"Working on a Long-Term AI Plan for the IDF"

The IDF's C4I Directorate established a dedicated Artificial Intelligence laboratory to develop this field of activity within the Israeli military. Special interview with the head of the laboratory team, Lt. Col. Nurit Cohen-Inger

Archive photo: IDF

Over the last five years, the IDF has been working on developments in the field of Artificial Intelligence (AI) for the various arms. The IDF Intelligence Directorate is regarded as the most advanced IDF element in this field, but it is not the only element involved in this activity. In the last two years, the IDF C4I Directorate, as the regulator of all IDF communication and cyber activities, has entered the field by establishing a dedicated AI laboratory with some twelve data scientists. Lt. Col. Nurit Cohen-Inger commands the laboratory team and serves as the IDF's Chief Data Officer (CDO).

"We have had some operational success stories," says Lt. Col. Cohen-Inger in a special interview with IsraelDefense. "This field of activity calls for commitment on the part of the IDF with regard to several aspects: establishment of excellence centers, revision of combat doctrines, investment in human capital, etc. The decision makers understand that and the matter is under deliberation in anticipation of the next long-term plan ('Gideon plus One')."

As far as Artificial Intelligence and algorithms are concerned, the IDF evidently requires data pools. These are massive databases into which data 'flows' from all of the IDF's sensors and data systems. The objective is to derive insights from this data. "The IDF has not yet 'boiled down' to a single data pool. There are critical compartmentalization issues that we must resolve," explains Cohen-Inger. "Yes, we do have different pools that share data. Intelligence data, operational data, and combat support data. These are different networks, but we can still derive common analytics products. Our job is to search for the contexts in all of those data – to ask the right questions."

The Objective: Operational Effectiveness

The Artificial Intelligence activity in the IDF is not restricted to a single course of action. As the military is a massive organization that must play in concert in order to win the war, algorithms will enter any field of activity that involves data. The range of relevant activities extends from the 2020 Administration in charge of the organizational ERP, including personnel aspects, through operational data, intelligence data (structured and unstructured) and decision making processes, all the way to the sensors and weapon systems. In the future, algorithms will provide insights for all of the IDF arms, from any activity that involves data out of which algorithms may derive insights.

"Our objective is to create a new military occupational specialty, 'Operational Analyst,' and staff such analysts in every brigade command center," explains Cohen-Inger. "We are working on a Long-Term Artificial Intelligence Plan for the IDF. We have to budget that plan and it will be included in the next long-term plan titled 'Gideon plus One.' The general staff realized that we must go for it. We are not yet certain about how to accomplish it – what is the path we should follow. To reach the warfighter, AI products should undergo a process through the systems of the IDF. We are looking for the right way to do it so that we can improve the effectiveness of the IDF."

One of the ideas currently under deliberation involves the establishment of a digital revision administration in the IDF that would coordinate the field of Artificial Intelligence, among other activities. A technological change on such a scale calls for a budget and for an operations tool such as a dedicated administration. "The planning requires answers to such questions as the required number of data scientists, their education, the computer infrastructure, etc. To determine what the IDF needs, we must engage in a long-term, strategic planning process. The General Staff is responsible for a part of the effort, while the C4I Directorate is responsible for the other parts with which it is familiar, like communication, spectrum, and cyber," says Cohen-Inger.

"I make a point of traveling around the IDF a lot and encouraging the discourse on Artificial Intelligence. The initiative includes lectures to senior IDF command courses for officers at the rank of Lieutenant Colonel and above. There are workshops on this subject, intended to identify the problems we may be able to solve by extracting information using algorithms. Some of the information is available within the IDF while the rest is available outside the military. We cooperate intensively with other security agencies. Everyone knows everyone else, we share information and look for ways to solve problems together."

Between Man and Algorithm

The Artificial Intelligence Laboratory at the IDF C4I Directorate operates under the Lotem Division, which is responsible for the build-up of the technological force. The researchers at the laboratory have master or doctoral degrees, and the laboratory maintains cooperative alliances with some of Israel's academic institutions for the benefit of developing algorithms for the IDF. Cohen-Inger explains that for every new project the IDF wants to implement, they look first for relevant projects civilian elements had accomplished – in academic circles, the web and commercial companies. If they spot interesting projects whose insights are available to the public, they may take these projects and use them in the IDF, subject to the appropriate modifications.

Within the IDF, the various arms cooperate fully. "All of the arms understand that they must address the field known as Artificial Intelligence jointly," says Cohen-Inger. "Every arm has some activity in this field, at various levels. The laboratory of the C4I Directorate is a multiple-arm operation. A part of the future structure of this activity will include a specialist in charge of implementing AI in large IDF staff elements. These specialists will be data scientists whose educational background is less extensive than that of the scientists serving at the laboratory.

"One of the fields we are working on at the laboratory is text analysis. Another activity involves the visual world, which includes video images. The sensors of the IDF produce massive amounts of data we must store and search to extract insights. Without algorithms, that would be impossible. One of the challenges is tagging the data – improving it for the algorithm. If you enter poor data, you will receive poor results."

Another aspect of the use of AI systems in the IDF involves the interface between the human element and the algorithm. At this point, the IDF avoids discussing autonomous algorithms. The parties involved refer to a machine designed to help the human element. IDF sources explain that incorporating a learning system in the decision-making process associated with the employment of fire is not a simple undertaking. To do it, they install the system and initially, the human element will make the decisions and the system will learn how the human element operates. Subsequently, the system will shift into 'recommendation' mode: it will present the optional decisions to the human element and the human element will correct it whenever the recommended decision was inappropriate. Finally, the system will be expected to make decisions more effectively, or at least just as effectively, as the human element. The Iron Dome system is one example of such a system.

"Assimilating such a system can take months or even more when fire employment decisions are concerned," explains Cohen-Inger. "There are several learning modes. There is real-time learning by the system. In this mode, the user provides real-time feedback to the system, and a decision-making mechanism corrects the system. Even if the algorithm works effectively in the laboratory, the challenging part is getting the algorithm into the operational cloud of the IDF, where real-life data flow."

Endpoint Processing

Along with extracting the data from the IDF's primary data pools, the AI activity is also relevant to the end devices – the sensors. "The sensor sends massive amounts of data, it never lies, and without algorithms, it will be nearly impossible to extract insights from those data," explains Cohen-Inger. "We will not be able to utilize non-standard data if a sensor provides such data. One of the challenges we face in this context is uniform cataloging of data. Standardization of the format in which the sensor delivers the data to the central data core. Some of our success stories in the past two years were associated, among other things, with this activity – extracting data from sensors."

IDF sources explain that the sensors' endpoint processing capabilities have evolved, so that some advanced sensors now come complete with dedicated GPUs (Graphic Processing Units) for analyzing visual data, while other sensors even feature factory-installed algorithms. The sources explain that the prevailing trend in the IDF is to operate through two channels. The first channel involves the acquisition of intelligent sensors capable of performing a part of the endpoint processing in real time. The second channel involves the development of algorithms that would analyze the data in the central data pools.


In the photo: Lt. Col. Nurit Cohen-Inger