Highway of Algorithms—Driven by Data
“Change is the only constant in life,” said Heraclitus, and this fall, that truth felt more real than ever. As the leaves began to turn and the air grew crisp, I embarked on a journey that took me to three major conferences: FOSS4G in St. Louis, the National Credit Union Association (NCUA) Conference in Phoenix, and the Geospatial Professionals Network GIS-Pro Conference in Portland, Maine. Each event offered a unique glimpse into the evolving world of artificial intelligence, much like the shifting colors of autumn leaves.
The road ahead felt like a highway of algorithms, all driven by the data that fuels them. Just as fall symbolizes a time of transition, these conferences highlighted the need for adaptability in the rapidly changing landscape of AI.
FOSS4G Conference—St. Louis
At the FOSS4G conference in St. Louis, I teamed up with Tim Klawa from Figure Eight Federal to present on managing risk in AI systems using a Lean AI/ML approach. Our focus was on using just the right amount of training data to balance mission effectiveness and risk. We emphasized that fear of the unknown shouldn’t hold us back; instead, it should motivate us to tackle challenges directly.
We discussed the importance of aligning training data campaigns with business goals while minimizing various risks. One of the highlights was introducing our Adaptive Labeling Optimization approach. This method uses human-machine collaboration to improve labeling efficiency and identify gaps in data diversity. By finding the right balance between data quantity and quality, we can ensure that AI systems perform their best while reducing risks.
The audience was engaged and curious, asking insightful questions about how to implement these strategies in their own projects. It was encouraging to see so much interest in smart-sizing training data and managing AI risks effectively.
National Credit Union Association (NCUA) Conference—Phoenix
Next, I headed to Phoenix for the NCUA IT Subject Matter Expert Conference. There, I joined a panel discussion on the evolving role of AI in financial regulation. We talked about the urgency of addressing data risks in the financial sector. Waiting too long to properly curate and label data can lead to significant problems, especially in areas like fraud detection and compliance.
I stressed the need for accurate and consistent data labeling in financial environments. If data is mislabeled, it can lead to incorrect decisions and even regulatory penalties. We also discussed the careful use of synthetic data. If not properly validated, synthetic data can introduce bias into AI systems. Proactively identifying and correcting biases in training data is crucial, especially in financial services where fairness and equal opportunity are essential.
We wrapped up the discussion by focusing on continuous data monitoring and updates. As market conditions and regulations change, it’s vital to keep AI models relevant. This means regularly checking and updating the data that trains these models. The session was a reminder that in the world of finance, staying ahead of risks isn’t just important—it’s necessary.
Geospatial Professionals Network GIS-Pro Conference—Portland, Maine
Finally, I traveled to Portland, Maine, for the GIS-Pro Conference. With nearly three decades of experience in geospatial intelligence, it felt like returning home. I delivered a presentation titled “Boxing the GEOAI Compass: A Data-Centric Approach to Managing Risk and Promoting AI Adoption.” The phrase “boxing a compass” means making a complete change in direction, and that’s exactly what we discussed.
We talked about shifting focus away from AI models themselves and emphasizing the importance of training data. While AI models and their outputs often get the spotlight, the real key to a successful AI program lies in high-quality training data. Using an adaptive labeling strategy, we can iteratively improve data quality and reduce risk, especially as labeling campaigns scale up.
The audience was highly engaged, and we had nearly 40 minutes of questions after the presentation. One of the most thought-provoking questions was about the future of AI labeling: Will we need more or fewer human labelers? I believe we will need more. The vast amounts of data that generative AI systems are trained on aren’t always representative of society at large. Human labelers play a crucial role in ensuring that AI outputs reflect diverse perspectives and align with human values.
Embracing Change
As fall settles in and the leaves change color, it’s clear that AI is also undergoing a transformation. We must face the challenges in AI with courage, particularly when it comes to managing training data risks. The message from each event was clear: high-quality, diverse training data is the key to ensuring AI’s future success.
By implementing tailored AI risk frameworks and continuing to refine how we manage and label data, we can ensure that AI remains trustworthy and capable of driving real-world results. Just like the changing seasons, we must adapt and grow, embracing the new while learning from the past.
My journey across these conferences felt like a drive down a highway of algorithms, all driven by data. The experiences reminded me that in the midst of change and uncertainty, there is always a path forward. As we move into this new season, both in nature and in technology, let’s embrace the change and steer toward a future powered by data and guided by human values.
This fall has been a season of learning and reflection. The conferences reinforced the idea that while technology advances rapidly, it’s the human element—our ability to adapt, question, and guide—that shapes its impact. As the leaves continue to fall and we prepare for winter, I look forward to the challenges and opportunities that lie ahead in the ever-evolving field of AI.
