The AI pilot: can we remove Maverick from his cockpit?
Image credit: Landmark Media
Implementing autonomous technology into fighter jets will mean convincing human pilots to cede control.
Towards the end of the 20-minute drive from central Belgrade to Nikola Tesla airport, travellers with time to spare can explore the inside of a huge steel and glass doughnut. A curious piece of architecture, the Aeronautical Museum Belgrade appears to have split: as you pass by, you are separated from the museum by fences and then an aeroplane graveyard – a jumble of planes and helicopters that track the history of Serbia’s (and the former Yugoslavia’s) – air force all the way back to the Second World War. Perhaps the museum was once going to be bigger. Perhaps it’s a marketing ploy: ‘if this is what we leave outside, just imagine the stuff we keep out of the alternating summer rainstorms and new year’s blizzards.’
Inside the museum are further examples of aircraft from Serbia’s somewhat tumultuous history. But it’s the three shabbiest exhibits that are given pride of place, dangling on wires, or propped up at dramatic angles: trophies of the 1999 Nato bombing campaign. First is the tail of an F-16 Fighting Falcon. Second, the canopy of an F-117 Nighthawk (the ‘Stealth Fighter’). Finally, there hangs one of the first examples of what the public would come to know over the next 20-plus years as a ‘drone’ (an increasingly loaded term which militaries have since attempted to acronym-ise themselves away from: referring to them first as ‘UAVs’ or ‘unmanned aerial vehicles’, and now the du jour ‘RPA’ – ‘remotely piloted aircraft’). Dangling like a battered seagull pinata is one of the original General Atomics RQ-1 Predators, which until 2001 flew without weapons solely as information, surveillance and reconnaissance (ISR) platforms. It feels wrong to anthropomorphise weapons of war – but it really does look sorry for itself.
Despite spotty performance (a second Predator over Yugoslavia suffered a technical fault and was allegedly ploughed into the ground by its crew, hoping to smash it into small, unrecognisable pieces), once the Predator was armed with a pair of Hellfire air-to-ground missiles and began operations in Afghanistan post-9/11, its perceived operational successes – coupled with powerful surveillance applications, relative low cost and lack of physical danger to pilots – led to a battlefield revolution. In 2022, drones are omnipresent: the US and the UK operate the Predator’s larger and more technologically sophisticated successor, the MQ-9 Reaper, while other countries – including China, Russia and Iran – operate less sophisticated (though outwardly similar) systems. Even insurgent groups have been known to use jury-rigged, off-the-shelf civilian drones to carry and release hand grenades, or simply for surveillance.
Over a similar period, the West – and the US in particular – has poured billions into the development of next-generation fighter jets like the F-22 and the F-35. All this raises the question: as budgets increase and technologies mature, are we headed toward a world of pilotless air forces – and what are the implications if we are?
“It depends on your vision of future combat aircraft,” says Philip Sabin, professor of strategic studies at King’s College London’s Department of War Studies. “If you think that we’re now in a world where the human eye is, in effect, outdated and the engagements will be at such ranges, or in circumstances like night or bad weather, where the human eye just can’t see, but other things can... if you think that – and there are many that do – we might as well move to not having a pilot.
“On the other hand, there are many who think we need to have a situation where a pilot can look down and see properly. Radar can easily be ‘spoofed’ – the enemy makes you think what you’re hitting has the signature of a missile launcher when actually it’s a dummy or, even worse, an apartment block. Humans contribute this general sense through vision, but also a common sense: being able to operate in situations other than [very specific missions]. That flexibility is the key attribute that humans bring.”
The cameras on modern military drones make poor replacements for the human eye, Sabin argues. To perform effectively in a multirole fighter, like the F-35, even with all the technological bells and whistles, the ability of the eye to continuously, passively ‘scan’ for threats and then instantly focus to the exclusion of everything else is deceptively difficult to replicate.
“We’re familiar with drones as they exist now – the Predator and so on – being remotely piloted using communication links and a camera, which allows the crew to see what’s going on,” says Sabin. “The big difference between the camera and a human is that humans have this wonderful capacity to combine almost 180-degree vision across a whole hemisphere with very precise views of the centre. The trouble with cameras is you need to choose. It’s the same with a computer: you choose either to view a small part of the picture in high resolution or the whole of the picture in the moderate resolution. You can have one or the other, but you can’t have both at the same time. That’s what’s most lacking with cameras: cameras, compared to what the human eye can do, don’t give you peripheral vision at the same time as you’re seeing what’s directly in front of you.
“Now, let’s think about how that applies to drones. It’s not so much of a problem when you’re using drones to target enemies on the ground. In fact, the drone can zoom in better than the human eye unaided. But it’s like looking through a straw: you have a very, very small picture, and that’s fine if you can lock that picture to a particular target on the ground. But imagine trying to fly a plane like that. One reason that drones have [often] not been approved for flight in a normal airspace is the question: who’s looking out the front?”
Nevertheless, the dream of the all-seeing, all-capable multirole fighter drone isn’t completely dead; indeed, the US has technically flown drone ‘fighter jets’ for years. It’s less glamorous than it sounds: the QF-16 is an unmanned, remote-controlled variant of the F-16 Fighting Falcon (‘Q’ being US Department of Defence shorthand for unmanned aircraft. Like its forebear, the QF-4, it isn’t really a variant: the QF-16s are essentially surplus aircraft which, while perhaps a bit doddery (the first flight of the F-16 was in 1974), are still reliable enough that they can be used as target practice for newer US aircraft, air defence systems and so on. They’re not going to swoop into a warzone and save the day – but they’re proof that the idea of a pilotless fighter is technically possible, if not currently practical.
“If you don’t have a pilot in the plane, you can go down one of two routes, or a mixture of the two,” says Sabin. “One is you have the person flying the plane using sensory inputs provided from a whole series of systems. They’re looking at screens showing the top-down view, as well as information from the radar on the aircraft itself and so on. The pilots have these virtual inputs fed to them and they are sending commands to the plane. That’s the way that the Predators and so on operate. All you need is to make sure that you’ve got sufficiently strong and reliable data links and sufficient sensory inputs. One nightmare if you move to a completely unmanned air force – in terms of the people ‘in the cockpit’ – is that the enemy can jam them. If you’re dependent on the datalinks, and you lose those datalinks, you lose your air force. It fails catastrophically. It would not be easy, but it would not be impossible.”
Sabin adds that the alternative is that you put sufficient artificial intelligence (AI) in the aircraft that it makes its own decisions. “Now, that’s perfectly feasible. Modern airliners, effectively, fly their flight plan. The pilots, in a way, are a luxury: necessary for public relations, but unless something goes catastrophically wrong, these planes can take off, fly their routes and land completely autonomously. The limitation of that, again, is when you’re operating in an environment where the enemy is trying to interfere with you and you don’t want these systems to, say, kill lots of children. Can you depend on AI? It’s the same debate in military that we have for driverless cars. And if it’s hard enough to do in a civilian context, it’s even harder to do it in a military one.”
The driverless car comparison is particularly apt: you can test a driverless car until the wheels fall off to show that its onboard AI is orders of magnitude more capable than the average human driver and still end up in endless hypothetical what-if scenarios. For more than a decade, people have debated the merits of the driverless car, and the only two conclusions seem to be that there is no level of proficiency that will guarantee zero fatalities on the road, and that there is something peculiarly more galling about being deliberately killed by a machine than accidentally killed by a human. If the public is unwilling to hand over control of a Prius to a computer, they’re unlikely to do so with a missile.
The challenges for remotely piloted, or autonomous fighters, don’t stop there. Removing the pilot has two advantages: the safety of the pilot and the potentially increased manoeuvrability of the aircraft (drones don’t suffocate if they turn at high speed). But much of the appeal of Predator and the Reaper is that the per-unit cost of each is far lower than a manned fighter jet. The same wouldn’t be true of a remotely piloted or autonomous multirole aircraft.
“The practical reality is that they would be extremely expensive,” says Peter Lee, professor of Applied Ethics, and the director of Security and Risk Research and Innovation at the University of Portsmouth, who researches the ethical, operational, > < and other human aspects of UK drones and autonomous weapons. “For that aerial combat vehicle to be effective, it would have to survive in a modern, high-tech combat environment. It would need very good defensive aids: it would need to know when a missile was incoming, it would need chaff, and it would need flares. Like the [F-35] Lightning and other next-generation aircraft, it would need much flatter panels for a low radar signature. That is expensive technology. Then in order to put that in the sky, it would need basic things like detect and avoid systems, so it didn’t accidentally fly into a mountain.”
Such an aircraft, Lee believes, would represent a generational change – even if the technology were available, it would put the operators of the aircraft on the wrong end of the cost-benefit analysis that makes current drones so popular.
“I’ve always believed that humans aren’t the weakness in a fast jet,” Lee says. “[Yes], you could build an [pilotless] aircraft that could perform to much higher g-tolerances. But when you think about the complexity of creating all that versus the relatively straightforward challenge of: can you build a ground-to-air missile that will take it out of the sky? It’s much more expensive to build a very capable, high-end offensive weapon than it is to build a missile defence system to defend against such technology.”
But while replacing manned, multirole combat aircraft with similarly sized and capable drones may be science fiction (at least for the foreseeable future), the US and UK are exploring other roles for RPAs. The next major development in drone technology is likely to be in miniaturisation and ‘swarming’ technology, by which a large number of small drones are able to work with some degree of autonomy from human ground crews while feeding data to each other. Drone swarms – whether designed to drop small munitions or operate as ‘suicide’ drones that operate more like intelligent, networked missile barrages – highlight an awkward truth in the development of traditional air combat: the whole enterprise is simply becoming too darn expensive.
“So, at the moment with surface-to-air missiles, they’ve gotten really big and costly, but it’s still worth firing them against aircraft, [as those] cost even more,” says Sabin. “If you can split [one aircraft] into a variety of reusable platforms, you get to the point where, for the enemy, it’s costing more to fire one surface-to-air missile than it is for the enemy to replace the drone it’s targeting.
“The beauty of AI – and one way it may turn out better than remote piloting – is that [with the latter] you save the risk to the pilots, you save g-force limits and so on, but you don’t save human costs,” Sabin adds. “It actually costs more to have [aircraft] remotely piloted because you need lots of crew, different crews taking over from one another, and a big infrastructure to support them. The great thing about AI is that it costs a lot to design the [software], but once you’ve got it, it’s trivially expensive to have a hundred, a thousand or a million [aircraft].”
In November 2017, American F-22 Raptors – arguably the world’s most capable air-superiority fighter jets – carried out strikes on Taliban-controlled opium factories in Afghanistan. Like the F-35, the F-22 is considered a ‘fifth-generation’ fighter aircraft, a key delineation between fourth and fifth generation being advanced stealth capabilities. The Taliban had no air force, no credible air defence systems. The F-22 programme was cancelled by the US Senate in 2009, with a total of 187 aircraft produced. In the year it was cancelled, the cost per F-22 was estimated at $138m. Eight years later, they were being used to blow up sheds processing opium. Even before the cost overruns of the F-35, the F-22 was perhaps the best argument for small and specialised drones over small numbers of high-cost aircraft with no true enemies to fight.
“Even now with a defence spending boost because of the Ukraine crisis, we’re getting to a point where the traditional manned combat aircraft, which are so rarely used at anything like their full capacity, is becoming completely unaffordable. And therefore, I think we’ll have to go into something else, something more effective,” says Sabin.
However, swarms of smaller drones will not necessarily change the problem of flying blindly into the wrong target. More drones, more cameras, more autonomy, and better software all still bring us back to the paradox of the driverless car: even if mistakes are less likely, they will still, undoubtedly, occur.
“If you’re a state or a country that wants to operate within international law and international human rights law... you still need to be able to see what’s happening,” says Lee. “You need to understand where your targets are versus collateral damage... It’s hard [for machines] to interpret images – it’s much harder to interpret moving images. It will require massive amounts of computing power. AI doesn’t work like the human brain; the human brain has got amazing shortcuts. I’ve done aerobatics, so I know when I do certain things the picture I expect to see. And if I don’t see it, then I know I’ve done something wrong. So, another major limitation of an effective UAV is: how does the AI autonomously interpret the images that it is seeing to put together a three-dimensional picture? How can you teach AI to interpret moving images at high speed?
“I would describe myself as a pragmatist,” Lee concludes. “I am not a drone romantic; I see massive limitations. I could do a lot more damage in a lot less technologically advanced ways – with a lot less money – than try to create some of these [sci-fi] drones.”
Like Tom Cruise’s Maverick in ‘Top Gun’, pilots are, for now, safe. Between the deployment of the eye-wateringly expensive F-35 and the increasing number of combat drones in development around the world, there is still a place for a number of fighter pilots should the West ever be forced to return to days of air-to-air combat. That number may be shrinking as more countries expand their air forces with drones like the Reaper, its successors, or its substitutes – but the fastest and most reliable thing in the sky remains the reactions of a human pilot.
Sign up to the E&T News e-mail to get great stories like this delivered to your inbox every day.