Enhancing General Aviation Safety: The Integration of Generative AI in Preflight Weather Planning

Abstract

Artificial Intelligence (AI) permeates daily life and industries through enhanced machine learning, natural language processing, and computer vision. Generative AI, a significant advancement, mimics human-like text and makes data-driven decisions, particularly in generative AI applications. Exploring novel AI applications is crucial, such as using generative AI to aid General Aviation (GA) pilots in preflight weather planning, with the aim to enhance pilots’ awareness and reduce weather-related accidents. However, the rapid evolution of generative AI raises many concerns like volatility, security risks, and decision-making biases. Preflight weather planning is vital in aviation, with weather-related incidents comprising a significant portion of accidents. Despite advancements, GA pilots’ interpretation of weather information remains subpar. This paper examines how generative AI may have potential to assist GA pilots to perform preflight weather planning, while also addressing risks and suggesting ethical research directions.

Introduction

In 2024, Artificial Intelligence (AI) has become integrated into daily life and a wide range of industries. Enhanced machine learning, natural language processing, and computer vision have increased AI system capabilities, which includes producing text similar to that produced by humans and making data-based decisions. The AI landscape is continuously evolving, particularly in generative AI. With the new capabilities, it is important to consider new uses of AI. One example is the use of generative AI to aid General Aviation (GA) pilots in preflight planning.

The purpose of the current paper is to examine how the evolving field of generative AI may aid in preflight planning for weather. To accomplish this, the paper first provides an overview of the current state of generative AI and the task of preflight planning for weather. Next, the paper describes the potential for generative AI applications to preflight planning for weather as well as the challenges inherent to applying generative AI to the preflight planning tasks. The paper concludes with research recommendations. This paper focuses on the most commonly used generative AI, ChatGPT, and the authors used the currently newest model from OpenAI.com, ChatGPT4, when interactions with a chatbot were necessary.

The State of Generative AI

Generative AI can be thought of as a machine-learning model that is trained to create, develop or build, rather than make a prediction about a specific dataset. A generative AI system is one that learns to generate more objects that look like the data it was trained on (Zewe, 2023). Generative AI is used by a wide array of professionals. This includes the tasks of preparing outlines for presentations, drafting report and letters, preparing exams, and many more. Even the medical and dental research industry utilizes this technology, assisting with medical diagnoses, treatment plans, and patient education (Bagde et al., 2023; Cahan & Treutlein, 2023). Generative AI significantly boosts human abilities, yet it still relies on human judgment for crucial decisions.

On one hand, generative AI has offered solutions that were previously unimaginable, from advanced voice assistants (e.g., Siri, Alexa, Google assistant, etc.), home automation (e.g., Amazon Echo [Alexa], Google Home, Apple HomePod, etc.) and enhanced entertainment experiences (i.e., streaming services using AI to personalize media recommendations), to personalized healthcare (for example, the health data collected by smartwatches and smartphones can be collated by the AI models and pertinent information can be provided to the user’s physician, allowing them to further personalize their own healthcare (Schork, 2019)). These developments have added convenience, efficiency, and accessibility to everyday life. On the other hand, the rapid pace of generative AI evolution brings uncertainties and concerns (Khogali & Mekid, 2023; Krügel et al., 2023; Körber et al., 2018; Sætra, 2023), which calls for a careful approach to maximize its advantages while also recognizing and addressing its limitations. This includes the following:

Volatility

One major issue is the unpredictability of AI’s advancement, as the ever-evolving nature of AI can lead to inconsistencies in user experiences and a steep learning curve for new technologies. An example of this is, from the author’s experience with ChatGPT, that at various times during a day, or different users, entering the same prompts, can receive different responses from the same chatbot model.

Security

Consumers also face privacy and security challenges, as AI systems often require access to vast amounts of personal data to function optimally, raising questions about data handling and consent. In its own words, ChatGPT states, “I cannot access real-time internet data. This limitation is in place primarily for privacy and security reasons. It helps improve consistency and predictability of the results, without the variability and other risks/errors that may exist with real-time online data. This design choice also helps to maintain user privacy and the integrity of the interaction, as I don’t access or retrieve personal or up-to-the-minute information from the web.” Thus, to achieve the full potential of real-time internet data processing, a certain level of security risk must be afforded, since the AI would have access to real-time information, which may contain sensitive information.

Ethics

Furthermore, ethical implications of AI exist. Biases inherent to the decision-making algorithms and the responsibility for AI-driven actions, add layers of complexity for consumers (Kalpokiene & Kalpokas, 2023; Krügel et al., 2023). The AI models are only as bias free as the data fed into them, and the process of how they were trained has the possibility of creating internal biases in the model (Bender et al., 2021). These and other ethical concerns exist.

Thus, while consumers benefit from the conveniences and advancements brought by AI, they must also navigate a landscape filled with rapid changes, ethical dilemmas, and potential risks.

The Importance of Prompt Engineering

A proposed key aspect of applying AI is prompt engineering. In the context of AI, specifically within the realm of large language models (i.e., large transformer-based language models, trained on web page text, like ChatGPT), prompt engineering is an emerging field that involves crafting inputs (i.e., “prompts”) to elicit desired outputs from AI systems. It is akin to a form of communication art, where the prompt engineer skillfully designs questions or statements to guide the AI toward generating the most relevant, correct, and useful responses. This technique is crucial because the effectiveness of an AI’s response heavily depends on how a query or command is structured (White et al., 2023).

In essence, prompt engineering is about understanding the nuances of the AI’s language processing capabilities and using them to achieve specific goals or extract specific information. It is particularly important because, as AI systems become more sophisticated and widely used in various industries, the ability to reliably extract valuable insights or responses becomes paramount. Effective prompt engineering can significantly enhance the utility of AI, making it a more powerful tool for businesses, researchers, educators, and consumers. It bridges the gap between human intent and AI interpretation, ensuring that the interaction with AI is more efficient, effective, and aligned with user expectations. As AI continues to evolve, the role of prompt engineering is becoming increasingly essential in harnessing the full potential of these technologies (White et al., 2023).

Interestingly, some believe that AI model advancement will make the process of engineering prompts unnecessary (Center for Humane Technology, 2023). Considering the rapid improvements and increased capabilities of AI models over the past year or two, it may be that prompt engineering is not a worthy investment of research and development resources. That is, AI may understand how we think and communicate much faster than we could possibly understand how AI “thinks” and communicates, making it an unnecessary endeavor to pursue prompt engineering as a field of study.

Nonetheless, prompt engineering research may still provide insights into how to specify to the AI how it may best aid the user. This seems particularly likely in the distinct and niche process of GA preflight weather planning.

GA Preflight Weather Planning Process

During preflight planning, the GA pilot conducts an in-depth analysis of diverse information, including geographical maps, aeronautical charts, aircraft specifications, and current and forecasted weather. Preflight planning for weather (the focus of the current paper) involves gathering and reviewing weather information to consider the implications for a planned flight. After considering the weather information, the pilot makes a go/no-go decision. They also prepare alternate plans, such as selecting alternate airports or routes, in case the weather changes unfavorably during the flight.

Pilots need to access a variety of weather products (i.e., static and dynamic displays of both observed and forecast conditions) for the departure, enroute, and destination locations. This includes METARs; Meteorological Aerodrome Reports), TAFs; (Terminal Aerodrome Forecasts), winds aloft, satellite imagery, radar, a variety of products that display the movement of weather fronts and the presence of high or low-pressure systems (e.g., clear dry skies, or cloudy and wet skies, respectively), pilot reports (PIREPs), and other products. Pilots also review NOTAMs (Notices to Air Missions) for other weather-related information that could affect the flight, such as temporary flight restrictions due to severe weather.

Reviewing the weather data is crucial for flight planning as it affects the chosen altitude, fuel consumption, and estimated time of arrival. Pilots also assess the risk of encountering adverse weather conditions such as icing, turbulence, and thunderstorms. adding to the situational awareness.

Throughout this process, it is essential for pilots to remain updated on weather changes, continuously monitor conditions, and be prepared to adjust their plans, as necessary. Effective weather preflight planning is a dynamic process that requires vigilance, knowledge, and sound judgment to ensure a safe flight.

Challenges to Aviators

Reviewing the weather information and understanding the implications for their planned flight is a complex cognitive task with a series of steps.

The first step is to access (e.g., online) the most important weather products. While this may seem straightforward, pilots’ behavior indicates otherwise. That is, while numerous types of weather products are available, pilots tend to view only the most basic products (Blickensderfer et al., 2023). This may indicate that pilots have difficult remembering which products are important to look at and where to find them.

Adding to the difficulty of the task, pilots need to review weather products for both current observation and forecast conditions, Additionally, pilots need to review the applicable products for their departure, en-route, and arrival locations.

Once the information is accessed, the next step is to interpret the information, which requires skill and experience, as the weather products are complex and sometimes present conflicting information (Blickensderfer et al., 2021) and have low usability (McSorley et al., 2019).

Ultimately, this means the pilot has to mentally integrate various types of information for different locations and corresponding to various times of day/en route time. It is a large amount of information for a pilot to integrate and remember in order to develop a robust situation awareness.

Other challenges exist as well. One is the dynamic nature of weather conditions. Weather conditions can change rapidly, and forecasts may not always accurately predict real-time changes. This uncertainty adds to the complexity of planning flights over longer distances or in regions prone to volatile weather. The need for contingency planning also poses a challenge. Pilots must not only plan the primary route but also consider alternative routes and destinations. This requires additional time and resources to ensure safety in case of unexpected weather changes during the flight.

Lastly, as part of their pre-flight decisions, pilots must balance the desire to complete a flight with the need for safety. For example, they may feel pressure of personal or business obligations which, in turn, can mean that pilots must weigh the risks of flying in less-than-ideal weather conditions against the importance of the flight.

These challenges require pilots to be well-trained, continuously update their meteorological knowledge, and maintain a cautious and flexible approach to flight planning, always prioritizing safety over convenience.

Implications for Using Generative AI in GA Weather

A number of opportunities may be possible for AI to perform a role during the preflight planning phase, in terms of weather assessment and decision-making.

Gather Information

Future generative AI models could be used to search and gather weather resources. This would assist the pilot by providing resources that the pilots may not have thought of previously. Additionally, prompting the pilot with alternative resources may function as a catalyst for the pilot to check additional resources they may have forgotten to utilize (Bin-Nashwan et al., 2023).

Customized Weather Information

Additionally, generative AI could customize weather information based on specific flight plans. By integrating historical, predictive, and real-time weather data, AI systems could provide scenario-specific forecasts, helping pilots anticipate and prepare for likely weather conditions, in a manner that is not possible without the aid of a local weather expert. An example of this could be a pilot wanting to take a flight from Orlando (MCO) to Atlanta (ATL) on a specific day, at a specific time, generative AI could show departure, en-route, and destination weather information. The predictive nature of generative AI, coupled with its ability to learn and adapt from past weather events, would allow pilots to receive the most current and relevant weather information (OpenAI, 2023).

Efficiency

Generative AI technology also has the potential to streamline the communication of weather-related NOTAMs, ensuring that pilots are promptly and accurately informed about critical updates that could impact their flight plans (OpenAI, 2023).

In essence, generative AI could empower aviators in general aviation with enhanced situational awareness, tailored weather forecasting, and real-time updates, leading to safer, and more efficient flight operations (OpenAI, 2023).

Challenges to Generative AI Assistance

While AI offers significant advantages in assisting aviators with preflight weather planning, it also faces several challenges. Once again, a major issue is the reliability and accuracy of the data fed into AI systems, as well as the data available to it (Zewe, 2023). Weather forecasting is inherently complex and unpredictable, and AI models are only as good as the data they analyze, and what information it was trained with (Azamfirei et al., 2023; Bender et al., 2021). Inaccurate or incomplete data can be given (Borji, 2023) which can lead to misleading predictions, potentially compromising flight safety.

Usability of Generative AI Output

Another challenge, but also an opportunity to aid pilots, is the AI’s ability to interpret and present nuanced meteorological information in a manner that is easily understood by pilots. Since pilots do not have extensive meteorological training, the challenge is for the information reported by the generative AI to be displayed in a way that a GA pilot could effectively process and understand the information. Without usable displays, the utility of collating and reporting this data would be lost—the pilots would not understand what the data meant for their potential flight.

Over-reliance

Similar to other forms of automation, a risk of over-reliance exists (Körber et al., 2018; Khogali & Mekid, 2023). That is, as pilots develop trust in AI tools and enjoy the reduction in their own mental workload, a risk exists that they might become less vigilant in cross-checking information or less skilled in self-interpreting traditional weather information. If over-reliance occurs, pilots may accept the AI’s information (which may not be completely accurate) without cross-checking. In turn, pilots may be misinformed about the weather, leading to flights into unexpected weather hazards. Thus, ensuring a balance between AI assistance and pilot expertise is crucial for safe and effective preflight weather planning.

Source Transparency

When examining generative AI models, many pull information from the internet with varying degrees of accuracy (Borji, 2023). Most do not disclose the exact sources for the generated information. For aviation weather tasks, the lack of source reporting is a challenge. The FAA requires aviators to use weather services that meet FAA/NWS (National Weather Service) standards. Without controlling for reliable resources, and citing those resources, there is no way to determine the accuracy or reliability of the information generated by the AI (Azamfirei et al., 2023), aside from pilots manually cross-referencing the information with known reliable resources (which would defeat the purpose of using generative AI to collect and display data).

Future Generative AI Recommendations

A variety of needs exist to increase AI capabilities for assisting GA pilots conduct preflight planning for weather.

Recommendation 1: Include Real-time Data

The widely available generative AI models such as Chat GPT do not have access to real-time information, and thus cannot develop a comprehensive weather report for preflight planning. While the current restrictions set on chatbots like ChatGPT are most likely necessary, for generative AI to be useful to aviators in weather planning, it must have access to real-time data.

Recommendation 2: Develop a Model Specifically for Preflight Planning for Weather

Currently, companies such as Openai.com allow the user community to develop specified versions of their model for particular tasks. Referred to as “GPTs,” GPTs are a tool for anyone to create a tailored version of ChatGPT that would helping their daily life, at specific tasks, at work, or at home and then share that creation with others (OpenAI, 2023). Importantly, each GPT can be kept internal to the creators, not allowing the general public to access it or the information retained therein.

A possible solution to the complex task of preflight weather planning could be to train a generative AI model specifically on available information from trusted weather sources and aviation information sites, such as the FAAs website, the Aviation Weather Center’s website, and other FAA approved sources. This would allow the generative AI model to have access to real-time information on weather and important aviator notices, leading to a more accurate, and more reliable application for preflight planning.

This model could be restricted to only accessing these vital resources, eliminating the possibility that the model would pull inaccurate or unreliable information. Also, this diminishes online security issues, as it would only have access to the pertinent weather and aviation data, and it could even have an addition layer of security, potentially requiring login credentials to access this database.

Recommendation 3: Increase Source Transparency

The proposed model for preflight planning for weather must have source transparency. Source transparency is also important for public access generative AI’s. In both cases, source transparency will assist users to more easily determine the credibility of the information given to them by the model. In turn, source transparency will help prevent propagation of misinformation.

Recommendation 4: Conduct Research on Prompts

By researching how prompts affect the AI, we can learn how to effectively communicate user needs. This may be helpful in considering human-AI interactions overall, but also in specific use cases, like preflight planning for weather. A thorough understanding of prompts, should include how the AI models interpret the user inputs. Other research on prompt engineering will reveal if prompt engineering is necessary.

Recommendation 5: Conduct Research on Pilot Skill Decay due to Over-Reliance

Similar to other forms of automation, a risk of over-reliance exists. More specifically, as pilots develop trust in AI tools, a risk exists that they might become less skilled in self-interpreting traditional weather forecasting methods (Körber et al., 2018; Khogali & Mekid, 2023), and this possible skill decay should be researched.

Conclusion

In summary, generative AI holds significant potential for enhancing preflight weather planning in General Aviation by helping users aggregate, visualize, and interpret weather information more effectively. However, to maximize the accuracy and reliability of AI-generated data, we recommend specific enhancements to current models’ capabilities. Ongoing research is essential to further refine these tools, ensuring they provide precise and actionable insights without undermining the pilot’s decision-making authority, or causing over-reliance.

While AI should not supplant pilot judgment, it can augment pilots’ understanding of both current and forecasted weather conditions, thereby mitigating confusion and reducing weather-related incidents in the GA domain. Through continued development and careful implementation, generative AI has the potential to become a valuable asset in promoting safer and more informed flight operations.