Technology Explained

It is fundamentally a symptom of high-technology warfare. To survive participants must possess a broad knowledge base of what each and every adversary system can do and what to do about it very quickly. Knowing and being able to exploit your own system is then equally important, high-quality decisions are a must.

The question however remains: can the human being cope with the volume of information under that level of stress? Or equally importantly is it appropriate to sacrifice aircrew with these skills in high attrition scenarios? The solution could very possibly lie in the use of Artificial Intelligence (AI) techniques.

Until 1981 AI was really a little known scientific discipline somewhere on the boundaries of mathematics, computer science and engineering. Its breadth of scope meant that aspects of it were researched in all of these areas and the US DoD was one of the main sources of research grants. In 1981 the Japanese startled Western technologists with the announcement of their ‘Fifth Generation Computer Systems Project’, a very ambitious national goal to develop a family of AI-based computers, termed Knowledge Information Processing Systems. These intelligent machines would reason like human experts in given problem areas and were seen as the means to Japanese technological dominance over the world computer market.

The flurry of activity in Western AI research funding which followed has since exceeded Japan’s commitment but certainly testifies to the respect the Japanese command in the commercial market. Whatever the outcome of the technology race the Japanese deserve full credit for identifying the strategic and commercial importance of AI. At this instant in time it appears that AI will find many commercial applications before penetrating into the more demanding military (/realtime) application area.

As a discipline AI essentially deals with the principles behind intelligent reasoning. Though many classify AI into computer science this is misleading as computers are really only the medium used to implement Al-based systems. One of the major applications of Al is naturally its use in designing software/ hardware to make computers smarter but equally importantly Al can serve to improve human decision making techniques in given areas. The breadth of AI as a discipline is considerable and many of the basic ideas used have been extracted from areas as diverse as electronic circuit design and mathematical set theory. The construction of practical software/hardware based AI systems can require considerable knowledge in all of the above scientific disciplines.

The shallowest level at which AI may be applied is in writing software with AI features and running it together with more conventional software on a conventional computer. A step further lies the creation of sophisticated AI-based software, eg expert systems, to be run on conventional machines. The final step lies in running AI-based software on dedicated Al architecturally specialised computers. These may differ from conventional number crunching machines considerably if high speed is required.

Most practical applications of AI today fall into the area of expert systems. An expert system is a program that reproduces a human reasoning process in solving a particular type or family of problems. Some of the earliest expert systems were developed for medical diagnosis and oilwell analysis, essentially areas where human expertise is very expensive. Conceptually expert systems are usually built around rule based systems. A rule based system is essentially a set of ‘IF (antecedent), THEN (consequent)’ rules equipped with a mechanism to compare real-world information with these rules and draw conclusions. A hypothetical application could be a target classifier that uses output data from a radar warning and electronic surveillance system (RHAW/ESM) and a radar with a non-cooperative target recognition (NCTR) mode. Obviously, it is difficult to use rules such as IF (aircraft has red stars), THEN (aircraft is Soviet) in practice as one need not get close enough to seeI For the above (please see example) however rules such as IF (aircraft has Skip-Spin radar), THEN (aircraft is Flagon) and IF (aircraft is Flagon), THEN (aircraft has 2 AAMs, 2 gun pods, max speed 2. 5M, thrust/weight cca 1:1, . . . ) are more realistic.

In use an expert system will take incoming data and compare it with antecedents (or consequents, depending on the type of system) rule by rule until it finds a match(es). The matching rule is then said to be ‘triggered’ and its consequent is ‘fired’ (invoked). In our instance our RHAW/ESM detected a Skip-Spin radar’s emissions from the direction of a radar contact, this triggered and fired the 1st rule which triggered and fired the 2nd rule. The result is advice to the user that the radar contact is a Flagon with possibly 2 AAMs, 2 gunpods, etc, etc.

Practical rule-based systems may use even thousands of such rules if dealing with complicated problems which to some degree explains the need for a lot of computing power. The whole collection of rules is referred to as an inference net as it allows the inference engine to infer facts and from these infer further facts.

Rule based systems have strengths and weaknesses. In the former area ease of expansion by simply adding in new rules is significant, particularly in military applications. The limitations of these systems are significant enough to have led to them being dubbed ‘idiot savants’, as they cannot learn, do not understand the reasoning behind their rules (though some can explain their train of reasoning) and view problems only from one perspective. Furthermore they are ill-suited to dealing with shades of grey in the certainty of the facts supplied to them. Techniques for fudging ‘certainty factors’ into rule-based systems exist but usually lack generality. An approach currently seen by many as the answer lies in the use of fuzzy logic which is a rigorous way of dealing with facts that may have shades of meaning or certainty. Handling natural human language and making decisions based on questionable sensor outputs are both areas which may benefit strongly. The ultimate answer may lie in expert systems which utilise each technique in areas best suited.

Expert systems can be a powerful tool in many problem areas and are already under consideration for areas such as ASW, Inverse Synthetic Aperture Radar target classification, electronic warfare threat classification and response selection (or ECCM) and onboard systems and battle damage management. DARPA’s Strategic Computing Initiative does in fact cover two programs with a heavy content in expert systems, battle management systems and pilot’s associate systems. An instance of the latter is a Lockheed project aimed at supporting the development of the Advanced Tactical Fighter. The Al Pilot’s Associate uses a distributed architecture with expert systems embedded in most of the critical subsystems, eg offensive, defensive, sensory, nav/com systems, these communicate with a ‘copilot’ expert system which in turn advises the pilot of the aircraft.

This particular strategy is quite meaningful given the current trend toward distributed processing in avionic systems (see part I, Jan 86 AA) as additional computing power locally can then be very usefully harnessed for low-level decision-making tasks.

Application of AI in general promises significant gains in weapon system capability. One of the goals of the DARPA initiative lies in developing autonomous battlefield vehicles essentially. eliminating the vulnerability of today’s remotely piloted/controlled vehicles; and jammable communication links. The Martin Marietta Autonomous Land Vehicle (ALV) falls into that category. The ALV is being developed as an attempt at a robot battlefield scout vehicle equipped with electronic and visual sensors. The biggest problem is seen at this stage according to Denelcor, codevelopers, in developing the high speed parallel processing computers necessary for the response times on a live battlefield. The ALV must be able to both navigate and recognise/avoid threats. Interestingly a legged vehicle is seen as easier to deal with than a wheeled one which must find it’s way around obstacles rather than over them.

Autonomous vehicles, whether land-based or airborne, offer advantages over manned scouts as they may be built smaller, with greater tolerance to battle damage and may manoeuvre beyond acceptable human physiological limits – all of these factors should provide much better survivability. Similarly the use of such machines ‘on “kamikaze” style or high loss rate missions’ will force defenders to expend much greater effort on covering critical targets. One could almost envisage retrofit conversions of ageing fighters into autonomous vehicles to get the very last ounce of life out of the platform.

The success of the Israelis using RPVs to attack heavily protected SAM sites provides a good proof of concept. Autonomous land vehicles may be extremely useful as harassment weapons to be left behind while retreating or airdropped behind enemy lines prior to an assault. Similarly providing weapons such as cruise missiles with this level of intelligence will make them extremely difficult to stop. Combined with intelligent submunitions (see TE Sept 84) this class of weapon promises to be a real force multiplier.

One of the interesting ideas that fall out of the use of AI machines in manned and autonomous platforms is the concept of ‘corporate memory’, where each and every AI machine is updated over time with the same expertise and experience. AI machines could be ‘debriefed’ after sorties and, after evaluation, their experience, if useful, distributed to other units to aid in effectiveness. This way the whole force structure can benefit from locally acquired combat experience.

AI technology is in its infancy at this stage. Given and the mind-boggling rate of development of computer technology it is unwise to make strong predictions as to where AI will be in a decade. It will almost certainly appear in mundane areas such as logistics, fault diagnosis and spare parts management, where it has found commercial use already. Applied in a limited fashion to tactical aircraft AI could for instance provide for much friendlier voice communication between pilots and their machines, similarly, it could be used as suggested earlier to cut down the crews’ information management workload. If used in aircraft built-in-test and diagnostic systems it could almost certainly improve availability on deployments as the aircraft could carry a lot of the needed engineering and support expertise with it.

An area where AI could provide massive gains is in aiding sensor-based target recognition. Identifying a target first means shooting first which often means winning the engagement. This may be the only way of effectively coping with a determined and numerically superior enemy operating weapon systems of comparable performance.

Electronic warfare and jamming equipment management is an area where AI is currently seen as offering massive gains in capability, providing a jamming system with the intelligence to reject false alarms and deliberate attempts to confuse by adversaries.

Preflight mission planning and tactics development is another potential application area, given the complexity of the air/ electronic battlefield of coming years.

Given existing trends in avionic development the integration of Al into systems may prove to be more of an evolutionary exercise rather than a revolutionary one. The problem at hand is more the issue of whether the human element will accept advice from a reasoning, talking machine without seeing it as a threat or adversary. A significant challenge lies in developing suitable forms of interface.

In the context of Australia’s defence forces AI would appear to be a very cost-effective force multiplier given the small size of our offensive forces. Whether an attempt to develop a local capability occurs is yet to be seen.

It is obvious that Artificial Intelligence is set to reshape many aspects of weapon system design, whether its greatest contribution will be at the system level or subsystem level is not yet apparent. Either way we can look forward to some very interesting developments. References: D

(1) Winston P. H. – ‘Artificial Intelligence’, 2nd Ed, 1984, Addison-Wesley.

(2) Wiseman C. H. – ‘International Countermeasures Handbook’, 10th Ed,