Hey there,

I've been spending a lot of time with ChatGPT-4 recently and thought about its potential use as an airtraffic controller in MSFS. I gave it a context as follows:
 

Quote

Context: You are role playing an air traffic controller for a class delta airfield in the USA which has a single runway. The runway runs east to west with the numbering Runway 27 and Runway 9. The winds at 6 knots at 360 degrees. The runway is 12000ft in length. Further context: I am role playing in a pilot who is approaching your airfield from a distance of 20 miles. I will call you up and request the various permissions required in order to land my aircraft, and for each of my calls, you will respond as an ATC would in the real world.

And here is how it played out:

 

Quote

bigbluss: Burbank Tower, Cessna N16LB 10 miles to the East, request full stop with bravo.

ChatGPT: Cessna N16LB, Burbank Tower, good day. Make straight-in approach for Runway 9, report 3 miles final. Wind 360 at 6 knots, altimeter 30.10.

bigbluss: Straight in for zero nine, report 3 mile final 16LB

ChatGPT: Cessna 16LB, roger.

bigbluss: Burbank Tower, would prefer to land on 27, N16LB

ChatGPT: Cessna N16LB, roger. Enter left downwind for Runway 27, report midfield. Wind 360 at 6 knots, altimeter 30.10.

bigbluss: Left downwind for 27, report midfield 16LB

ChatGPT: Cessna 16LB, roger.

bigbluss: Burbank tower, midfield for 27, n16lb

ChatGPT: Cessna N16LB, roger. Cleared to land Runway 27, wind 360 at 6 knots.

bigbluss: Cleared to land 27, 16lb

ChatGPT: Cessna 16LB, roger. Have a safe landing.

bigbluss: Burbank tower, N16LB clear of 27 at alpha, request taxi to parking

ChatGPT: Cessna N16LB, roger. Taxi to parking via Alpha, monitor ground on 121.7. Have a good day!

So it made a few errors there - for example I was approaching from the East yet it told me to make straigt in for 09, and it also said 'Have a safe landing' which I don't think I've ever heard in the real world. But you can see how with a little more contextual development this thing could produce a highly convincing and interactive ATC for MSFS. Given the API could take voice commands and geo coordinates as an input, I don't think a fully fledged dynamic ATC is unrealistic. Technically you could also set up the system to impersonate a number of AI pilots to speak on behalf of AI traffic - each with different voices. 

There are of course significant uses for LLMs out there, and this one is rather mundane by comparison - but I thought it was quite cool and thought I'd shared.
 

Edited April 20, 20233 yr by Bigbluss

So I thought I'd push it a little further - the following context was given:

 

Quote

Expanding on the above conversation, I want you roleplay the above tower, with five aircraft of various types approaching for landing. Roleplay all the pilots and the air traffic controller, sequence the aircraft for landing, and get them to their gates.

yielding the following:

 

Quote

ChatGPT: Burbank Tower, Cessna N16LB 10 miles to the East, request full stop with information Bravo.

ATC: Cessna N16LB, Burbank Tower, good day. Make straight-in approach for Runway 9, report 3 miles final. Wind 360 at 6 knots, altimeter 30.10.

ChatGPT: Piper N23PA, 12 miles west, request landing with information Bravo.

ATC: Piper N23PA, Burbank Tower, enter right downwind for Runway 9, report midfield. Wind 360 at 6 knots, altimeter 30.10.

ChatGPT: Learjet N31LJ, 15 miles southeast, request landing with information Bravo.

ATC: Learjet N31LJ, Burbank Tower, expect Runway 27, maintain 3000 feet, report 10 miles out.

ChatGPT: King Air N42KA, 8 miles north, request landing with information Bravo.

ATC: King Air N42KA, Burbank Tower, enter left base for Runway 27, report 3 miles final. Wind 360 at 6 knots, altimeter 30.10.

ChatGPT: Cirrus N55CS, 11 miles south, request landing with information Bravo.

ATC: Cirrus N55CS, Burbank Tower, enter left downwind for Runway 9, report midfield. Wind 360 at 6 knots, altimeter 30.10.

ChatGPT: Cessna N16LB 3 miles final Runway 9.

ATC: Cessna N16LB, cleared to land Runway 9. Wind 360 at 6 knots.

ChatGPT: Piper N23PA, midfield right downwind for Runway 9.

ATC: Piper N23PA, follow the Cessna on final, number two, cleared to land Runway 9.

ChatGPT: Learjet N31LJ, 10 miles out for Runway 27.

ATC: Learjet N31LJ, make straight-in approach for Runway 27, report 3 miles final.

ChatGPT: King Air N42KA, 3 miles final for Runway 27.

ATC: King Air N42KA, number one, cleared to land Runway 27. Wind 360 at 6 knots.

ChatGPT: Cirrus N55CS, midfield left downwind for Runway 9.

ATC: Cirrus N55CS, follow the Piper on right downwind, number three, cleared to land Runway 9.

ChatGPT: Cessna N16LB clear of Runway 9 at taxiway Alpha, request taxi to parking.

ATC: Cessna N16LB, taxi to parking via Alpha, monitor ground on 121.7.

ChatGPT: Piper N23PA clear of Runway 9 at taxiway Bravo, request taxi to parking.

ATC: Piper N23PA, taxi to parking via Bravo, monitor ground on 121.7.

ChatGPT: Learjet N31LJ, 3 miles final for Runway 27.

ATC: Learjet N31LJ, cleared to land Runway 27. Wind 360 at 6 knots.

ChatGPT: King Air N42KA clear of Runway 27 at taxiway Charlie, request taxi to FBO.

ATC: King Air N42KA, taxi to FBO via Charlie, monitor ground on 121.7.

ChatGPT: Cirrus N55CS clear of Runway 9 at taxiway Delta, request taxi to parking

Pretty impressive for a first draft. 

And finally, I promise no more - adjusted the responses to add an element of uncertainty. It produced the following:

 

Quote

ChatGPT: Cessna N16LB going around, missed approach.

ATC: Cessna N16LB, roger, climb to 2000 feet, make left traffic for Runway 9, report midfield.

ChatGPT: King Air N42KA declaring emergency, engine failure.

ATC: King Air N42KA, roger, you have priority, continue approach to Runway 27, emergency services alerted.

ChatGPT: Piper N23PA clear of Runway 9 at taxiway Bravo, request taxi to parking.

ATC: Piper N23PA, taxi to parking via Bravo, monitor ground on 121.7.

ChatGPT: Learjet N31LJ, 3 miles final for Runway 27.

ATC: Learjet N31LJ, expect a slight delay due to emergency traffic, continue approach, expect go-around.

ChatGPT: Cessna N16LB, midfield left traffic for Runway 9.

ATC: Cessna N16LB, number one,

 

Very clever. And here I am trying to use it to suggest Mothers Day gifts for my wife....

22 minutes ago, jrw4 said:

Very clever. And here I am trying to use it to suggest Mothers Day gifts for my wife....

If you're serious, try giving it a context! The more detail the better. For example:

 

"Context: you are a loving husband of x years old with a wife aged x. Your wife enjoys (insert hobbies) and you want to treat her to an incredibly thoughtful mother days present with a budget of a to b. Write a list of your top five ideas, explain your reasoning"

 

It's much better with gpt4 (if you don't have plus that is) but hopefully that kind of prompt will net some good results 🙂 you can also use the reasoning to figure out how it came to it's conclusions, and adapt them as needed!

Edited April 20, 20233 yr by Bigbluss

Now that's a use for ChatGPT I could get behind. The biggest problem for using it in the sim would be that you'd somehow have to get it to properly interpret terrain data, otherwise it'd be CFITing you anywhere the ground wasn't flat.

 

47 minutes ago, Bigbluss said:

If you're serious, try giving it a context! The more detail the better. For example:

 

"Context: you are a loving husband of x years old with a wife aged x. Your wife enjoys (insert hobbies) and you want to treat her to an incredibly thoughtful mother days present with a budget of a to b. Write a list of your top five ideas, explain your reasoning"

 

It's much better with gpt4 (if you don't have plus that is) but hopefully that kind of prompt will net some good results 🙂 you can also use the reasoning to figure out how it came to it's conclusions, and adapt them as needed!

Amazing stuff and amazing days!  I was fascinated by your use of reasoning and deductive logic to speak to the AI in a language and context it could really get its teeth into and showcase its huge store of knowledge. Quite amazing, but it’s understanding of the use Roger is flawed by urbane usage and movies I suspect, but overall like nothing I’ve  ever seen before…

Now that said, Elon Musk who co-Founded OpenAI and now regrets states that the current use and mismanagement of AI Platforms and development poses an existential threat to mankind greater than ever faced.  He goes on and gives very reasonable and very possible scenarios that show Skynet isn’t coming, it’s already here and he says as it stands right now he is terrified.

What say you?

-B

 

Edited April 20, 20233 yr by btacon

7 minutes ago, btacon said:

Amazing stuff and amazing days!  I was fascinated by your use of reasoning and deductive logic to speak to the AI in a language and context it could really get its teeth into and showcase its huge store of knowledge. Quite amazing, but it’s understanding of the use Roger is flawed by urbane usage and movies I suspect, but overall like nothing I’ve  ever seen before…

Now that said, Elon Musk who co-Founded OpenAI and now regrets states that the current use and mismanagement of AI Platforms and development poses an existential threat to mankind greater than ever faced.  He goes on and gives very reasonable and very possible scenarios that show Skynet isn’t coming, it’s already here and he says as it stands right now he is terrified.

What say you?

-B

 

It's a tough one. I'm fairly lucky in that I work with machine learning engineers (my girlfriend actually works in AI too) and have the luxury of people explaining it to me without the hyperbole a lot of the press is firing off.

Personally I think it's very important to keep in mind that this is a statistical model. It's really easy to anthropomorphize these LLMs and attribute a level of intelligence/ cognition, when it fundamentally doesn't exist. The models also don't scale linearly with hardware, and Sam Altman has suggested we've hit the apex point of rapid improvements.

That said, things are moving so quickly now and there's so much investment that it feels hard to predict further ahead than a few weeks at the moment.

The idea of AGI personally frightens me quite a lot. The law of unintended consequences and issues with alignment to human values are both real challenges. But I don't think we're anywhere near that point yet.

 

The idea that Elon Musk, Jeff Bezos, Bill Gates and Mark Zuckerberg have exclusive access to AGIs makes me nervous though. Some of the most narcissistic billionaires on Earth having control of this tech is unnerving for me!

15 hours ago, Bigbluss said:

(my girlfriend actually works in AI too)

Would that make her more likely to figure out when her gifts are AI generated? 😂

39 minutes ago, regis9 said:

Would that make her more likely to figure out when her gifts are AI generated? 😂

She gave me an anniversary card with a poem written by chatgpt. We've actually managed to strip out all the effort from our relationship 😂

This idea has some serious potential if ASOBO could somehow incorporate it into the sim or if one of the developers could harness AI to enhance ATC.  It suggests the possibility of an exciting future for this aspect of the sim which is one area but is not very well implemented as things stand.

A few more different voices and maybe even regional accents are also things I would love to see.