Information Governance | Compliance | Artificial Intelligence (AI)
At a recent industry panel, Bill Murphy, Chief Executive Officer of RedZone Technologies, moderated a discussion featuring three distinguished information management experts:
- Reggie Henry, CAE, Chief Information & Performance Excellence Officer, American Society of Association Executives (ASAE)
- Candace McCabe, CIP, Sr. Solution Architect - Governance, Risk, and Compliance, Walmart
- Mark Patrick, CIP, Records Officer, US Department of Defense
The panel was part of the AIIM Information and Data Leadership Symposium on October 1, 2024 in Arlington, VA and explored crucial topics in information management, from the value of data to the challenges of innovation. Here are my key takeaways:
The True Value of Information Management
Candace McCabe emphasized a critical point: "No one cares about compliance." She explained that when promoting information management, it's essential to focus on the value it brings rather than compliance alone. Decision-makers are interested in the benefits of compliance, such as findable, accessible, and accurate information, rather than compliance for its own sake.
Mark Patrick agreed, adding, "Your selling point is not compliance. It's that you are making jobs easier." This perspective shift is crucial for information management professionals seeking to gain support for their initiatives.
Leveraging Unstructured Data: ASAE's Innovative Approach
Reggie Henry shared an exciting project at ASAE, where they're building a large language model (LLM) trained on years of documents and resources. This initiative came after realizing that only 25% of ASAE's vast library was readily accessible to members.
The goal? To create an LLM that can provide accurate, industry-specific answers to member questions. Henry described this as moving from a "choices economy" to an "answer economy." In a choices economy, users are presented with various resources to learn about a topic. An answer economy, on the other hand, provides immediate, precise responses to specific queries.
This project hasn't been without challenges. Henry discussed navigating legal and ethical questions, such as whether bias testing was necessary for the LLM. By relying on precedent - noting that the source content had never undergone bias testing - they determined a practical approach to move forward.
Data Accessibility and Democracy
An audience member raised concerns about data feeling trapped in a "Bermuda triangle" of security, retention, and ownership issues. The panelists offered their perspectives on improving data accessibility:
- At ASAE, Henry introduced the concept of "data democracy," aiming to make all data not just accessible, but also shareable and usable.
- The Department of Defense is working towards "zero click" records management, inspired by the cybersecurity concept of "zero trust." Patrick explained that this initiative aims to reduce the steps required to access desired data, making policy execution more feasible for users.
Quality of Data: Finding the Right Balance
The panelists agreed that good data is findable and usable. Patrick emphasized the importance of thinking beyond policy to execution: "If users can't execute, it's bad policy."
To improve data quality and usability, the panelists suggested:
- Reducing data volume: Patrick recommended starting with "taking out the trash" by implementing default deletion dates for records.
- Understanding your data volume: Henry noted that most organizations underestimate their unstructured data volume.
- Identifying strategic data: ASAE had to determine what data was truly needed to help stakeholders do their jobs better.
Interestingly, the panelists agreed that data doesn't have to be perfect to be usable. They suggested scoring data quality and making strategic decisions based on these scores. This approach is particularly relevant when considering data use in AI applications, where organizations must decide if complete (but possibly imperfect) data is sufficient.
The Role of Human Intervention in Managing LLMs
As organizations increasingly adopt large language models, the panelists highlighted the ongoing need for human oversight:
- McCabe advocated for user involvement in vetting and continuously training LLMs.
- Henry stressed the importance of "feeding the beast" - providing data to the LLM, a task requiring specific skills.
- Patrick noted the growing demand for IT professionals who understand records management, especially in government AI implementations.
McCabe pointed out that information management practitioners already possess many of the skills needed to advance AI in organizations. "We should learn from the past and reuse that knowledge," she said, drawing parallels between current AI needs and past data center implementations.
In conclusion, as the information management landscape evolves, professionals in this field must balance compliance with value creation, leverage new technologies wisely, and continue to advocate for the strategic importance of their work. By focusing on these areas, information managers can drive innovation while maintaining the integrity and accessibility of their organization's most valuable asset: its data.
