Questions of whether you would trust artificial intelligence to handle anything from your accounts to your medicines could potentially worry some. After all, you are placing responsibility at the hands of a machine, and despite being in the 21st century where technology is everywhere, it still seems intimidating to some. So the question is: would you trust AI to fly an armed combat jet?
Psibernetix is the developers behind a new software called ALPHA that is used to fly unmanned jets in simulations, with the potential of assisting pilots in real-world missions in the future, NewScientist writes.
To the ones concerned out there, Psibernetix highlights that unlike many other AI-systems, ALPHA’s behavior is verified at each step, which means that it won’t act unpredictably. The system was initially developed to fly aircraft in a virtual air combat simulator, but has now transitioned into a “friendly co-pilot system that can help human pilots using the simulator”, the website writes.
ALPHA differs from other popular AI-systems, as it uses a “fuzzy logic” approach, called Genetic Fuzzy Tree-system, rather than the usual deep learning neural networks that mimics the human brain.
“Rather than emulating the biological structure of the brain, fuzzy logic emulates the thought process of a human,” Nick Ernest, CEO of Psibernetix told the website. Ernest explains how this makes it easier to work out each step the system took to produce an outcome.
So how does the system actually work?
By classifying data in terms of English-language concepts, i.e. a plane “moving fast” or being “very threatening”, ALPHA develops rules on how to respond most appropriately. This is where the system decides whether to fire a missile or take act based on a combination of how fast and threatening an opposing aircraft appears to be.
This is the key to ensuring ALPHA avoids computational overloads that can slow other fuzzy logic systems. “Without the GFT structure, ALPHA would not be able to run or train, even on the largest supercomputer in the world. With it, however, it can run on a Raspberry Pi and training can occur on a $500 desktop PC”, Ernest told the website.
ALPHA will also depend on instructions from a human commander on how to carry acts out, hence why it will never fire unauthorised. “We created the ability to have human overrides at every single level in ALPHA’s logic, and it is perfectly loyal to commands,” says Ernest.
What perhaps is deemed the most important aspect of ALPHA is the validation and verification, a process ensures that its software is reliable and appropriate to do the job it is assigned to, which is a vital factor when dealing with life and death-decisions.
This article was first published at: https://www.newscientist.com/article/2113709-ai-pilot-helps-us-air-force-with-tactics-in-simulated-operations/