TLDR: The Future Atmospheric Conditions Training System (FACTS) is a new platform that uses IoT sensors for real-time atmospheric data and Generative AI for personalized, place-based learning in climate resilience education. It creates localized challenges, provides adaptive feedback, and has been found effective and easy to use in initial evaluations. FACTS aims to improve climate awareness and adaptation skills by combining environmental data with AI-driven educational content, with potential applications beyond viticulture to other outdoor fields.
Climate change is a pressing global issue, leading to more frequent and intense extreme weather events like heatwaves, droughts, and storms. These changes significantly impact various societal sectors and industries, including agriculture. For instance, traditional wine-producing regions are struggling with rising temperatures and unpredictable rainfall, affecting both the quantity and quality of their produce. To address these challenges, climate resilience education is becoming increasingly vital, equipping individuals and industries with the knowledge and skills to adapt.
A novel platform called the Future Atmospheric Conditions Training System (FACTS) has been introduced to advance climate resilience education through personalized, place-based learning experiences. This innovative system combines real-time atmospheric data gathered by Internet of Things (IoT) sensors with carefully selected resources from a Knowledge Base to create dynamic, localized learning challenges. Learners’ responses are then analyzed by a Generative AI-powered server, which provides personalized feedback and adaptive support. Initial user evaluations have shown that participants found FACTS both easy to use and effective in building knowledge related to climate resilience. These findings highlight the significant potential of integrating IoT and Generative AI into learning technologies that adapt to atmospheric conditions, enhancing educational engagement and fostering climate awareness.
How FACTS Works
FACTS is designed as an adaptive educational technology that merges Generative AI and IoT elements to deliver learning activities tailored to real-time, local weather conditions. It acts as a learning companion, helping users adapt to changing climate conditions by drawing on research and expertise from regions already familiar with such challenges. The system uses a proof-of-concept scenario where users take on the role of a farmer transitioning to viticulture (wine grape cultivation) in a region not historically suited for it. FACTS helps them associate local weather conditions with relevant viticultural activities.
The system is comprised of three main components:
-
Client Application: This is a smartphone application that presents and manages interactive learning activities. Users can see a list of detected ‘atmostate’ stations nearby. By selecting a station, a challenge appears, summarizing local weather conditions (temperature, humidity, light intensity) and a list of possible actions. Users select actions they believe are relevant for the current conditions and submit their choices. The application then displays AI-generated feedback and detailed explanations for each selected and unselected action.
-
Atmostate Stations: These are small modules equipped with IoT sensors that capture real-time local weather conditions. They collect data and send it to the server via WiFi. These stations are strategically placed across an area to monitor diverse atmospheric conditions. Each station typically includes an ESP32 micro-controller, a DHT20 temperature and humidity sensor, and a BH1750 light intensity sensor.
-
Server Application: This component leverages Retrieval-Augmented Generation (RAG) based Generative AI, supported by a Knowledge Base (KB) for factual accuracy. Its primary functions are to generate KB-related resources for contextual educational challenges (e.g., a list of activity proposals with illustrative images) and to evaluate user responses to challenges based on local weather data and corresponding knowledge from the KB.
The adaptive learning process in FACTS involves an initialization phase, which occurs when the Knowledge Base is updated, and a general adaptation process that runs each time a user starts a learning task. During initialization, the KB (e.g., 20 scientific articles on viticulture in the Mediterranean region) is converted into vectorized values using an embedding model (mxbai-embed-large) and stored in a vector database (ChromaDB). This allows for efficient retrieval of relevant information. ChatGPT-4o is used to generate practical activities and DALL-E for illustrative images, which are then stored in an activity database.
In the general adaptation process, when a user begins an activity, the client collects real-time atmostate data and presents a challenge with randomly selected activities and images. The user’s selections, combined with the atmostate data, form a prompt used to retrieve relevant documents from the vector database. This retrieved information, along with the atmostate data and user selections, is then passed to a Small Language Model (Mistral-7B) to evaluate the user’s response and generate a debriefing, which is sent back to the client.
Also Read:
- Generative AI Workshops Drive Transformative Change in Business Operations
- MultiFluxAI: A New Approach to Intelligent Data Orchestration in AI Platforms
User Evaluation and Future Potential
A study involving twenty participants, none with prior viticulture knowledge, evaluated FACTS based on Perceived Usefulness and Perceived Ease-of-Use. Participants explored the system and completed at least two challenges before providing feedback. The results indicated a high level of satisfaction, with FACTS receiving a median score of 4.5 for perceived usefulness and 4.75 for perceived ease-of-use. Users found the system informative, interesting, and useful for learning new concepts, appreciating the combination of AI and atmospheric data as original and relevant for outdoor-related learning.
The positive reception confirms FACTS’s effectiveness for teaching topics like viticulture, agriculture, and climate change. A key strength of FACTS is its RAG-based architecture, which allows for high adaptability. By simply modifying the documents in the Knowledge Base, the system’s domain expertise and the associated weather-adaptive challenges can be shifted without altering the underlying code. While the initial KB focused on viticulture, feedback suggests potential use cases in other areas such as agriculture, forestry, beekeeping, and marine biology, all closely aligned with climate resilience education.
However, challenges remain, including defining what constitutes a good Knowledge Base, as it must include connections between desirable actions and specific atmospheric conditions. The risk of ‘hallucinations’ from Generative AI, though mitigated by RAG, is also a consideration. Furthermore, the system’s dependence on domain-specific expertise and appropriate language models in target languages highlights the need for more resources in underrepresented languages to avoid exacerbating the digital divide. Future work will involve larger-scale studies targeting both the general public and professionals in various sectors. For more details, you can refer to the original research paper: Harnessing IoT and Generative AI for Weather-Adaptive Learning in Climate Resilience Education.
