Jean-François Bonnefon conducted a landmark study on people’s attitude towards self-driving cars. Bonnefon found that people loved the idea of utilitarian self-driving cars – cars that would kill their passenger to save innocent bystanders – but wouldn’t likely buy one.
Most of us hope Bonnefon’s research is moot. Self-driving cars should save tens of thousands of lives. Total automation will almost eliminate human error as a cause of accidental death on the roads.
But what about autonomous vehicles that are designed to cause death? If scientists can prevent loss of life with AI, can they exploit that same technology to deal death as efficiently as possible?
This combat AI flies like a (super)human
A doctoral graduate from the University of Cincinnati has developed an AI pilot capable of defeating a human pilot in a dogfight. ALPHA – developed by Nick Ernest in collaboration with the Air Force Research Laboratory – isn’t the first artificial intelligence capable of simulating combat flights. But, according to the flying whizz it’s just defeated, it is the best.
ALPHA is “the most aggressive, responsive, dynamic and credible AI” US Air Force Colonel Gene Lee has ever seen – and he should know. Lee has flown and supervised thousands of missions, and has sparred with AI pilots for years. ALPHA was the first to best him.
“I was surprised at how aware and reactive it was”, Lee told popsci.com. “It seemed to be aware of my intentions and reacting instantly to my changes in flight and my missile deployment. It knew how to defeat the shot I was taking. It moved instantly between defensive and offensive actions as needed.”
How ALPHA works
ALPHA works on an algorithms that form a “genetic fuzzy tree”, a subtler, more complex form of logic than traditional binary (1 and 0, yes/no) logic trees.
“Fuzzy” logic allows a computer to deal with imprecise data in a more human way than binary logic does. Binary logic can decide if something is true or not – but fuzzy logic can determine to what extent it’s true.
ALPHA uses this sort of logic to process hundreds of thousands of variables in rich detail, allowing a more dynamic understanding of a flight situation. That’s the “fuzzy” part of the system.
The genetic system is a little more intuitive. Early versions of ALPHA were tested against a pre-programmed version of the system, which acted as a norm. ALPHA flew thousands of simulated missions, and the code that proved most useful in those flights was combined – resulting in the best possible flying AI.
Previous fuzzy logic tree systems have been impossible to scale. Today’s computers can run such a system with four of five data inputs, but more than that and the computer’s processing power was insufficient. Ernest’s system breaks larger fuzzy problems into smaller ones. The system is so efficient that it can now run on a smartphone.
Artificial intelligence won’t make pilots obsolete
Ernest’s solution to this computing problem is important, but the science is unlikely to live long in the memory. That’s why his decision to couple his tech to such an evocative display – a life-or-death combat simulation – is so canny. People won’t remember what a “genetic fuzzy tree” is – but they’re sure to remember than an AI can outmanoeuvre a fighter pilot.
But the fact that an AI can beat a pilot – even a pilot as decorated as Lee – in a combat simulation shouldn’t come as a surprise. Most domestic flights are already highly automated, with pilots and co-pilots present to monitor, rather than lead operations.
It’s also worth remembering that we’re essentially talking about a computer game. This exercise was a simulation, taking place in virtual space. It’s no surprise that a computer program excelled in that environment, with fixed, predictable parameters, environment and objectives.
So, are even fighter pilots going to lose their jobs? Not any time soon. But ALPHA will likely be used – one day – to pilot assistant, “wingmen” aircraft that provide human pilots with real-time advice. It could also fly unmanned aerial vehicles with extreme precision. In aviation, as elsewhere, the trend is towards human-AI co-operation.
Category: Artificial Intelligence