Photo: General Atomics
Artificial intelligence (AI) is slowly getting involved in anything and everything. In some fields, like say medicine, AI will likely help. When it comes to education, they’ll likely be used for cheating. In warfare, they’ll be outright dangerous, and eventually unbeatable.
Like it or not, it’s defense contractors that’ll probably bring AI to an uncomfortable-for-most level of development, and no matter how much press time things like ChatGPT get, it’s military AIs that will probably change the world.
Take the latest gimmick coming our way from General Atomics. Back in 2009, the company best known as GA-ASI introduced the MQ-20 Avenger (aka Predator C), a combat drone that is operationally ready but technically still in development, with new variants (like the ER) and new capabilities being announced all the time.
The most recent development of the Avenger front is that the drone was flown by multiple artificial intelligence systems in various scenarios, and it all seems to have worked to the company’s liking.
More to the point, GA-ASI conducted three missions with an equal number of AI pilots, using both real-world and simulated Avenger drones. The idea was not only for the AIs to fly these things, but to conduct “multi-objective collaborative combat missions.” The missions took place back in mid-December at the company’s facility in El Mirage, California, but we’re only now learning about them.
At the core of the exercise was something called Reinforcement Learning (RL) architecture. RL teaches machines to react and respond through trial and error while interacting with their environments. For the Avengers flights, GA-ASI created three so-called RL agents (the AIs) and their accompanying algorithms and placed them in “operationally relevant environments.”
The first was the single RL agent, and this one had to navigate the real drone while avoiding threats. The second, called multi-agent RL, operated both a live Avenger and a virtual one and coordinated efforts to chase down an unspecified target, while also avoiding various threats.
Last but not least, the hierarchical RL agent was tasked with using data from sensors to choose the best course of action in a given situation “based on its understanding of the world state.”
Now, all of the above sounds scary and all, but it’d probably be even more so if we had some actual details of what the missions were all about, how they unfolded, and perhaps even an insight into the hows and whys of the AIs’ decisions during the flights. We’re not given any of that, but the defense contractor says the tests “demonstrated the AI pilot’s ability to successfully process and act on live real-time information independently of a human operator to make mission-critical decisions at the speed of relevance.”
It’s unclear at the time of writing what will come of the lessons learned last December. General Atomics seems determined to advance the use of AI in its military machines and hopes for rapid integration of such autonomy capabilities.
Take the latest gimmick coming our way from General Atomics. Back in 2009, the company best known as GA-ASI introduced the MQ-20 Avenger (aka Predator C), a combat drone that is operationally ready but technically still in development, with new variants (like the ER) and new capabilities being announced all the time.
The most recent development of the Avenger front is that the drone was flown by multiple artificial intelligence systems in various scenarios, and it all seems to have worked to the company’s liking.
More to the point, GA-ASI conducted three missions with an equal number of AI pilots, using both real-world and simulated Avenger drones. The idea was not only for the AIs to fly these things, but to conduct “multi-objective collaborative combat missions.” The missions took place back in mid-December at the company’s facility in El Mirage, California, but we’re only now learning about them.
At the core of the exercise was something called Reinforcement Learning (RL) architecture. RL teaches machines to react and respond through trial and error while interacting with their environments. For the Avengers flights, GA-ASI created three so-called RL agents (the AIs) and their accompanying algorithms and placed them in “operationally relevant environments.”
The first was the single RL agent, and this one had to navigate the real drone while avoiding threats. The second, called multi-agent RL, operated both a live Avenger and a virtual one and coordinated efforts to chase down an unspecified target, while also avoiding various threats.
Last but not least, the hierarchical RL agent was tasked with using data from sensors to choose the best course of action in a given situation “based on its understanding of the world state.”
Now, all of the above sounds scary and all, but it’d probably be even more so if we had some actual details of what the missions were all about, how they unfolded, and perhaps even an insight into the hows and whys of the AIs’ decisions during the flights. We’re not given any of that, but the defense contractor says the tests “demonstrated the AI pilot’s ability to successfully process and act on live real-time information independently of a human operator to make mission-critical decisions at the speed of relevance.”
It’s unclear at the time of writing what will come of the lessons learned last December. General Atomics seems determined to advance the use of AI in its military machines and hopes for rapid integration of such autonomy capabilities.
Google News