Greater Explainability in AI Will Drive Increased Engagement — JPMC Head of Marketing Data and Analytics

(US & Canada) Dr. Tiffany Perkins-Munn, Managing Director, Head of Marketing Data and Analytics at JPMC, speaks with David Arturi, Head of Financial Services at Lydonia, in a video interview about AI adoption in the workplace, the approach to bias and transparency in AI-driven initiatives, and the role of AI agents and the need for human intervention.

GenerativeAI (GenAI) has brought people closer to understanding the possibilities with AI, says Perkins-Munn, as she shares her opinion on AI adoption in the workplace. She adds how GenAI has made AI tangible for regular job holders as something they can connect with, making it more accessible as a theory.

According to Perkins-Munn, the perception of AI being a threat to jobs is gradually changing into AI being a useful tool. She believes that as this thought seeps more into society, and with the natural inclusion of AI as an enabler, the idea that AI is here to take over will die down.

Having said that, Perkins-Munn affirms that AI will probably replace certain jobs that are logic-based and functionally easy to execute. In doing so, it will also open a myriad of other jobs and opportunities. She takes the instance of the AI-born role of a prompt engineer and the data career.

The upside of what AI and ML have to offer surpasses its downside. Further, as people get more involved with it, there will be less concern over AI becoming sentient and taking over the world.

When asked about her approach to bias and transparency in AI-driven initiatives, Perkins-Munn first highlights the importance of having diverse data sets to mitigate bias. She notes that organizations must ensure that training data represents diverse customer segments to avoid skewed results.

Moreover, regularly auditing datasets for potential biases is critical, says Perkins-Munn. For AI executors, she advises the implementation of fairness constraints in AI models with techniques like adversarial de-biasing or equal opportunity methods to create algorithmic fairness.

Perkins-Munn also advocates for periodic audits of the outputs to detect and correct unintended biases while using tools and frameworks designed to detect algorithmic biases. This could include fostering diversity and having AI development teams bring varied perspectives and reduce unconscious bias.

To reduce bias in financial products, Perkins-Munn suggests paying attention to whether there are any biases in credit scoring, loan approvals, or insurance offerings.

Speaking of transparency, she mentions explainable AI and recommends using interpretable AI models wherever possible, especially for decision-making processes that impact customers. Perkins-Munn further refers to leveraging the powerful Local Interpretable Model-Agnostic Explanations (LIME) and SHapley Additive exPlanations (SHAP) techniques for explaining the predictions of machine learning models.

Delving further, she states that as AI becomes more explainable and people understand the factors behind AI-driven decision-making, it will increase people’s engagement.

In this scenario, clear communication is the key when it comes to informing customers when they are interacting with AI systems. For example, platforms like Google make it explicit by labeling responses as AI-generated or providing jargon-free explanations of how AI is utilized in marketing activities, enhancing transparency.

It boils down to communicating with customers to help them understand AI through explainability, effective communication, and model documentation, says Perkins-Munn. Reflecting on the challenges, she mentions the creation of black boxes that are not auditable.

The focus should be on building an AI system that is auditable by both internal teams and external regulators to ensure credibility. Additionally, customer control is fundamental, states Perkins-Munn, as she stresses the need for that partnership.

Organizations must empower customers to have control of their data by understanding how it is used in AI systems and being given options to view, correct, or delete their data.

Moving forward, Perkins-Munn reiterates how AI and ML have been around for years, but GenAI has fuelled the interest of people. It has connected the larger society to what AI is and enabled them to participate in a conversation around AI as a user of the tool. Therefore, it is crucial to assess the future scope of AI, engaging with customers and the public while having regulations in place.

Sharing her opinion on the role of AI agents, Perkins-Munn states that she does not see AI agents becoming autonomous. She adds that while the agent can mimic autonomy, it would not be real as human intervention is critical in getting the agent up and running and for its continuous improvement.

Therefore, Perkins-Munn believes that the idea of autonomy needs to be redefined, as it will not be autonomous with human engagement and decision-making going into the process. This also makes the focus on bias paramount to ensuring everything is thought through in all stages of using the tool.

Concluding, Perkins-Munn states that there will always be a dichotomy where some people fully embrace digital solutions, while others refuse to engage with online systems altogether. The same principle applies to autonomous systems where there will be a need for human intervention because not everyone will feel comfortable relinquishing control to automated technologies.

CDO Magazine appreciates Dr. Tiffany Perkins-Munn for sharing her insights with our global community.