Trust but Verify – A Principle Engrained in AI Data Collection

I read an article recently about how the trust but verify principle applies in the workplace and it got me thinking about how it should also apply to the expansion of Artificial Intelligence (AI) in our society. At its core, this principle encourages us to start with trust, while also committing to accountability (i.e. verification). As we expand our use of AI in the world, the idea of human-centered AI ensures that processes and decisions are subject to human review, embedding human responsibility into the core of our AI systems and applying this important principle of trust but verify.

A powerful use of AI – traffic studies

Leveraging AI for the data-intensive task of capturing FHWA 13 vehicle classifications, vehicle counts, and speed, not only leverages this new technology to collect highly accurate data but improves the safety of workers by removing the need to work in dangerous conditions and environments. This symbiotic relationship allows the human professional to focus on areas requiring creative thinking, system planning, emotional intelligence, and strategic judgment, while relying on AI to handle the tasks that are typically repetitive, data-intensive, or demand a level of speed and accuracy.

Verification made possible with video validation

But what about the verification part of trust but verify? With more traditional data collection sensors, there is no way to go back and validate the data that was captured. This is where video streams play a critical role. With AI-driven traffic studies, we are putting eyes on the road (via video cameras) so that AI can quickly and accurately count, classify, and measure the speed of each vehicle on the road, faster and better than a human eye. But how do we know? Because we can always go back to the video to check the results.

• Need to perform QC on the study? → A human can review the video to compare results.
• Not sure why count data suddenly dropped off and then came back? → A human can review the video and determine if traffic stopped.
• Seeing a significant difference between new and historical truck classification? → A human expert in classification can review the video to validate the results.

Conclusion

When AI-powered edge processing is leveraged to automatically analyze live video streams of active traffic and deliver ground truth traffic studies, we are able to leverage AI to empower and enhance the efficiency of human workers, while also leveraging video to enhance the human oversight need to validate the results.