AI-Powered Trip Planning: MIT-IBM Team Develops Personalized Travel Agent

Navigating the complexities of travel planning, from transportation and accommodations to meals and lodging, can be daunting. While large language models (LLMs) seem like a promising tool due to their ability to interact using natural language and provide commonsense reasoning, they often struggle with the logistical and mathematical intricacies required for effective trip planning. A recent study found that state-of-the-art LLMs provide viable solutions only 4 percent of the time, even with additional tools and APIs.

To address this challenge, a research team from MIT and the MIT-IBM Watson AI Lab has reframed the problem, combining LLMs with algorithms and a complete satisfiability solver. This innovative approach aims to create a user-friendly AI travel broker capable of developing realistic and logical travel plans.

“We believe a lot of these planning problems are naturally a combinatorial optimization problem,” says Chuchu Fan, associate professor in the MIT Department of Aeronautics and Astronautics (AeroAstro) and the Laboratory for Information and Decision Systems (LIDS), and a researcher in the MIT-IBM Watson AI Lab. Fan’s team focuses on developing safe and verifiable control systems for robotics, autonomous systems, controllers, and human-machine interactions.

The team’s framework leverages LLMs as translators, converting natural language descriptions of travel plans into problems that a solver can handle. Solvers are mathematical tools that rigorously check if criteria can be met, making them ideal companions for LLMs in complex planning scenarios. This hybrid technique allows users to plan trips without needing programming knowledge or extensive research into travel options.

The AI travel agent operates in four steps: parsing the user’s travel plan prompt, converting the steps into executable Python code, using APIs to collect data, and employing an SMT solver to execute the steps. If a solution is found, the LLM provides a coherent itinerary to the user. If constraints cannot be met, the framework identifies conflicting constraints and suggests alternative measures.

The researchers tested their method using GPT-4, Claude-3, and Mistral-Large against other baselines. The new technique generally achieved over a 90 percent pass rate, compared to 10 percent or lower for the baselines. The team also explored the addition of a JSON representation within the query step, which further improved the method’s ability to provide solutions with 84.4-98.9 percent pass rates.

The MIT-IBM team also tested their framework on other domains, such as block picking, task allocation, the traveling salesman problem, and warehouse optimization. The results demonstrated the generalizability and robustness of their approach.

“I think this is a very strong and innovative framework that can save a lot of time for humans, and also, it’s a very novel combination of the LLM and the solver,” says Hao.

The research, which was recently presented at the Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics, was funded in part by the Office of Naval Research and the MIT-IBM Watson AI Lab.