A breakthrough in artificial intelligence is set to revolutionise aerial combat, specifically the mix of fifth generation fighters and long distance missiles which characterises modern day air warfare.

Christopher Jay | Sydney

The key step was the adoption of a variant of the programming approach known as ‘fuzzy logic’ to control the combat flight paths and missile launch patterns of advanced jet fighters, such as America’s F-22 and F-35 fighter aircraft and UCAVs (unmanned combat aerial vehicles). Hitherto, the typical assumption in artificial intelligence (AI) combat simulation software was that assessing attack angles and missile launch flight paths was a task for precision calculation.

The trouble is that so many of the complex intellectual assignments, which at first blush seem a natural for computers, entail rapid processing of thousands of variables. This is acceptable if you have a bit of leisure time to get through the massive computation load, say two or three seconds. But when a gaggle of missile-laden fifth generation jet fighters are storming in at each other at heights above 40,000 feet and a joint closing speed in excess of 1,500 mph (2,400 kph), once you reach the weapons engagement zone, out 20 miles or so, you may not have time to process thousands of relevant variables. Whether it’s going to be done by a human pilot, a computer or a mix of the two, it has to be the next thing to instantaneous.

“ALPHA can consider and coordinate the best tactical plan and precise responses within a dynamic environment over 250 times faster than its human opponents could blink.”

Make this a combat mix of manned fighters handling overall battle strategy and unmanned combat drones doing much of their own combat housekeeping autonomously, subject to top level human direction, and you run into the problem of dimensionality. Issues include a vast number of inputs and outputs, uncertainty, randomness, countering disruptive tactics by opponents, application of shoals of combat rules, all this to be compressed into processing milliseconds. Modern fighters do take sophisticated software aloft, but handling exceptionally complex decisions near-instantaneously has until now been the remit of a super computer.

That means more weight in a situation where every extra kilogram is a trade-off somewhere, plus expense in an arena where unit costs are already a concern.

For Nick Ernest, a US doctoral candidate in aerospace engineering at Ohio’s University of Cincinnati, the initial reality was that an expert human fighter pilot making intuitive guesses and instant estimates under conditions of uncertainty could outperform the pre-existing combat software.

During a three-year fellowship funded by the Dayton Area Graduate Institute and the US Air Force Research Laboratory, Ernest asked himself: can humans cope with high speed informational/decision-making overload? The answer is that we break the problem into a cascade of rapid-fire sub-decisions, use near-enough approximations and make informed guesses in uncertain situations. A conclusion of “enemy threat level high” galvanises action without requiring an exact probability to several decimal points.

Fuzzy logic

Computers using fuzzy logic employ linguistic programming, with ‘if this-then that’ decision-making. From the many sub-variants of fuzzy logic, Nick Ernest chose what is called a genetic fuzzy tree (GFT). The calculations and decision issues for each sub-part of the overall problem are assembled in a series of fuzzy inference systems (FIS), which are interconnected to allow interchange of results and feedback to improve overall performance.

Genetic algorithms can test out slight variations in software strings, with software sets pitted against each other to see which prevails. The programming language used is CYTHON, a superset of the PYTHON language deployed for computationally complex calculations.

In its early days, the Nick Ernest software was known as LETHA (Learning Enhanced Tactical Handling Algorithm). Now that a generalised account has been cleared for public airing, it has been renamed ALPHA. The software was modified and adapted for use first by competing against multiple versions of itself, then against the official USAF simulator training combat software and finally against experienced human pilots, starting with retired Colonel Gene Lee.

Lee, a former air combat instructor, imported his learned experience into the software. In October, 2015 he undertook a series of lengthy simulated aerial combat engagements against a considerably improved version of ALPHA, with disconcerting results.

Even when handicapped by deliberately inferior performance, in simulated air battle after air battle, the ALPHA software missiles shot the Colonel out of the sky. He described ALPHA as “the most aggressive, responsive, dynamic and credible AI seen-to-date”.

This result, aired by the University of Cincinnati media unit on June 27, 2016, caused a real stir in the US technical community. This included a considerable number of fighter pilots. It is said to be the first time that a set of AI simulator aerial combat software has unequivocally trounced highly trained and experienced professional fighter pilots, both in one-on-one confrontations and multi-aircraft engagements.

Cost effectiveness

A striking feature is that ALPHA, using human-style computing, sidesteps the computational overload, running on the equivalent of a $500 household computer. Yet ALPHA can consider and coordinate the best tactical plan and precise responses within a dynamic environment over 250 times faster than its human opponents could blink. Dr Ernest and some associates have set up a company called Psibernetix in Liberty Township just north of Cincinnati itself. The company “is ready to begin working with both local and international clients through a negotiable combination of on-site visits and tele-coordinating from our office on both short and long term contracts”.

Given the usual practical R&D application times, it will be a while before ALPHA is guiding combat unmanned aerial vehicles, or mixed human and drone formations over combat space. But if you’re looking ahead, to say the 2020s, you need to start investigating and thinking now how this development is going to shape the future of aerial defence.