ASCE – Artificial Intelligence for Complex Information

In the architecture, engineering, and construction industry, practitioners rely on ever-changing technical standards to manage and execute infrastructure projects. These crucial documents are updated inconsistently on varying schedules, leading to complications in staying informed about the latest guidelines.

Natural language processing which is a form of artificial intelligence (AI) — is transforming how we interact with complex information, and by harnessing the power of large language models, (another form of AI) altering large volumes of text into interconnected references is possible. This powerful combination of AI technologies streamlines access to information and makes the complex material more user-friendly, thereby instilling a sense of confidence in practitioners’ work.

AlfKa, an engineering services firm, manages statewide training programs for the Florida Department of Transportation (FDOT). AlfKa recognised the need for new AI tools so they initiated a three-month pilot project, using the FDOT Design Manual (FDM) to create a technical agent the project team named the FDM Navigator. The FDM is the guiding technical document FDOT uses to design all projects — construction, reconstruction, and resurfacing — on its state highway and national highway systems.

The first step in creating the technical agent was to define a clear goal: to provide users with prompt access to the FDM via a user-friendly interface. The next step in the development was to narrow the scope of the application to verify that it would function within the parameters of technical information in the manual.

The team began by cross-referencing the manual’s three key sections — Development and Processes, Design Criteria, and Plans Production — within the main database. For this pilot project, these limiting guardrails were enforced to assist in funnelling the formulation of answers.

The project team established a classification system to direct user queries into four initial categories: development and processes questions, design criteria questions, plans production questions, and general integrative design questions. Thus, if the agent is unable to find a proper classification, it would generate a response requesting a more detailed question or only addresses the technical content of the FDM.

The development of the database within FDM began by categorising content into text, figures, and tables, then converting these data into vectors for easy integration. The FDM’s standardised structure helped create semantic data vectors, making it seamless to cross-reference within the manual.

The tool’s structure was designed to be easily updatable, with section numbers for identification, allowing revisions to be streamlined. Different manual editions allow creation of an update history, helping users track changes and evaluate past criteria.

Preprocessing the data with conversational AI applications enhanced the Navigator’s efficiency by integrating text into its core informational elements, removing redundancies, and tokenizing the text. Tokenizing is a key element which enhances the model’s ability to establish statistical connections between words, sentences, and technical meanings within the content.

The FDM consists of specific terminologies unique to its technical purpose, which general conversational AI agents often struggle to understand. To address this, additional model training was required to fine-tune the technical agent to the specifics of the manual terminology. Interactions could span multiple exchanges, building context around the user’s questions and helping the Navigator classify and tailor its responses better.

The Navigator follows a three-step workflow when responding to user queries. The technical agent has to:
1. Classify the question into one of the five categories shown in Figure 3.
2. Provide a response that meets the answer criteria predefined for each question category.
3. Use embeddings to retrieve the most relevant information from the database.

The crucial step in testing the technical agent was the inclusion of common guardrails against misuse or user abuse, such as hate speech or misinformation. With the help of widely available datasets from public conversational AI agents that provide information on monitoring adverse content, implementing these safeguards requires less effort.

The deployment of the FDM Navigator, like other IT-based engineering support tools, restricts the technical tool to registered users. The registered-user system also allows for the traceability of user queries which supports provision of additional data to strengthen the underlying technical agent and its database. Additionally, it provides valuable insights for enhancing the underlying manuals and guidelines.

– Enhanced cross-disciplinary insights: The Navigator supports users in discovering connections to technical areas beyond their expertise, facilitating more informed decision-making.

– Accelerated learning curve: Tailored for beginners, the tool lowers the learning curve for comprehending complex technical manuals, enabling faster adaptation to new jurisdictions.

– Historical data navigation: The tool allows experts to view previous versions of technical manuals, so they can effectively evaluate changes and access older projects.

– Enhanced planning and predictive analytics: Predictive analytics uses user interaction data to recommend modifications to the design manual based on patterns identified.

– Streamlined manual updates: The vectorised database effectively executes changes and ensures comprehensive updates across all integrated areas.

– Reduction in information requests: The tool reduces the need for manual intervention by technical staff, resulting in less work and increased efficiency.

– Scalability: Owners can extend the tool’s capabilities by including more information to widen their knowledge base and improving the integration of various processes and standards within an organisation.

AI tools, such as the FDM Navigator can be incorporated into engineering software to develop personalised design tools to cater to different agencies. These tools provide a foundational layer of quality control by highlighting items that do not fit current requirements and creating connections that link relevant technical criteria within engineering design software.

Nonetheless, while these tools accelerate data processing, the ultimate decision to apply technical criteria to certain engineering projects remains dependent on a professional’s judgement.