However, Scaramuzza notes that there are significant challenges to deploying this technology in real-world scenarios such as warfare, as Swift relies on having reliable information on the drone's speed, location, and orientation in real time, and must learn the course ahead of time. Changes in conditions, such as wind direction, could render all of its learning useless. Despite these limitations, the success of Swift highlights the potential of AI in various applications, including the potential for AI-controlled drones in warfare.
Key takeaways:
- The Swift AI system, developed by Davide Scaramuzza and his team at the University of Zurich’s Robotics and Perception Group, has beaten human champions in drone racing. The AI system was able to learn real flying skills in a virtual world and apply them to real-world contests.
- The AI system's success in drone racing has implications for warfare, as AI-controlled FPV drones could potentially outperform human-guided weapons. However, the current system has limitations, such as the need for reliable real-time information and a known map of the course, which prevent it from being deployed in warfare scenarios.
- Scaramuzza believes it will take decades before such technology can be used in the military field and has called for a Geneva convention to stop killer robots. He warns of the potential dangers if AI systems like Swift are unleashed without proper regulation.
- Despite years of discussions at the United Nations, there has been little progress in drafting international law to regulate the use of AI in warfare. The current proposal calls for some sort of international law by 2026, but previous attempts have been watered down due to resistance from various states.