Israel's Physical AI Push Needs to Account for the Invisible Electromagnetic World
During the war, at times Israelis in central Israel opened navigation apps and saw themselves located in Lebanon or Iran. The streets had not moved; the invisible map above them had changed.
That incident was a strange disruption of ordinary life caused by the electromagnetic environment. Engineers and soldiers already understand that people live inside an electromagnetic world they rarely notice. Radio waves carry calls, fix positions and connect devices, yet most users see only signal bars.
As more decisions are delegated to autonomous cars, drones and robots, that invisible environment becomes part of what physical AI must learn to navigate.
Israel's new national AI plan lists physical AI as a strategic direction, defining it as machines that perceive, decide and act in the real world, across transport, health, space and agriculture. The plan proposes sandboxes and a national physical test range, where autonomous systems can be validated under controlled, instrumented, realistic conditions.
But the plan raises an engineering question: what counts as the physical world? Roads, weather, walls and people count. Radio waves should count too. Daniel Ferber argues the plan does not explicitly include the electromagnetic environment, but it should be part of the discussion. A wall is not only what a camera sees; it is also something radio signals must pass through, reflect off or route around.
Everyday experience reduces Wi-Fi or cellular networks to a signal-strength icon, yet the underlying physics is far less tidy. Radio behaves like impolite invisible light: some surfaces block it, some reflect it, and the path is rarely straight. A signal can arrive multiple times by different routes, separated by an instant.
Engineers have turned that chaos into an advantage. Modern wireless systems use multiple antennas to exploit multipath, converting complexity into extra capacity. If technology depends on the physical environment, testing it means understanding that environment. For autonomous machines, this determines what they can transmit, where they think they are, and which machines they can coordinate with. Unlike a wall, the electromagnetic environment can be deliberately altered by an adversary seeking to disable a machine.
Ukraine is an extreme case. RUSI researchers documented widespread jamming of navigation and command frequencies, and an old-fashioned response: small drones trailing a thin fiber-optic cable, trading wireless freedom for jam resistance. The same research notes that too many friendly radio-controlled drones in one airspace interfere with one another. Jamming is not only an enemy weapon; it is a property of congested spectrum.
The lesson is not that autonomy needs perfect connectivity. Often the opposite. With one-way Earth-Mars signal delay reaching about 20 minutes, NASA cannot drive a rover like a remote-controlled car; it must push more thinking onto the machine. Autonomy is partly built for moments when the network cannot be trusted. But the network remains part of the world a mission must understand. Autonomy does not replace communication; both are equally important. A rover must decide for itself, but a rover that can never send discoveries home loses much of the mission's meaning. Autonomy lets it act; communication gives those actions meaning beyond the machine.
The same logic holds closer to Earth. A single robot can navigate, decide and adapt alone. Put ten robots on one mission, and communication makes intelligent machines behave like one intelligent system: sharing what they see, dividing labor, warning one another and changing plans together. Intelligence lets people solve problems; communication lets them solve problems together.
Another part of AI already shows how to prepare for situations too rare or dangerous to meet first in the field. Waymo says its system has driven billions of miles in virtual worlds, and its simulation can generate rare scenarios on demand, from severe weather to unexpected road objects. NASA checks digital twins before sending commands across the long delay to Mars.
Radio deserves the same treatment. Tools for modeling signal propagation already exist: terrain, materials, antenna patterns, transmitter and receiver positions. More modeling is being tied to digital twins of real places. The next generation of digital twins should map not only where walls are, but what walls do to signals. The goal is not to replace field testing, but to fail safely in simulation thousands of times before entering the field.
For most people, the radio world becomes visible only when it fails: a dropped call, vanished Wi-Fi, or a navigation app suddenly insisting Tel Aviv is somewhere in Lebanon. Autonomous machines will not have that luxury. AI is being taught to see, reason and move through the world. The next step is to teach and test it in the invisible world as well.