Recently, many articles and references have been published about the role of Artificial Intelligence (AI) in military systems, most of which are contrary to the truth. In this article, within the limitations of classification, we present some of the current realities.
AI appears in two main categories in military applications.
- Utilities in technical and tactical work processes (OCR, search, automatic translation, information extraction and structuring, speaker recognition, speech recognition and speech-to-text conversion (STT), area scanning and change detection, big data processing, etc.).
- It is a utility to improve the decision-making process, but cannot be used to make actual decisions.
Naturally, much of the interest has been focused on the use of artificial intelligence in decision-making processes and day-to-day operations within defense agencies, which has seen great advances in recent years thanks to the HPC (High Power Computing) revolution and the ability to process “big data” quickly and efficiently.
The use of AI in IDF systems involves a very careful operational approach, and maintaining this approach is crucial.
Current orders and procedures clearly state that artificial intelligence applications will not be used in life-or-death incidents without human involvement.
In other words, attack decisions are not made solely by AI-based algorithms, and likely won't be so in the foreseeable future. Decisions will be made solely by competent humans, including with close legal support, just as is the case with AI applications in healthcare that affect human lives.
Decision makers will no doubt use AI to extract the best information and recommendations, including presenting past precedents, to make the best informed decisions, but as appealing as it may sound, AI systems are not designed to replace lawyers or legal experts.
For example, in the Israel Defense Forces' “Lavender” system of “Iron Sword” fame, the AI layer adds some consideration to intelligence filters but does not make decisions.
Beyond scientific research, the use of AI tools has led to the development of a wide range of evaluation methods to ensure the accuracy of products against defined performance thresholds and operationalize the systems. This process includes testing the stability of algorithms over time and under different conditions by creating random samples and developing a number of regulatory mechanisms and approvals before AI systems are authorized to provide recommendations for human use.
The verification and decision leading to an attack will in any case involve at least two separate indications from independent sources and will be carried out exclusively by the human element.
AI products that are made accessible to operational consumers will also go through an “explainability” process to verify their appropriateness for use.
AI products can also be classified according to their resolution usage: the first layer deals with “single pieces of information” (which are very voluminous, and this is one of the main enablers of AI) and extracts most of the information components from them for future searches. This usage has given rise to different specializations of information items (audio, images, text, etc.).
The second layer deals with “intelligence entities” that combine lots of information, allowing AI algorithms to generate comprehensive insights from them.
The third (futuristic) layer deals with a wide range of events linking together a vast number of “intelligence entities” and building the ability to predict events that different entities may take part in in the future.
The three-tier model is well suited to defense information systems, but it is also appropriate for many civilian systems, such as medical, scientific, transportation, and agricultural.
AI implementation
Implementing AI tools in military systems, especially in environments that affect human lives, is extremely complex and challenging, which is why only the human element is declared as the “content expert” and nothing can be decided by AI software algorithms alone.
In a decision-making system, it is the “content experts” who actually make the decisions, and it takes a long time before the recommendations of the AI system gain enough trust to become part of the “content experts’” decision-making process.
Training operational users to use machine learning tools requires special knowledge and a lot of time, but in any case, it is important to reiterate that decisions made by AI systems will not be made in offensive military systems that could affect human lives. We are very far from this situation.
In civilian systems, AI systems find applications making life-or-death decisions, even if only indirectly. This is perhaps one of the main reasons why, while other kinds of AI systems are becoming ubiquitous, civilian autonomous systems such as self-driving cars are penetrating our lives at a much slower pace than expected.
Brigadier General Jacob Nagel (Ret.) is a Senior Fellow at FDD and a professor at the Technion. He previously served as National Security Advisor to Prime Minister Netanyahu and Chairman (Acting) of the National Security Council. Advisor Jacob Bardugo is a media expert covering technology and has been a senior commentator on radio, television and newspapers for many years.
