JournAI: Steering the Future of Mobility with AI
Written by Zaahir Imam, Aarohi Kapadia, Dr. Sant (Winn) Leelamanthep, and Lisa Yu Li
The Honda Research Institute (HRI) Team is at the forefront of integrating advanced technologies into practical applications within the mobility sector. Our ongoing project, JournAI, represents a cornerstone of HRI’s commitment to transforming urban mobility through intelligent data use and AI integration.
Overview
In the rapidly evolving landscape of mobility, HRI recognizes the need for innovative approaches to data analysis and service optimization. JournAI, our latest project, leverages Large Language Models (LLMs) to develop and validate responsive, user-centered transportation solutions by analyzing mobility trends. This tool is designed to utilize mobility patterns from a given dataset (or multiple datasets) to generate and validate potential ideas through combining statistical models, classical machine learning (ML) and LLMs for researchers and product teams to explore, ensuring 99P Labs remains at the cutting edge of the mobility sector.
What JournAI Can Do
JournAI leverages data on people’s mobility patterns to generate product ideas, customer personas, and associated stories. Additionally, it features a conversational agent that enables users to interact with the data, facilitating the generation of valuable insights.
Enhancing Mobility Pattern Segmentation
JournAI relies on human-in-the-loop parameter identification to determine features of interest within mobility datasets. Additionally, it requires human input for dataset selection. The tool utilizes Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN), a clustering algorithm with the capability to automatically select hyperparameter epsilon. Epsilon is a critical parameter in density-based clustering algorithms, representing the maximum distance between points for them to be considered part of the same cluster. By automatically tuning epsilon, HDBSCAN optimizes the clustering process, enabling efficient analysis of large datasets on travel patterns to predict future mobility needs.
Following data clustering, the extracted insights are harnessed to prompt a language model via Chain-of-Thought with Self-Consistency (CoT-SC) method. This enables us to analyze travel patterns and their correlations with demographics, geography, and transportation modes. This information can be leveraged to identify potential pain points and forecast demand for additional services. This predictive capability allows HRI to tailor its services to meet anticipated changes in urban transportation dynamics, such as the increase in demand for electric vehicle (EV) charging stations or the optimization of route planning for shared mobility services. JournAI’s success stems from HRI specific context injection and multi-agent collaboration.
Improving Service Personalization
Through its advanced data analysis, JournAI can identify opportunities based on user preferences and mobility requirements. JournAI identifies potential pain points for different mobility personas and creates potential solutions for these pain points. Additional information like market size and potential impact is also generated. The workflow starts by clustering mobility data to reveal user patterns and defining specific groups. From these, JournAI derives stories to identify clear pain points, leading to the development of tailored solutions. If desired, each solution can be assessed through market sizing to ensure viability and significant market impact.
Generating Robust Responses
JournAI can surpass traditional LLM-based tools in output quality, operational efficiency, and privacy. It customizes outputs for mobility data analysis, ensuring more actionable insights. JournAI also streamlines interactions by using standardized prompts, improving efficiency and consistency across users. We apply prompt engineering best practices and techniques, including clarity, consistency, chain of thought reasoning with self consistent, and multi-agenting prompt chaining. Furthermore, it adheres to stricter data privacy standards, offering a secure alternative for handling sensitive mobility data, unlike some traditional LLM-based tools like ChatGPT, which do not necessarily adhere to the same privacy standards.
Limitations of LLMs for the Future of Mobility
Replace Human Decision-Making
While JournAI provides valuable insights, it does not replace the nuanced decision-making of human experts. The development and potential impacts of solutions often involve complex socio-economic factors that require human oversight. JournAI supports these decisions but cannot make them autonomously.
Limitations with AI Today
JournAI faces limitations across AI, product, and human factors. AI challenges include dependency on input data quality (“Garbage in, Garbage out”), ensuring effective prompts, and avoiding hallucinations — incorrect outputs. Product-wise, there is a risk of advancing poorly conceived ideas that may miss critical market pain points. Lastly, the success of a product like JournAI can be undermined by misuse, poor implementation, or inadequate change management, affecting adoption and effectiveness.
Operate with People in Mind
JournAI requires careful implementation and comprehensive training to be effective. It functions best when integrated into the workflows of researchers and product developers, complementing rather than replacing human expertise. We believe that a collaborative approach will enhance productivity and innovation, leading to more practical, user-centered mobility solutions. Proper training ensures that users can fully leverage JournAI’s potential as a tool, optimizing decision-making and fostering an environment of continuous improvement.
Looking Ahead
The insights from our semester-long development of the JournAI tool are incredibly promising. As we move forward, we see JournAI being integrated more deeply into HRI’s strategic planning for future mobility solutions, ensuring that services not only meet current demands but are also preemptively aligned with future trends.
There may be additional areas for solutions based on Machine Learning and Large Language Models beyond research and product idea generation and validation and HRI should dedicate resources appropriately. For example, within logistics and supply chain management, LLMs could predict disruptions and optimize operational routes to maintain productivity and improve efficiency. Also, in the near future, LLMs could be used to create detailed and realistic training modules or simulations, which could be incredibly useful for training staff or customers on future products.
We are excited about the potential of JournAI to transform mobility. By harnessing the power of LLMs and other AI technologies, HRI is driving towards a more connected, efficient, and sustainable future. Stay tuned for updates on JournAI and other innovative projects by following HRI Innovations here on Medium and the official 99P Labs LinkedIn page.
~ Team 99P Labs 2 from the Corporate Startup Lab at Carnegie Mellon University~
This project is part of 99P Labs ongoing commitment to enhancing mobility through innovative technology. It is not intended to be a standalone solution but a component of 99P Labs broader strategy to lead the future of transportation.
