BACKGROUND

The innovation of generative artificial intelligence (gen AI) tools offers new opportunities in curating, synthesizing, managing, and transferring knowledge. Gen AI tools have potential for leveraging state departments of transportation’s (DOTs) resources to assist with tasks such as producing an initial draft of a literature review, generating executive summaries, and responding to user prompts and questions with detailed insightful feedback from the library of existing resources. Gen AI tools are poised to revolutionize accessing knowledge by reducing the time and effort for retrieving relevant information, using and synthesizing information, enhancing decision-making, facilitating best practice exchange, and providing engaging personalized user experiences. 

As DOTs' interest in adopting gen AI tools for knowledge management grows, research is needed to create suitable knowledge management gen AI tools for transportation agencies.  

OBJECTIVE

The objective of this research is to explore gen AI's potential for enhancing knowledge management at transportation agencies by (1) developing a scalable framework to use gen AI tools for improved knowledge management, (2) designing a trustworthy human-AI collaboration prototype that leverages DOTs' resources to address common transportation systems management and operations (TSMO) scenarios within the established framework, and (3) implementing a pilot deploying the identified prototype to demonstrate its efficiency and effectiveness. 

Accomplishment of the project objective will require at least the following tasks.

RESEARCH PLAN

The National Operations Center of Excellence (NOCoE) Knowledge Center is the centerpiece and most accessed site of NOCoE’s website. It houses over 2,600 resources on diverse topics of interest for TSMO practitioners, including case studies, research reports, Federal Highway Administration (FHWA) resources, and state and local agency resources. It also has a search function so users can quickly find and shortlist resources of interest. Given the wide variety of available resources, NOCoE is an ideal candidate for enhancing its knowledge management capabilities. The pilot of this research shall focus on the NOCoE's resources.

At minimum, the proposed tasks should include:

• Review, identify, and synthesize the current gen AI tools and capabilities that are suitable for knowledge retrieval, generation, and management at DOTs.
• Acquire stakeholders’ feedback on how everyday NOCoE users would use gen AI tools and how state DOTs and other transportation agencies would be interested in using gen AI tools. A TSMO expert group shall also be identified to provide feedback on gen AI tool fine-tuning in collaboration with NOCoE.
• Organize and prepare the NOCoE Knowledge Center's 2,600 resources by cataloging, cleaning, and aligning them with defined gen AI objectives and NOCoE website TSMO needs, while addressing ethical and technical considerations.
• Use the findings from Tasks 1 through 3 to develop a trustworthy (considering robustness, generalization, explainability, transparency, reproducibility, fairness, privacy preservation, and accountability) and scalable framework for the NOCoE and other transportation agencies to use gen AI tools for improved knowledge management.
• Identify common TSMO scenarios (e.g., freeway management, arterial management, traffic incident management, road weather management, special event management, emergency transportation operation, parking management, work zone management) for the gen AI tool application with input from the TSMO expert group from Task 2. Include training curriculums/modules for the TSMO-related workforce.
• Design a trustworthy human-AI collaboration prototype for the NOCoE knowledge center, develop user prompts for human-AI interactions, and establish a mechanism to incorporate a feedback loop for continuous evolution of the gen AI tool with the identified TSMO expert group’s input in collaboration with NOCoE.
• Develop evaluation criteria and testing protocols for the gen AI tool that cover ethics, efficiency, and effectiveness (e.g., the Exemplar Super-COmpression and Regeneration based on prompTs [ESCORT] test).
• Deploy a prototype gen AI tool pilot, including software framework and user interface (UI) development through the NOCoE website. 

The research plan should build in appropriate checkpoints with the NCHRP project panel, including at a minimum (1) a kickoff teleconference meeting to be held within 1 month of the contract’s execution date, (2) at least one face-to-face interim deliverable review meeting, and (3) web-enabled teleconferences tied to panel review and/or NCHRP approval of interim deliverables.

The research plan will describe appropriate deliverables (which also represent key project milestones), including:

• An interim report and panel meeting. The interim report should include the analyses and results of completed tasks, a plan for the remaining tasks, and a detailed outline of the final research product(s). The panel meeting will take place in Washington, DC, after the panel review of the interim report. The interim report and panel meeting should occur after the expenditure of about 40 to 50 percent of the project budget. No further work is allowed until the interim report is approved by the NCHRP. 
• Final deliverables. The final deliverables shall include a useful resource for practitioners and stakeholders of transportation and planning agencies, and a research report that documents the research process.  
• A technical memorandum titled “Implementation of Research Findings and Products,” including a hardware improvement strategy for transportation agencies and a stakeholder communication plan.
• A slide deck for use in webinars with presenter’s notes that presents the research findings and conclusions.
• A pilot deploying the identified prototype within the NOCoE knowledge management system to demonstrate its efficiency and effectiveness.
• Training materials for NOCoE staff to operate, manage, maintain, and upgrade the prototype gen AI tool. 

STATUS: Proposals have been received in response to the RFP. The project panel will meet to select a contractor to perform the work.