You’ve seen them all over the place and by now you know what an AI-generated graphic looks like. You know those cartoonish line drawings that attempt to depict real-world circumstances in an oddly appealing way that does draw your eye. What I don’t get is they seem to be designed to appeal to my generation, you know, the folks who watched the moon landing on television (I missed my best friend’s birthday party for it) and then proceeded to systematically and effectively destroy the world for the generations to follow.
From what little I understand about the online culture of the various letter-designated spawn of the last few cycles of human reproduction, I believe these cute cartoons are “cringe” to most of them and not to be taken seriously. That suits me fine because beyond their curb appeal lies the potential for serious misinformation of the kind some of us feared when we first encountered AI.
Earlier this week, the graphic that accompanies this blog, which was first circulated by a flight school, was making the rounds on the forums. I’m going to give the offending school a pass and not identify them. Assuming they’ve done the right thing and taken the image down we can just chalk it up to experience without vilifying anyone.
If you’ll notice, either the school or AI came up with a new acronym for spin recovery and it’s catchy. The problem is that it graphically depicts using the ailerons to “stop the rotation” even though the accompanying text correctly says to use opposite rudder. The aileron miscue is circled on the cartoon.
It’s doubtful that any lives will be lost because we’ve all done it and survived. On that very first spin (or maybe the first two or three) when that wing dropped your first impulse was to correct with the stick or yoke. When I did it, my instructor gave me a pretty hard slap on the hand while she simultaneously yelled “Right rudder!”
Trouble is, I’m told a lot of young folks learn differently these days and the pictures carry more weight than the words so I guess there’s some risk with this graphic, but that’s not the real problem here. As boring as flight instruction is, the paltry wages are earned when the student pulls mixture instead of carb heat or switches the fuel off instead of selecting a different tank. Making those mistakes is as much a part of the learning as correcting them is part of the teaching. A catchy looking cartoon can’t replace the actual experience.
Time marches on and it’s unrealistic to expect that AI won’t continue to become an integral part of much of what we do, except of course in AvBrief’s research and presentation of the news. That will always be done by humans here. AI has no place in writing the news, but I digress.
There may be some safe and practical uses for it in aviation, but there will be challenges.
It’s more than a little ironic that in some of those proposed applications the weak link will be the humans involved. The brainiacs at Georgia Tech have come up with a system that may use AI to allow drones to identify and then geolocate traffic in the pattern of non-towered airports based on the position and intention radio reports of pilots in that pattern. The goal is to give the drones a sense and avoid capability based on radio chatter that will allow them to mix safely with crewed aircraft at uncontrolled fields.
I’m not sure if any of the techies devising the system have ever flown at an uncontrolled airport on a sunny Sunday, but if they’ve come up with a way to wade through the stepped-on transmissions, the partial position reports, the silence of those on the wrong frequency, the pilots saying north when they mean south and those reversing the runway numbers, and turn that into a coherent picture of the traffic, then they may be on to something we can all use. But I have my doubts. I can’t see weekend warrior radio performance ever being used as the baseline information for any safety system, but it does show how ambitious and optimistic AI proponents are about the technology.
Presumably the Georgia Tech folks aren’t using the free version desktop AI that generated the flight school cartoon and that like anything else you get what you pay for. I also presume that Jeppesen Foreflight has created the systems and redundancy necessary to support the application of AI to mission critical flight information with its latest offering.
On the surface, the new ClearNOTAMs feature seems to be the ideal use for AI. The concept is simple. AI reads and then deciphers the increasingly convoluted and very difficult-to-read NOTAMs that cover everything from gopher holes on the runway edges to presidential TFRs and then gives you a fifth grade reading level summary of what’s in it. It also spreads that information wherever it might be useful on any of the myriad screens pilots consult for getting from here to there. There are also numerous customization options and configurations that would seem to make it easier for pilots to actually comply with that rule that says we must read and understand applicable NOTAMs before we fly.
Considering what’s at stake for Foreflight Jeppesen to take this leap, I would have to assume that it’s been tested every which way from Sunday to make sure its translation of NOTAM gobbledegook results in a faithful plain language rendition of the NOTAM. Assuming that’s the case, it might be one the most useful additions to flight planning since the moving map.
What would be even better is if the FAA just wrote NOTAMs in a readily understandable format. That would seem to be asking too much, though, so free enterprise has again filled a gap in a government-provided service that will cost us a little extra.
Learning to fly is a deeply personal experience to everyone who takes the plunge. AI, no matter how sophisticated, can never replace that feeling in the pit of your stomach when that wing drops and the world looks like a flushing toilet.
Nor can it replace the sting, both physical and mental, of a slap on the hand delivered as the bottom falls out of your world. It’s a lesson I never forgot.
“I would have to assume that it’s been tested every which way from Sunday to make sure its translation of NOTAM gobbledegook results in a faithful plain language rendition”.
That’s pretty big assumption especially since there are solid arguments that it is technically impossible to do that kind of testing of LLMs.
Personally I will never knowingly use what is commonly called AI for anything related to flying. The risk of gross errors, hallucinations and other LLM artifacts precludes its use in any life and safety realm.
I really hope that dude looks forward at some point….
I must be missing something with that acronym. S for “stop” the rotation, check. P for “power to idle”, check. I for “neutral stick”? N for “recover”? I don’t think that’s a helpful thing to memorize. I was taught PARE, (power to idle, ailerons neutral, rudder opposite, elevator to break the stall), with emphasis on reducing power first. I think this AI graphic and technique is inane AND misleading.
FAA employees get lots of training on the specific rules needed to turn English into NOTAM gobbledegook, so it can then be returned to English by AI…. What’s wrong with this picture?
And another thing….how does his 172 become a low-wing airplane by step 3?
I also watched the moon landing but find this cartoon far too busy to facilitate understanding of the spin recovery procedure. Confusing aileron for rudder is unforgivable, AI or not. Someone tried to re-invent the wheel with this cartoon and failed.
Regarding NOTAMS, I’m convinced that what I refer to as the flight planning cottage industry lobbies congress and the FAA to keep rendering NOTAMS indigestible.
And finally Russ, that “slap on the hand” was likely meant to be a hammer blow to the head. But like successfully coming out of a spin, the recovery can be exhilarating and deliver new horizons one never imagined were there. Looks to me like you’ve been more than successful in your personal deft spin recovery technique. Thank you for leaving AI out of it.
I step two and three, does that show the right wing has folded?
No, silly. The right wing in Step 3 is obscured by a cloud. Doesn’t everyone practice spin recovery just above a cloud deck?
The AI (anthropomorphism alert!) felt that it was necessary to depict two aircraft in different orientations for each step of the spin recovery process. If a few details (like wings) were omitted in the process, that was just to portray the sensation of panic in the student.
For me, the funniest cyber-brainfarts were placing the “Before Initiating Spin” checklist under “After Recovery”, the graphic for the “N(?) Recover” step showing the pilot looking to his right as he’s about to head-on collide with another plane, which is plainly visible because his yoke is now mounted at the top of his panel, so he has no readable instruments to distract him. But I wonder if he doesn’t have a bigger problem than entering a spin, given the long, serpentine trail of smoke he is leaving in the graphic.
Although not a safety-of-flight issue, that plane’s owner should sue his avionics shop for installing the radio antennas backward.
Sadly, this was a notable example of AI hallucinations only because it was So Bad. The evil of AI is that most of the AI-generated errors I’ve encountered were so subtle that most folks would never have noticed. And that’s the True Evil: we no longer have to worry about Big Brother. The manipulative tools and techniques that the advertising industry created and honed for a century, and well-documented in any number of dystopian sci-fi novels, are now available to a fifteen year old.
Long ago, long before I became a CFI-I, the old timers said just spin it down and hope there is a reasonable ceiling below the clouds in which you can recover from the spin. I was surprised in the “needle, ball, airspeed” era how few of the old timers actually had an instrument rating. I guess now I could, with appropriate weather, call approach and request a block altitude to practice going up, spinning down repeatedly. I guess might be kinda fun. Just make sure the ceiling is there.
SPIN is sure a catchy acronym. I’m just struggling to relate “N” to the idea of smoothly recovering. Maybe I’m missing an obvious connection?
Everyone needs to be cautious of AI for sure. A perfect example on a recent Social Media discussion on practicing a 180 power off approach led to a discussion about what to do with a constant speed prop in that situation and what happens to the prop when the engine quits or loses oil pressure. I was floored to find that many seemingly experienced pilots had no idea how the prop actually works or the dynamics of flat pitch vs. coarse pitch. To find a reference for my comments I searched for an answer from someone like Hartzell and was amazed at what AI generated. In addition to several functional discussions about how it really works, there were AI generated answers that were completely opposite to how it actually works and that is exactly what appeared to have been copied and pasted in some of the comments by these pilots.
A friend of mine with a 180 Comanche, he had just gotten a prop overhaul. While downwind, he felt a slight vibration and then although the engine was running, the airplane reacted and started downhill. First occurs denial. Full throttle and heading down still. Then react and head for the runway now. Turns out the prop hub’s restrictor devices that keep the prop from going to flat pitch had broken. And the prop, even with full throttle went flat. He said it was just like a drag chute or spoilers on the plane. He made it, barely, to the runway. Something we hope to never experience, but know it could.
“ Nor can it replace the sting, both physical and mental, of a slap on the hand delivered as the bottom falls out of your world. It’s a lesson I never forgot.”
As Roy T. Bennet would put it,
“Some things cannot be taught; they must be experienced.”
That AI image gets worse the more one looks at it…
Gary, you nailed it. A bad visual can plant the wrong control input faster than the text can correct it.
Artificial sweetener has never really been a true substitute for sugar…..