The Path to Ethical AI is Through Regulation
The conversation about AI in regulatory compliance that we started in the last post will not be complete without discussing the ethical issues of AI. There are two main reasons I wanted to bring it up for the attention of regulatory compliance professionals. The first reason has to do with the risks brought about by biased data used to train AI algorithms, something that we can experience both as business users and as individuals. Biased outcomes are due to the inherent bias in the data used to train (i.e., to “teach”) the algorithm to recognize patterns. In short, the data commonly available for training reflects discriminatory prejudices of society and will lead to unfair outcomes. We need to be aware of such biases as individuals and users of AI-based products. At an enterprise level, model risk management is an evolving field of technology, which we at Citation Compliance™ are following to keep our perspective on the market current.
Business leaders responsible for implementing AI-based products (such as heads of compliance) need to plan for ways to mitigate the inherent ethical biases of data, which will differ depending on the type of product, the jurisdiction, the scale of operations and other factors. Amazon’s Sage Maker Clarify is one example of tools available to detect bias in raw data, in the model (algorithm) or the specific features of the AI model, with regulatory compliance cited as one of the use cases.
This segues to the second reason regulatory compliance professionals will benefit from following the AI ethics dynamics. The emerging regulation to control related risks will soon become part of many regulatory practices in addition to privacy regulations, making it directly relevant for regulatory professionals. The existing antidiscrimination legislation applies to biased outcomes, but only to the point at which individuals are responsible for decisions. With more products incorporating AI capabilities, more services and processes outsourced to AI, individual accountability becomes less black and white, and the scale of potential bias balloons to unprecedented proportions.
The way to mitigate it is through governance and regulation, involving dedicated audits, more stringent data governance protocols, enforcement of traceability and monitoring of AI outputs, and related training. The US ad EU is leading the way with initiatives such as the Equal Credit Opportunity Act and this Regulatory framework proposal on artificial intelligence, and more stringent regulations are in the works. Corporations are developing internal best practices and tools, appointing executives to oversee ethics governance boards, getting ready to compete on AI ethics and providing input and context for the upcoming regulation. Speaking of tributaries for regulation, there are over 70 standards published and emerging to establish a baseline for AI governance across industries. A leading standards developer, IEEE has launched a certification program for ethical AI design.
This activity points to the fact that we should expect widespread AI-related regulation in the near future in all professional fields using data. When it happens, Citation Manage™ will ensure you are prepared. We scan hundreds of regulatory publications daily and update the system with all changes, ranking them on applicability to your business. This way, you will be notified of mission-critical updates such as AI ethics regulations when it comes to your jurisdiction and will have ample time to design compliance protocols.
Arrange a demo today to see the platform in action.
