MIT has developed Air-Guardian, an AI system designed to work collaboratively with airplane pilots to enhance flight safety. Using Liquid Neural Networks (LNN), Air-Guardian can monitor pilot and AI attention and intervene when necessary to prevent potential incidents. This human-AI partnership is designed to enhance safety without relinquishing pilot control.
How Air-Guardian Enhances Flight Safety:
- Monitoring Human and AI Attention: Air-Guardian simultaneously monitors the human pilot's attention and the AI system's focus. It identifies discrepancies in their attention patterns.
- Intervention When Necessary: If the pilot misses a critical aspect of the flight, the AI takes control of that specific element while ensuring that the pilot maintains overall control.
Air-Guardian can be particularly valuable in scenarios where flight safety is crucial. For instance, it can intervene when an aircraft is flying close to the ground, where gravitational forces may lead to pilot incapacitation. Additionally, the AI can assist pilots when they are overwhelmed with excessive information, helping them identify crucial data.
The advantages of Liquid Neural Networks (LNN) include:
- Offer transparency in decision-making, making it possible for engineers to understand how the AI makes choices. This sets them apart from traditional deep learning models often considered "black boxes."
- LNNs can learn cause-and-effect relationships within data, reducing the risk of learning incorrect correlations. This makes them more robust in real-world applications.
- Require fewer computational units (neurons) compared to traditional deep learning networks. Their compact nature enables them to operate efficiently on edge devices with limited processing power.
The compactness of LNNs is particularly advantageous for edge computing scenarios where real-time decision-making is essential. Examples include self-driving cars, drones, robots, and aviation, where cloud-based models are not practical.
The insights gained from Air-Guardian's development can be applied to various scenarios where AI systems collaborate with humans. This includes tasks across applications, automated surgery, autonomous driving, and more. LNNs could also contribute to the development of autonomous agents, capable of making and explaining decisions, and hold applications in various domains.
MIT's development of LNNs is likened to the emergence of "transformers" in 2016, which led to the development of large language models. LNNs could lay the foundation for a new wave of AI systems, enabling powerful AI on edge devices such as smartphones and personal computers.
Air-Guardian and Liquid Neural Networks offer a promising outlook for AI collaboration with humans, enhancing safety and expanding AI applications across various fields. LNNs, with their explainable AI capabilities and compactness, represent a new foundation for AI systems.