In the future, commercial jets will be flown by a pilot and a dog. The pilot, to reassure the passangers that the aircraft is flown by someone capable. And the dog, once that they both are inside the cockipt, to bark and bite at the pilot if he even attempts to touch any of the controls.

As a rated pilot Musk has a pilot’s expectation of what an aviation autopilot does, but constant complaint in media and Internet message boards is that most reasonable people expect the function to completely automate the driver function. So they supposedly kill themselves and other people with that assumption and the word is that Elon lied and misled them just to get richer.

It doesn’t help that the “FSD” software package actually is an attempt, so far unrealized, at providing so-called Level 5 driver automation. Opinions on this run a full spectrum, from “it will save lives and in fact already has” to “unrealistic wishful thinking that will inevitably lead to wide scale carnage.” And of course Musk is a liar who pathologically lies so he can get richer.

So it should be no surprise that there is a general public array of misconceptions regarding aviation automation. After all we have drones and cruise missiles that do all sorts of things without a pilot on board so why not airliners?

There will be no Pilot 1. No pilot Zero. Ever.

Toy problem. Solved.

Now go fly a balloon. To Montana or something.

https://www.axios.com/2023/02/03/fully-autonomous-flight-planes

At best it substitutes the judgment of programmers and engineers, who almost by definition do not understand and cannot understand what they’re trying to automate because they’ve never done it, for the judgment of professionals who quite literally bet their lives on their knowledge, expertise, and a little bit of paranoia.

At worst, the automation requires the operation to be modified to adapt to the limitations of the automation. Loss of capacity, fluidity, reactivity, and increase in costs results. The labor savings is ephemeral; it simply replaces the on-board crew with a retinue of office staff, which have every possible financial and legal incentive to bloat their ranks and study everything to death.

Mr. Smith I sometimes disagree with, but on this subject he’s 100 percent on target.

But even this, once you look at it closely, seems desperately far away. It’s another one of those things where once you get into the details it’s a lot harder than it first looks even to get to where it could be done with absolutely _no_ pilot interaction with the aircraft or ground controllers between pushback and runway threshold even 95% of the time. And getting from there to 100%, where it can recover the system from something like a flat tire, is _much_ more work.

(Sorry for the separate post; I wanted to add this as a reply to my earlier post but the “Reply” button here hasn’t worked for me for quite a while.)

This has been yet another round of, “Wow, that looks impressive” improvements, and in certain ways it _is_ impressive. But that invariably triggers a fresh round of “We’re obviously just about to have AGI (Artificial General Intelligence)” claims and similar from people who haven’t been following the field and haven’t seen the same claims over and over (and over) again since the 1970s and even earlier, which continue not to pan out. And these people not only don’t know the answer to, “Well, what’s different _this_ time?”; they don’t even know to ask that question.

While ChatGPT has some serious implications for improving Siri’s “understanding” of simple commands, but it’s not only done nothing at all for having computers reason, but made things worse on that front. It produces answers that sound more authoritative but are actually just as often wrong as a random Internet search and can’t provide references that let you check whether it’s wrong or right; you can only decide if its answers are correct by already knowing the answer or doing the same research you’d do without ChatGPT. As Ian Bogost said, “Once that first blush fades, it becomes clear that ChatGPT doesn’t actually know anything—instead, it outputs compositions that simulate knowledge through persuasive structure.”

Like the one who flew up a mountain pass without any guarantee that his single-engine plane could reach the altitude at the top of the pass, and allowed himself to get into a situation where he wouldn’t have been able to turn around. (He managed to get through the pass, obviously, but it could have been another “pilot error” fatality.)

Or one who took off from SMO in “scattered” weather, and flew through a cloud into clear sky above the clouds. He called the tower with a “weather report” that “scattered” (less than half the sky covered with clouds) was now “broken” (more than half clouds). He was hoping to bamboozle them into helping him out, but the controller insisted he answer the question, “Where are you”. He had no idea.

I suppose an automated plane would handle that better — it would *always* be flying on instruments and not subject to the disorientation that occurs when you can’t see anything but white around you. But how good would it be at avoiding mountains? DOes it carry a complete contour map of the world?

I doubt that an automated pilot could do any better. And avoiding other aircraft in a sky full of even more aircraft than we have now (no pilots = cheaper flights = more craft)? Or dealing with the situation that Captain Sullenberger handled? I doubt it.