As the use of artificial intelligence (AI) continues to grow, data custodians are in a unique position to harness the capabilities of AI to improve the quality, accessibility, and integration of core traffic records. Using AI can help automate data processing, streamline end user reporting, and increase the capabilities of both the data sets and the humans who steward those repositories. However, like all technological advances, AI is not a solution to every problem and data custodians are in a unique position to provide guardrails and governance.
Pros of Artificial Intelligence
Automate data processing and cleaning: Data sets are growing exponentially, and as the size of data grows, so does the burden on data custodians to process and clean them. AI tools can automate repetitive tasks like data processing and cleaning, often with improved efficiency. This lowers the number of manual tasks that must be performed by humans and lowers processing time.
Increased data linkage capabilities: AI can help link disparate records from multiple sources when they do not share a reliable unique identifier. A trained model can consider variations in names and addresses, typos, dates of birth, and other fields together to assign a probability that two records belong to the same person or event.
Impute missing data/synthetic data generation: AI tools can impute data that is missing from a data set as well as test how realistic the imputed values are, especially when columns rely heavily on one another in non-obvious ways.
Improve data analysis: AI can review massive data sets quickly and use historical data to make predictive forecasts, often finding patterns and links that human analysts cannot see. Widely available low-code/no-code platforms combined with natural language processing bring data analysis and visualization tools to a wider range of end users.
Cons of Artificial Intelligence
Always keep a human-in-the-loop: While AI tools are powerful, no model or process is perfect and active human supervision, training, and oversight of AI inputs and outputs is vital. For example, as mentioned above, AI can impute missing data and create linkages across data sets. However, this can also work in reverse – AI can create systematic errors across massive data sets or create false matches and missed matches.
Watch for bias: Particularly with generative AI, all outputs must be reviewed for bias. AI tools and models are trained on existing data sets which can contain implicit or explicit bias. For example, studies have shown that language models can produce different recommendations based on demographic cues embedded in names and writing styles.
The answer may be a hallucination: Generative AI can produce information that appears authentic but is largely a fabrication. While tools and models have improved over the past few years, a 2025 study found that ChatGPT only reached approximately 80% accuracy. This accuracy varies widely with higher rates of error found in increasingly specialized knowledge domains.
Increased need for data security: Using AI tools brings an increased need for data security, especially when implementing tools from a third-party. AI models are often a “black box” and how they access, store, and process data can be opaque and opens new opportunities for breaches. It is critical to implement, monitor, and maintain a strong governance framework. AI tools should follow the same role-based access controls, data-sharing agreements, retention requirements, and audit procedures as other data systems.
Expensive up-front implementation: Developing and implementing AI tools, especially for more complicated tasks, can be expensive. The benefits of implementation may balance out the cost over time but up-front costs can be prohibitive.
Lowered barriers for data analysis: Improved access to data reporting and analysis is a double-edged sword. Users can bypass security protocols, exposing proprietary information. They can also lack the needed context to evaluate results.
Conclusion
AI in all its forms is a very promising technological advancement. It can be a force multiplier that supports core objectives and increases human capabilities. However, it is critical that AI is implemented in a thoughtful, strategic, and ethical way that improves core traffic records and gives those that rely on them greater insight.
Additional Resources
TxDOT AI Strategic Plan: https://www.txdot.gov/content/dam/docs/division/str/txdot-ai-strategic-plan-update-jan-2026.pdf
USDOT AI Activities: https://www.transportation.gov/AI
MIT AI Foundations: https://mitsloanedtech.mit.edu/ai/basics/
Eno Center Special Edition: Artificial Intelligence in Transportation: Uses and Lessons Learned: https://enotrans.org/issue/special-edition-artificial-intelligence-in-transportation-uses-and-lessons-learned-mid-august-2026/
How to avoid machine learning pitfalls: https://arxiv.org/html/2108.02497v4/
A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions: https://dl.acm.org/doi/10.1145/3703155