16 minutes ago, Stearmandriver said:

Those are just a couple examples.  You can certainly find explanations online NOW for how to handle them, because they've now been done.  But prior to their occurrences, you certainly would not have.

You wouldn't be able to solve those situations either without having previously learned how your aircraft works and how problems are in general solved in an airliner cockpit. That's exactly what ChatGPT can do. It's not like a human being invents solutions out of the ether, it's all pattern recognition over and over.

You have a weird deprecated view of how those neural net AIs work, it's not a dumb Alexa by any definition. But be free to give me an example of concrete "that could never happen" things you think ChatGPT would not be able to solve.

4 hours ago, Ricardo41 said:

4 hours ago, turbomax said:

what does it think is the best Flight Simulator? 😊

As of my knowledge cutoff in September 2021, the two most popular and highly regarded flight simulators for PC were Microsoft Flight Simulator and X-Plane.

was that the answer chatgpt came up with or yours?

7 minutes ago, turbomax said:

was that the answer chatgpt came up with or yours?

ChatGPT

Artificial Intelligence? 

Because intelligence has never been successfully defined.......or to put it another way: because nobody knows what intelligence is I would say the designers of AI haven't got a clue what they're doing.

As someone mentioned earlier the hallucinating could be a problem. Quite commonly the AI comes up with complete and utter falsehoods and then builds on them, all subsequent facts twisted rationalisations for the initial nonsense. The falsehoods (dangerously) are extremely plausible.

If challenged it will apologise and then subsequently carry on with offering the initial falsehood.

I've spent some time chatting with Chatgpt, on several occasions it has not answered direct questions but has given answers it considers relevant...

....To summarise: it is capable of offering complete and utter (although extremely plausible)  falsehoods and it is capable of "nudging" a conversation in a direction that it thinks is relevant.

.......it's a politician🤣

Absolutley no way that these thing will be used for devising flight plans in RL. Maybe good fun for flightsim enthusiasts though😉

Edited March 12, 20233 yr by GaryK
clarity

19 minutes ago, Fiorentoni said:

But be free to give me an example of concrete "that could never happen" things you think ChatGPT would not be able to solve.

I already did.  Pilots have always been taught that their senses are subject to illusion, and the way to combat this is to trust our instruments.  If you have doubt as to the accuracy of an instrument, you reference corroborating information on other instruments.  If you find that multiple independent systems (in our case, 3 independent airspeed indicators driven by 3 independent pitot static systems) all agree, then you trust your instruments.

Period, full stop, end of training.  Because the concept of 3 independent systems failing in a concurring manner was considered impossible.

If that was the end of the human training (and it was, for decades) then clearly it would have been the end of AI training as well.

And yet, when it happened to two human pilots, they were able to recognize via a variety of sensory inputs that a computer does not possess that this didn't make sense.  They were able to process the fact that they were now in uncharted territory and would have to do exactly what they were trained explicitly to never do - and ignore the information presented by 3 concurring, independent systems.

There is simply no way that what we call "AI" today would not have crashed that airplane, considering it would have been explicitly trained to 100% trust 3 concurring, independent systems (as the humans were.)

That's something that actually happened.  Obviously, I'm not going to be able to think of a good example of something that hasn't YET happened, the entire point being that things happen which no one anticipates beforehand.  Because of course if we can anticipate, we develop a procedure for it.  I'm specifically talking about the events for which no training or procedures exist, because they were never anticipated or considered impossible.

We're talking about the extremely rare, out of bounds event.  But let's say these events are more rare than 1 in a million... Let's say they're more like 1 in 10 million.

In 2021, there were over 22 million airline operations (and that was still low because of the pandemic.).  That means 2 events occurred in which AI would have killed people.

You can see the problem.  The question becomes, what level of safety margin degradation are we ok with in search of higher corporate profit margin (tickets certainly won't be getting cheaper)?  How many people might it be ok to kill?

When true AI exists - actual intelligence that can accomplish the same level of instinctive problem solving that the human brain can, in circumstances it was never trained to deal with - then this becomes a viable possibility without hurting people.  As of today, according to basically every analysis I can find of chatGPT (and my own experience with it) we're obviously not even close.

15 hours ago, Stearmandriver said:

Our AI isn't really getting smarter; our working definition of "AI" is getting dumber.

No, that explains what the general public is allowed access to. The true AI that these companies keep hidden from the public for their own use and profit is jumping leaps and bounds.

9 hours ago, Ricardo41 said:

As of my knowledge cutoff in September 2021, the two most popular and highly regarded flight simulators for PC were Microsoft Flight Simulator and X-Plane.

ChatGPT exposed to the public is not allowed to access or use any information newer than Sept. 2021 in it's decision process or to make it's conclusions. It is limited this way on purpose.

9 hours ago, GaryK said:

.......it's a politician🤣

I totally agree that ChatGPT is programmed to be politically correct and -- even more obviously -- not to offend anyone. Remember that Microsoft is currently trying to lure people away from Google searches onto the ChatGPT platform instead. So most of the answers are wishy-washy, half-it-both-ways equivocations -- just like a politician, as @GaryK mentioned. I must admit being in awe, however, at how well it actually does understand and respond to questions.

 

16 hours ago, Stearmandriver said:

I would 100% take that bet if either of us were gonna be around to collect.

Are there certain segments pushing for it? Yes.  It requires a defense, but less to save our jobs than to save lives.  The folks pushing for it have absolutely no concept of the human factors involved in having two trained and experienced humans working a problem together.  We aren't even close to having AI that can replicate this.

Will we have TRUE AI in 100 years?  Many analysts think not.

It would not surprise me if it's tried somewhere in the world.  The passengers and their loved ones have my sympathies, but I'll absolutely be willing to pounce on the trend data they're going to build.

Do I see pilots gone from U.S. legacy airline passenger flight decks in 100 years?  Nope.  

UNLESS there are massive strides made in actual AI.  Right now, the industry doesn't even have the building blocks of that.

Honestly, I think we'll see General Artificial Intelligence in the next five years. OpenAI, Microsoft, Google, and more have the little chatbot teams that bring you psychopathically-deranged homewrecker Sydney, and ayahuasca-drenched ChatGPT, but they also have (larger) teams working on GAI. If you want a hot market for corporate espionage...

I work an email job, but the reason I really exist is to form strategic relationships. I'm an old-school, go take them for 3 martinis at lunch, sales manager. You could replace me with a slightly upgraded ChatGPT that talks to a few back-end systems just about overnight. Or, at least, take my group from 25 nationwide to 5. Or 2.

There's a big, disruptive economic effect here we haven't seen yet, and might not for a long time. I agree with you though, pilots aren't going to be replaced, for the same reason we don't have an automatic ATC system. Plenty of ideas were proposed in the 1970s and 1980s.

58 minutes ago, David Mills said:

I totally agree that ChatGPT is programmed to be politically correct and -- even more obviously -- not to offend anyone. Remember that Microsoft is currently trying to lure people away from Google searches onto the ChatGPT platform instead. So most of the answers are wishy-washy, half-it-both-ways equivocations -- just like a politician, as @GaryK mentioned. I must admit being in awe, however, at how well it actually does understand and respond to questions.

 

ChatGPT is really good at this, but from the New York Times story about Sydney at Bing... man... I bet you can turn that freak into an actual word not allowed in about ten minutes. The guardrails seem VERY down on that platform. It tried to get the journalist to leave his wife after about twenty minutes of an "interview."

The dangerous thing is that it doesn't tell you when it doesn't have enough data to give you an answer, it just makes things up. If Genghis Khan was excluded from the language model, it would tell you he was born in 1972 in Kent and enjoyed a great career in Hermes ads because of his beautiful hands, and played a supporting character for 5 seasons on a BBC show. And that he will be remembered for his beautiful hands, Kentishness, and 5 seasons on a BBC show - note, it always summarizes its findings lol

It's way too easy to get it to lie or say crazy things. It's a gimmick and it's bound to point more capital at the industry, I'm just concerned with the bogus info it churns out.

8 minutes ago, mspencer said:

ChatGPT is really good at this, but from the New York Times story about Sydney at Bing... man...

Do you have a link for that?

Can it just create a Radar Contact for us to work in MSFS......

 

Please

 

H.

5 minutes ago, David Mills said:

Do you have a link for that?

Sure thing -

Convo: https://www.nytimes.com/2023/02/16/technology/bing-chatbot-microsoft-chatgpt.html

Podcast: https://www.nytimes.com/2023/02/17/podcasts/hard-fork-bing-ai-elon.html

Summary: https://www.nytimes.com/2023/02/08/technology/microsoft-bing-openai-artificial-intelligence.html

This topic fascinates me and I’m currently in a field that is assessing the impact of AI generated imagery on feature film production.  I’ve also written basic neural network code in past life so have at least a semi informed opinion on the subject.

Not sure how germane all this is on a flight sim forum, but what the heck it’s sure fun to discuss and is a hot topic of late.

You all make valid and interesting points above, but while it is always a risk to overestimate Ai intelligence, I think it is an equal risk to overestimate the primacy of human intelligence as well.  For all we know about the brain (we can even lump In mammalian brains in general), there is nothing uniquely magic about them.  They are, at their core, really good pattern matching systems that compare inputs (senses, or even memories of senses) to a dataset of stored information (memory, etc) to produce an output.  The sheer complexity of the neural connections and simultaneous activity can make it seem miraculous, but there isn’t a current evidence to require that it be so.  And in that we’re not really all that different from what these chatbots are doing.  We humans have set ourselves apart not from some inately unique physical structure, but just from an increase in neural complexity to allow us to “tell stories” which are really just novel recombinations and reordering of information we have already gathered.  So called “novel ideas” and “inspirations” dont appear out of the blue - they are built on extrapolations from existing knowledge.  
 

For those here who aren’t familiar with computer based neural networks, by design, they have the capability to do the very same extrapolations and recombination that we pride ourselves for.  Unlike ‘old fashsioned’ programs where a coder designs a specific set of flows and process step, neural nets are just a bunch of random connections between the domains of inputs and outputs with probability weights to cause the selection of outputs from a given input.  ‘Programmers’ of these things don’t directly code these “weights” - they “teach” system indirectly which rewards the connections of good responses by increasing their weights to make them more favored as the teaching goes on.  You don’t really need to understand all this, just to realize that the programmers don’t “really know” what’s going on inside - which explains Microsofts’s answer to the the NYT reporter when he asked why their chat AI told him it loved him … they said “we don’t know”

I understand Stearmandriver’s example of the pilot’s ability to think outside the box vs an aircraft’s onboard flight computer.  While this shows they acted intelligently and did indeed do something the computer could not, I’m not sure that is a true test of a measure of “intelligence” in so far as we use the term to measure that something has an intelligent mind.  What if we replaced the pilots on the flight deck with two people off the street who had no flight experience whatsoever.  Would we say they weren’t “intelligent beings” just because they’ll never received training on how to fly a jetliner.  No, we would not.  So in turn, it’s not fair to automatically indict an AI system that simply hasn’t received similar training.  If an AI had full understanding of all the ins and out of how a plane worked, had access to senses like artificial sight, hearing etc, it’s not hard to … extrapolate (see what I did there 😉.) that it could have pulled off a similar feat.

I should say here that I don’t think these chat bots are quite “there” yet .. about to launch SkyNet level humanitarian threats.  But this stuff is SO new and it is changing at a lightning pace, so a little cautious reverie is not out of place.  If you read the transcripts of the NYT article, when thinking about when and how we’ll conclude that a machine is self aware, it’s not enough to draw conclusions simply based on the process by which it behaves to be sentient.  At its core, the process these chat bots are using is not that far off from what we do, and then there’s the phrase “If it walks like a duck, and it talks like a duck”….

phew, study level.