A Chinese AI system has defeated a top human pilot in a simulated dogfight, according to Chinese media. The AI was pitted against Fang Guoyu, a Group Leader in a PLA aviation brigade and a previous champion in such contests.
“At first, it was not difficult to win against the AI,” said Fang in a report in Global Times, a Chinese state newspaper. But as the exercise continued the AI learned from each encounter and steadily improved. By the end it was able to defeat Fang using tactics it had learned from him, coupled with inhuman speed and precision.
“The AI has shown adept flight control skills and errorless tactical decisions,” said brigade commander Du Jianfeng.
The Chinese exercise of setting human pilots against AI aims to improve both. The AI gives the pilots a new and challenging opponent which thinks out of the box and can come up with unexpected tactics, while each dogfight adds to the AIs experience and helps it improve.
The AI was developed by a number of unspecified research institutes working with the aviation brigade, according to the report.
The event echoes DARPA’s AlphaDogfight competition last year which featured human and AI pilots fighting it out in simulated F-16s. In the initial rounds, different AIs competed to find the best. In the final round, the winning AI, Falco from Heron Systems, took on the human champion, an unnamed U.S. Air Force pilot. The AI triumphed, scoring a perfect 5-0 win in a series of encounters.
AIs have significant advantages in this situation. One is that they are fearless and highly aggressive compared to human pilots; another term might be reckless. They can react faster than any human, and can track multiple aircraft in all directions, identifying the greatest threats and the best targets in a rapidly changing situation. They also have faster and more precise control: Falco was notably skilled at taking aim and unleashing a stream of simulated cannon fire at opponents who were still lining up their shot. Whether these advantages would carry over into a messy real-world environment is open to question – further planned exercises by DARPA, the USAF and others may help settle the matter.
DARPA’s ACES program, of which AlphaDogfight was part, plans to port dogfighting algorithms onto small drones and test various scenarios of one-on-one, one-versus-two, and two-versus-two encounters in the next year. At the same time they are also preparing for combat autonomy on a full-scale aircraft. This may utilize existing ‘dumb’ QF-16 target aircraft, the drone versions of F-16s used for air-to-air combat practice.
The contest for AI supremacy between the U.S. and China is attracting increasing attention, with the National Security Commission on AI (NSCAI) concluding in March that, “for the first time since World War II, America’s technological predominance is under threat.” China has created hundreds of new AI professorships and developed an efficient ecosystem for AI start-ups with tax breaks and lucrative government contracts on offer.
AI fighter pilots are just a tiny piece in the military balance, and not a meaningful indicator on their own. However, the fact that China chooses to publicize the latest development sends a message that they are hard on America’s heels, if not drawing ahead, in direct military applications of AI. If their AI can really learn skills that rapidly from contests with human pilots, then, like DeepMind’s AlphaGo, it may now be competing with versions of itself and developing tactics and levels of skill impossible for humans.
Meanwhile, in the larger evolutionary contest between humans and AIs, the machines have just taken another tiny step forward in chipping away our superiority. The new Top Gun movie out later this year may be nostalgic on more ways than one.
Source: ForbesRelated posts: