The front panel of the Altair 8800 carried 36 indicator and status LEDs and 25 toggle switches, and in the basic configuration that was the whole interface. The machine that appeared on the cover of Popular Electronics in January 1975 sold as a kit for $439, and it had no built-in screen or video output. Anything else meant installing a serial interface card and connecting a terminal or teletype. The lights in the address and data rows were driven by the Intel 8080’s buses, and when the machine was halted, a lit LED meant a 1 and a dark one meant a 0.

The operator’s manual is careful about that condition. It warns that while a program is running, the LEDs may seem to give wrong indications. They were giving correct ones. A running processor changes its address lines far faster than an eye can follow, and an LED on a changing line glows at a brightness set by how often it is on. The panel showed an average of the bus, which was a true measurement of something nobody had asked about.

In May 1975 Dean McDaniel wrote a game for this panel. In Kill the Bit, a single lit bit circulates across the top eight address lights, and the player has to flip the switch beneath it at the right moment. A miss lights a second bit, which then also has to be cleared. The program has no output instruction for the display. It loads a byte from memory four times in a row, from an address whose high byte holds the current bit pattern, and fetching from that address puts the pattern on the address bus. The lights show the pattern because they show whatever is on the bus. The game’s graphics were a side effect of memory access.

Front panels did not last. Cheap terminals were arriving, and MITS soon sold a turnkey Altair that had no front panel. By the time the IBM PC shipped in 1981, a register display was no longer part of what a personal computer was expected to have. The lights moved out to the peripherals, and they kept the same property there: each one reported a signal.

The Hayes Smartmodem, introduced in 1981, had a row of them across the front: AA for auto answer, CD for carrier detect, OH for off hook, RD and SD for received and sent data, TR for terminal ready and MR for modem ready. Most of them correspond to lines on the RS-232 connector at the back. TR lit when the computer raised Data Terminal Ready on pin 20. CD lit when the modem heard a carrier on the phone line and asserted it to the computer on pin 8. RD and SD blinked with each character going in either direction. A user whose connection had dropped could see which side of the exchange had gone quiet, and whether the problem was the phone line, the modem or the computer, without running a diagnostic program.

Supra launched the SupraFAXModem 14400 in January 1992 at $399, less than most 9600 bit/s modems cost at the time. It replaced many of the individual lights with a two-digit green LED display that showed the connection speed along with codes such as RI for ring indicate and CD for carrier detect. The display cycled through its indications in turn, so carrier detect appeared only once every few seconds. None of the information had been lost. It now arrived when the firmware chose to show it.

That change spread. The indicator stopped being a lamp on a line and became a message composed by a program. The Wi-Fi icon, the sync spinner and the battery percentage are all reports written by software about hardware, and they are exactly as reliable as the software that writes them. For most indicators the difference goes unnoticed. For a camera light, the whole value of the light depends on nothing being able to intervene.

Apple’s earlier laptops were sold as having a hardware interlock between the iSight camera and its indicator. In December 2013 Stephen Checkoway and Matthew Brocker of Johns Hopkins published a paper showing that the camera could run with the light off. They did it by changing a register in the camera’s firmware so that the signal driving the LED no longer followed the camera out of standby. They had begun the study after the WebcamGate case, in which remote access software on school-issued laptops photographed students without their consent. The light had been wired to a signal, and the signal came from a program that could be replaced.

The current design goes further in the same direction. On the MacBook Neo, the camera indicator is drawn on the screen. Apple’s security documentation says that dedicated silicon in the A18 Pro, working with system software, is designed so that no untrusted software, including code running with root or kernel privileges, can turn on the camera without the on-screen indicator lighting up. The light is now a group of pixels, and keeping those pixels honest takes part of a system-on-chip.

Between the Altair’s LEDs and its bus there was a driver and a resistor, and nothing in that path could be told to report something else. Every indicator since has been as trustworthy as whatever sits between the lamp and the thing it reports. For nearly all of them that intermediary is a program, and no one minds. For the one light people rely on to tell them they are being watched, the industry has had to build a small piece of the 1975 front panel back into the processor.