A lesson from an F-16... and from AI

Ilias Lappas*

This morning I learned something interesting about the limitations of AI in engineering. I asked ChatGPT a seemingly straightforward question:

"What was the static margin of the F-16A and what was the one of the F-16C?"

The answer sounded convincing. It referred to NASA reports, discussed relaxed static stability, and concluded that later F-16 variants probably remained negatively stable. There was only one problem. It wasn't the whole story.

Fortunately, I happened to own a copy of one of my favourite aerospace textbooks:

‘Introduction to Aeronautics: A Design Perspective (AIAA)’, by Brandt, Stiles, Bertin & Whitford.

The textbook contains a complete worked stability analysis comparing an early F-16A with an F-16C. Using classical aircraft stability methods, the authors estimate:

Early F-16A: Static Margin ≈ - 8% MAC

F-16C: Static Margin ≈ +1% MAC

The increase in horizontal tail area shifts the neutral point sufficiently for the later aircraft to become slightly statically stable.

Was ChatGPT "wrong"? Not exactly.

The NASA reports it found discuss the early F-16 programme and describe the aircraft as operating with a moderate negative static margin. Other sources repeat similar statements.

The important lesson is this:

AI synthesizes the information it can access. It does not automatically know which source is the most authoritative, nor can it retrieve every textbook or proprietary engineering report ever published.

As engineers, we are trained to ask:

* Where did this number come from?

* Is this a primary or secondary source?

* What assumptions were made?

* Does another reputable source reach a different conclusion?

Those questions remain just as important when the answer comes from AI.

Ironically, the most valuable part of the conversation wasn't the initial answer. It was the discussion that followed. By comparing sources, questioning assumptions, and checking an engineering textbook, the final understanding became much stronger than either the AI or I could have reached independently.

Perhaps that's the best way to think about AI in engineering:

Not as an oracle, but as a very knowledgeable colleague whose work still deserves peer review.

As engineers, we've always reviewed each other's calculations.

Why should AI be any different?

Ilias Lappas*

 CEng FRAeS, Aerospace and Defence professional |Aeronautical Engineer | Pilot, University of South Wales · Harvard Business School Online