Tiny IDF Unit Is Brains Behind Israeli Army Artificial Intelligence

Teams apps predict enemy rocket launches and analyze videos

The operational research unit of the Military Intelligence Unit — the software unit of the Israeli armys J6/C4i Directorates Lotem Unit — doesnt look like the kind of place where state-of-the art artificial intelligence is being put to work.

There are no espresso machines, brightly colored couches or views of Tel Aviv from the top floors of an office tower. The unit conducts its work in the backwater of Ramat Gan and has the look and feel of any other army office.

But the unit is engaged in the same kind of AI work that the worlds biggest tech companies, like Google, Facebook and Chinas Baidu are doing in a race to apply machine learning to such functions as self-driving cars, analysis of salespeoples telephone pitches and cybersecurity — or to fight Israels next war more intelligently.

Maj. Sefi Cohen, 34, is head of the unit, which in effect makes him the armys chief data officer. As he explains it, his units mission is to provide soldiers in the field data-based insights with the help of smart tools. We embed these capabilities in applications that help commanders in the field, he said.

One example is a system for predicting rocket launches from the Gaza Strip. After Operation Protective Edge we developed an app that learns from field sensors and other data we collected what are the most likely areas launchers will be set up and at what hours. That enables us to know in advance what will happen and what areas should be attacked in order to fight them more effectively, he explained.

In one project the unit built a system based on neural networks whose purpose is to extract from a video a suspicious object and describe it in writing. It wont replace human observers, but instead of looking at five cameras, it will be able to be responsible for dozens, said Cohen.

Cohen said the amount of data at his disposal from the army is endless, reaching into petabytes (one million gigabytes) in some areas. It also makes use of data from outside sources and the apps it develops use open-source code. We return to the world things that we use, Cohen says, Models that are operational obviously do not go out.

Cohen got his start in combat signals corps. Near the end of his compulsory service he completed a course in Lotem and spent another 10 years at its command and control systems unit. Ive always loved algorithms. I was already involved with them in high school and worked in the field. When I saw drafted I wanted to combine the technology with a combat, he recalls.

Cohen set up the unit he now leads with the help of local high-tech executives. I convinced my commanders that we could use machine learning in combat, and from there I started to bring in more and more people, he said. The unit now comprises about 20 officers, all of them in the career army and holding advanced degrees in computer science, focusing on AI.

The units only female member left recently, so for the moment its an all-male team. Cohen says most are graduates of the armys elite Talpiot program; the one who isnt has a masters from the Technion Israel Institute of Technology. Everyone whos here is the tops. I learn a lot from them, he said.