By Wendy McCoy, Chief Strategist

As an early advocate for AI-enhanced solutions during my time at IBM, it's surprising to suggest that family offices carefully consider the deployment of cutting-edge AI tools. While these tools can certainly support lean staffing models managing vast amounts of data, it's crucial to assess the specific type of AI being implemented. Understanding how its model was trained, its impact, and the security measures in place is essential.

AI in this context isn't new. Algorithms enabling machines to learn without explicit programming have been around since the 1950s. Although these early algorithms were basic, they've evolved significantly. Today, most commercial AI applications are 'Narrow AI,' excelling in specific areas and easier to control regarding training data impact and result exposure.

For instance, Ledgex, a provider of family office software, uses Narrow AI to streamline operations by automating data extraction from financial documents like capital calls and K-1s, significantly reducing manual effort. As long as these applications follow best practice security protocols, the risk is minimal.

The greater risk arises with solutions developed in the race to Artificial General Intelligence (AGI). Companies like OpenAI, Microsoft, and Google use Large Language Models (LLMs) to offer broad capabilities across a wide range of topics. These models generate answers based on their training data, which often includes personal information. While the standard public interface is generally safe, fine-tuning the LLMs through the use of APIs for specific tasks, such as tax-related queries, can compromise access to the training data.

Dr. Prateek Mittal from Princeton University notes that no commercially available LLMs have strong privacy defenses. This presents a significant risk as these models may inadvertently learn and recall sensitive information. Google's recent AI integration efforts, despite being error-prone, highlight the pressure on companies to innovate quickly, potentially further compromising security and accuracy. Such challenges underscore the need for thorough evaluation when considering software incorporating LLM features.

Family offices need automation for effective data analysis. Narrow AIis a reliable solution that can save countless man-hours over time, provided strict cybersecurity protocols are in place. However, Generative AI using LLMs introduces additional risks, from simple errors to privacy concerns. Major players are working to enhance security and accuracy, but the potential for bad actors to exploit vulnerabilities remains high. Therefore, a thoughtful approach is essential.

About Ledgex

Ledgex was created by investment office professionals to address multi-asset data quality challenges. The company enables investment firms to manage complex asset portfolios with significant improvements in data accuracy, transparency, and timeliness. For more information or to book a demo, please call (888) 748-0933, email info@ledgex.com, or visit www.ledgex.com.