驱动变量的模式或“概况”。
Unsupervised learning, on the other hand, is ideal when there’s less information about the risks, but you want the data to help define itself by grouping like events (or transactions) together. This can be particularly helpful in fraud detection when you’re looking for anomalies or patterns in data without applying any preset rules.Techniques that can be used in unsupervised learning include K-means clustering and Apriori algorithms.

当关于风险的信息较少时，但希望数据通过将类似事件（或事务）分组在一起来帮助定义他们自身时，无监督学习是理想的，。在不应用任何预设规则的情况下查找数据中的异常或模式时，这对舞弊检测特别有用。可用于无监督学习的技术包括K均值聚类和先验算法。
Reinforcement allows a user to decide the best action based on the current state and learned behaviors that maximize the rewards. This approach is often used in robotics where the computer trains itself continually using trial and error. The machine learns from experience and tries to capture the best possible knowledge to make accurate business decisions. In fraud detection, reinforcement techniques can be helpful with some of the necessary data extraction and cleanup required to prepare data for analysis, for example.

强化允许用户基于当前状态和学习行为来决定*佳行为，从而*大化奖励。这种方法通常用于机器人技术，其中计算机不断地使用试错法训练自己。机器从经验中学习，并试图获取尽可能好的知识，以做出准确的业务决策。例如，在舞弊检测中，增强技术有助于准备分析数据所需的一些必要的提取和清理。
Transfer learning focuses on storing knowledge gained while solving one problem and applying it to a different, but related, problem. For example, knowledge gained while learning to recognize cars could apply when trying to recognize trucks. Transfer learning is good when problems are similar, the time to train a model is limited and results are needed fast. Bayesian networks and Markov logic networks are effective transfer learning methods. In an anti-fraud context, transfer learning can help uncover conflicts of interest by finding hidden patterns and relationships, such as in an unauthorized employee and vendor relationship.

转移学习侧重于存储在解决一个问题时获得的知识，并将其应用于不同但相关的问题。例如，学习识别汽车时获得的知识可以应用于识别卡车。当问题相似、训练模型的时间有限且需要快速获得结果时，转移学习是很好的。贝叶斯网络和马尔可夫逻辑网络是有效的转移学习方法。在反舞弊环境中，转移学习可以通过发现隐藏的模式和关系（如未经授权的员工和供应商关系）来帮助发现利益冲突。
Finally, there’s deep learning. According to IBM, deep learning attempts to mimic the human brain — albeit far from matching its ability — enabling systems to cluster data and make predictions with incredible accuracy. Deep learning is ideal when there’s lots of unstructured time series data or data that’s not independent. Deep learning drives many AI applications and services that improve automation, performing analytical and physical tasks without human intervention. Deep-learning technology lies behind everyday products and services, such as digital assistants, voice-enabled TV remotes and credit card fraud detection, as well as emerging technologies, such as self-driving cars.
*后，还有深度学习。根据IBM的说法，深度学习试图模仿人脑——尽管远不及人脑的能力——使系统能够对数据进行聚类，并以令人难以置信的精度进行预测。当存在大量非结构化时间序列数据或非独立数据时，深度学习非常理想。深度学习推动了许多人工智能应用和服务，这些应用和服务可以提高自动化，在不需要人工干预的情况下执行分析和物理任务。深度学习技术是日常产品和服务的基础，如数字助理、语音电视遥控器和信用卡舞弊检测，以及自动驾驶汽车等新兴技术。