With the news that an unbeatable AI ace has defeated a human expert in an air combat simulation - is it game over for fighter pilots? Not so fast, argues TIM ROBINSON.  

Retired Colonel Gene Lee, USAF in the air combat simulation. (University of Cincinnati) 

Breathless reporting recently that an artificial intelligence (AI) fighter pilot has conclusively defeated a human veteran pilot in 'dogfight' simulations may give the impression that a virtual Hans Joachim Marseille, R S Tuck (or Pete ‘Maverick’ Mitchell* (delete according to taste)) is ready to be patched in as a future upgrade to Taranis or nEUROn UCAVs.

Is this a foretaste of our doom at the hands of Cylon Red Barons? The air combat simulation, conducted by the University of Cincinnati saw a retired USAF Colonel with extensive AWACS, Aggressor and battlespace command experience beaten repeatedly by an ultra-aggressive AI program named ALPHA which used fuzzy logic to work out perfect tactics. However, it might be argued while the Colonel was undoubtedly an experienced tactician, was this a fair test? Would there be any difference if the AI was pitted against a current fighter pilot at the top of their game in the air superiority mission?

Secondly, the scenario was purely BVR (beyond visual range) – with it strictly limited to long-range missile fights, rather than close-in dogfights. Again, it is unknown what would happen once the fight entered the visual realm – where the ability of a human to visually assess energy states, predict moves and fool the enemy becomes more important than the long-range fight, which can be boiled down to radars, missile ranges and cranking. It might be argued, once adversaries enter the classic ‘knife fight’ in a phone box, the psychology of dominating and unsettling your opponent can compensate for less-than perfect moves.

Third, the scenario also takes place within a computer – with its own set of fixed rules and boundaries. But if the history of air warfare tells us anything, it is that creativity and improvisation rules and secures the kill. Once battle is joined, feints, deceptions and unexpected behaviour frequently turns the tables on the unwary. Even if programmed with ‘perfect tactics’ would an AI ace be able to ‘think out of the box’ when faced with an opponent who has not read the script? And as many video game players know, however insanely difficult an AI boss or opponent is, once a pattern is spotted or a loophole found, it can be ruthlessly exploited again and again. Unless the AI is able to 'learn' from its mistakes, would any fearsome robot Red Barons be limited to a very short period of time until countermeasures were developed? 

Fourth, It is also worth remembering that the veteran fighter pilots of the future, who may one day fly with or against UCAV AI, will be 'digital natives', and having been immersed in video games from an early age, will have a deep and intrinsic understanding of artificial intelligence, its limitations and how to beat it. Able to multitask, with lightning-fast reactions they will then be a new type of fighter pilot from those who have grown up straddling the pre and post digital age and whose brains are still adapting to this new world.   

That said, this research does point the way to the future – for robotic UCAV wingmen that can perhaps be unleashed on command like a pack of wolves. Manned/unmanned teaming is looking increasingly like the future for the fighter combat. Finally, the recent victory of an AI computer against a human Go player, ahead of predictions, shows that AI is now perhaps evolving faster than expected.

The white scarves and goggles brigade may be safe for now – but for how long?

Science fiction... for now?

19 July 2016