At an AI Startup Weekend in 2018, our small team embarked on a mission to explore the potential of AI in redirecting tourists and fostering sustainable growth. We developed a test version of models that integrated various datasets, predicting over tourism in specific geographies and redirecting tourists to less crowded areas with personalised travel itineraries. Our efforts even uncovered another use-case - assisting major hotel chains in predicting future hotel locations based on data availability.

Data Collection For AI Training

To train our models effectively, we gathered a diverse range of data. We analyzed travel intention through flight bookings and historical tourist data, sourced from travel booking platforms. Hotel room availability data, including reservations, dates, and durations, was collected from hotel booking platforms and partnering hotels. We also considered factors such as water treatment capacity (obtained from local authorities), waste treatment capacities (from waste management agencies), transportation infrastructure, and population growth (since utilities are impacted by population increase).

Training The AI Model

For predicting mass tourism and redirecting tourists, we leveraged supervised learning models such as decision trees, random forests, and gradient boosting algorithms. These models excel in handling complex relationships between data features and outcomes. The models were trained to identify destinations at risk of mass tourism and recommend alternative nearby cities or regions that can accommodate more tourists.

Deployment Of The AI Model

To ensure user-friendly access to the model's outputs, we opted for deploying an API. This allowed travel platforms to integrate the recommendations seamlessly into their existing applications. The API-based solution facilitated scalability and streamlined the AI building process, focusing on the core technology rather than user-facing interactions.

Modern AI Advancements

As of May 2023, AI has taken significant strides. Modern AI models possess the capability to handle multiple task types, eliminating the need for separate models for each aspect of the application. For example, a single model can now predict water treatment capacities, travel bookings, and recommend itineraries based on personality clusters.

Cutting-edge models like Bard, Google's generative AI, can access real-time internet data, expanding their learning capacity. Their ability to tap into unlimited data also unlimits their capacities. This clearly poses a question on how to "validate" training data set and how to bring ethics into AI but am sure we'll master it. We're not yet at the stage of general AI, and generative AI is still a type of narrow AI but it's clearly broadening 👀.

Wanna chat about your AI Product needs? Ping me :)
Cheers,
Pat 🩰