AI does not solve everything, but much, if done correctly
AI is the technology that is red-hot across most civilian technical fields, and it is here to stay. AI is an excellent tool in the toolbox for solving certain types of problems, though perhaps not the universal solution to all technical challenges that is sometimes marketed, not infrequently by those selling the service. All-encompassing AI works well when the result is permitted to be somewhat imprecise and when exactness, repeatability, and an understanding of the underlying reasoning are not primary requirements.
At Avioniq, we have for many years developed a variant of AI suited to the complex and sensitive military environment, where humans must remain "in the loop" for the foreseeable future and understand what is happening, as the stakes often involve human lives and difficult trade-offs between different options. But let us first clarify why AI can be problematic and why one cannot simply copy the all-encompassing AI solutions that exist in the civilian sector.
The type of AI that has attracted the most interest and admiration in recent times is the generative kind. ChatGPT and the image- and video-generating services that so seductively shower us with one remarkable solution after another are of this type. Generative AI, in brief, means that data is fed in, for example a question in text, an audio file, an image, or missile parameters, and a response is returned as data, such as text, audio, an image, or a series of manoeuvres in aerial combat.
A defining characteristic of generative AI is that the number of distinct combinations of input and output data is virtually infinite. When an image is generated from a text prompt, the result differs each time, even though the image is always a response to the text. It is therefore difficult to verify that the result is correct, or even good, since for these AI systems there is rarely a definitive answer as to what the correct response actually is.
To determine whether the answer delivered by generative AI is good, a domain expert must evaluate it, which is highly resource-intensive. In the case of ChatGPT, a large number of people, acting as domain experts for ordinary text, are used to assess the quality of responses. Specialised AI systems are also trained to evaluate the quality of other AI responses, based on the assessments provided by these human evaluators.
When it comes to AI for aerial combat, for example, the ambiguity around what constitutes a good response, and the fact that the answer may differ slightly each time, is a problem. Pilots are already a scarce resource, and the few who are available are not prioritised to serve as domain experts for AI training.
Companies that specialise in AI often sell it as though AI is the solution to the entire problem, hence the term all-encompassing. When AI generates answers to questions where the complete problem is contained within the question itself, as is the case with ChatGPT and image and audio generation, it works excellently to have AI experts solve the entire problem simply by being able to assess whether the answer is more or less correct in relation to the question. However, when the solution forms part of a larger context, with implicit information in the question that is not available as input data, where the boundaries of what AI should deliver are less clear, and where certain information and conditions reside in other systems, the challenge becomes considerably more complex.
Taken together, we have seen how AI can be a powerful tool, but that its application within military technology is not without significant challenges. All-encompassing AI solutions that function well in civilian contexts are often difficult to transfer directly to a military setting, where the requirements for precision and safety are considerably higher. But what does this mean in concrete terms for the development of AI in aerial combat? In the next instalment, we explore how these challenges can be addressed through specific examples and strategies for optimising the role of AI in the military arena.