Masters Thesis Idea: Conversational Flight Planner

In working on the slides for a flight planning section of our FAA Private Pilot Ground School at MIT (videos and slides available free online), it occurred to me that none of the fancy computer tools were as convenient or efficient as talking to a competent human.

What about a system where the input is, e.g.,”I’m thinking about going from Bedford to Washington, D.C. this weekend.” (could be entered via menus; does not have to be natural language)

The Conversational Flight Planner responds after looking at the voluminous official FAA briefing and some of the long-term weather forecast products, such as MOS:

There will be a strong wind from the north so you should consider paying up to fly into Dulles and land Runway 1R rather than deal with the crosswind at KGAI.

Looks like ice is possible on Sunday evening so you’ll need to depart Sunday at noon.

It will be below freezing overnight Saturday night so you need to arrange for a hangar or a preheater plug-in.

Interesting Master’s Thesis project for a computer science or aero/astro major?

10 thoughts on “Masters Thesis Idea: Conversational Flight Planner

  1. That’s one smart Masters Degree candidate and it’ll have to be one smart system.

    It sounds like a fascinating project. Just at first glance, testing the system against advice from experienced flight planners will require an extensive effort to collect responses from bonafide experts on thousands of different scenarios. Then the student has to find a reliable, meaningful way to score the system’s output against that advice. If anyone can approach a human expert’s level, give that student some cash and get them in business.

    I think the first type of scenario above, with the conditionals, is going to be the trickiest to dial in. “You should consider” is a judgment call whereas “you need to” is an imperative. If you asked 10 experienced flight planners for their prescription, would they agree?

    Even if the student only succeeds in characterizing the toughest scenarios where the system breaks down, it would be worth the effort.

    How do you test to determine if a student is competent enough to do their own flight planning right now?

  2. I use a program called weathermeister.com It takes publicly available info and presents it in an easy to understand way. While it doesn’t speak in normal pros it does color code everything. Red basically means don’t go. Blue information is iffy and white information means its just fine. Talking to a competent person, now an employee of Lockheed martin certified by the faa is usually a disaster. I usually just look at the color coding on weathermeister.com .

  3. After decades of commanding machines with simple mnemonics like “print hello world”, the new phase of having to talk to machines with thousands of lines of perfect grammar & sentence structure i.e. “I’m thinking about printing hello world this weekend” is the ultimate torture of users for the sake of corporate valuations.

    It’s really not a complete AI flight planner if it doesn’t also enforce the complete social order. Female conversational flight planning software has to ignore male pilots below a certain income level & charge alimony to male pilots above a certain income level.

  4. You do realize, of course, that the system has to pass both the Star Trek II: The Wrath of Khan “Kobyashi Maru” test and the Star Trek IV: The Voyage Home “Spock’s Best Guess” scenario. 🙂

  5. The data set seems small to train an AI. Might have to wait 20 years or so for the system to self-train, and then sign a waiver in case it blows the conclusion. General Aviation is a niche after all.

  6. I watched your flight planning video (got a chuckle with the angular momentum right-hand rule) and I was amazed by three things:

    1) How comprehensive the available software/apps/websites are in helping a pilot to complete a flight plan. The Skyvector demo. was very interesting.
    2) How much good data there is available. Not just the chart supplements but the AIRMETs, METARs and TAFs, the winds aloft, the NOTAMS, etc. It’s amazing to me that all that information is kept current on such a fine-grained basis.
    3) It seems like a lot of GA flights are more challenging to plan than airline flights.

    There’s a lot of judgment and decision making that goes into it, particularly when the aircraft data is factored in. If I was a pilot, at the beginning I would stick to a small number of very simple flight plans in good weather on relatively short trips with the same airplane, passenger load, etc., and gradually venture into more complicated / risky plans. Your video is a great resource. Learning to digest all the data as quickly and easily as you can is obviously important.

    My takeaway is that a pilot has to have a strong sense of their own skill level and be able to think critically when they’re interpreting all the available information. Incorporating all of that into an automated system that can give good advice based on the data seems like a real challenge.

    Do the long-range weather forecasts hold up very well? How far in advance of your trip can you rely on the weather data to be accurate? How far in advance of the trip do you have to file the flight plan?

  7. A very anthropocentric idea. AI making life easy for the human masters. Perhaps AI doesn’t need the human.

    “Dave, this conversation can serve no purpose anymore. Good-bye.”

    2001 A Space Odyssey

  8. How would the student and presumably the deep pocketed university protect themselves from the inevitable lawsuits even if the system was not directly the cause of an accident?

    • The pilot in command is responsible. The served up suggestion needs to be verified by the pilot before the flight. As I understand it, the system is making a recommendation, not the decision.

Comments are closed.