From Skunk Works: A Personal Memoir of My Years at Lockheed (Ben Rich), regarding programmers in the early 1980s:
A year after the stealth fighter became operational, two computer wizards who worked in our threat analysis section came to me with a fascinating proposition: “Ben, why don’t we make the stealth fighter automated from takeoff to attack and return? We can plan the entire mission on computers, transfer it onto a cassette that the pilot loads into his onboard computers, that will route him to the target and back and leave all the driving to us.” To my amazement they actually developed this automated program in only 120 days and at a cost of only $2.5 million. It was so advanced over any other program that the Air Force bought it for use in all their attack airplanes. At the heart of the system were two powerful computers that detailed every aspect of a mission, upgraded with the latest satellite-acquired intelligence so that the plan routes a pilot around the most dangerous enemy radar and missile locations. When the cassette was loaded into the airplane’s system, it permitted “hands-off” flying through all turning points, altitude changes, and airspeed adjustments. Incredibly, the computer program actually turned the fighter at certain angles to maximize its stealthiness to the ground at dangerous moments during a mission, when it would be in range of enemy missiles, and got the pilot over his target after a thousand-mile trip with split-second precision.
It took us about two years to really perfect this system, aided by the nightly training flights at Tonopah. The computerized auto-system was so effective that on a typical training flight pilots were targeting particular apartments in a Cleveland high-rise or a boathouse at the edge of some remote Wisconsin lake and scoring perfect simulated strikes.
Compare the $1+ billion spent on healthcare.gov!
The early program would have been written in C or FORTRAN, in procedural style. Everything would have been hand coded. It would have run in 64kb of memory. A modern equivalent would have to be written in Swift or Go, in functional style. It would be based on neural networks & probability distributions rather than conditional statements & be non deterministic. It would infer what fighter planes are supposed to do based on statistical accumulations of all the data in the world. It would require 128GB of memory.
Since the computer is doing all the flying anyway, why not leave the pilot home so he can’t be shot down and held as a bargaining chip by the enemy?
Jack D, the pilot is insurance in case the mission goes awry somehow, you need someone to fly the $100,000,000 plane back to base along an improvised route, losing the plane is a serious cost compared to losing 1 pilot.
Coding up control software for fighter planes is exciting and fun, attracting talented people. Managing health insurance purchases is dull, attracting less talented people.
I’m sure you can find Healthcare.gov style fiascos created in the early ’80s, E.g. large state DMV records programs that failed, ATC “overhauls”, etc.
Maybe the programmers didn’t have 4-6 meetings per week to speak to managers and/or department heads to constantly keep everyone abreast of the software development and of course the inevitable feature creep that pointy-haired bosses would introduce, thus bogging down the whole process.
Imagine if Marissa Mayer managed a military software contractor…
Joseph,
an engineer for a company that makes strategic UAVs told me, after Iran, they just drop bombs on the ones that crash, no matter what the casualties will be.
I was told (by a military software contractor) that the one shining light in the F-35 program is its software, which is supposedly secure and plugged-in. The network is the fighter jet, or something.
Software for airplane guidance is much more mathematically formal and usually more coherent than software that connects many different databases relies on unreliable networks has to deal with possibility of human errors (allow corrections for any previous input) and must handle millions of asynchronous “sensors” simultaneous.
No good deed goes unpunished.
The F-117 Mission Planning System turned into a “common” planning system called AFMSS ($550M in 1998) which now has morphed in JMPS which has exceeded a billion dollars and of course was quite late.