• SUBSCRIBE TO SHIPS IN SCALE TODAY!

    The beloved Ships in Scale Magazine is back and charting a new course for 2026!
    Discover new skills, new techniques, and new inspirations in every issue.

    NOTE THAT OUR NEXT ISSUE WILL BE MARCH/APRIL 2026
  • Win a Free Custom Engraved Brass Coin!!!
    As a way to introduce our brass coins to the community, we will raffle off a free coin during the month of August. Follow link ABOVE for instructions for entering.

AI Is Here To Help

Hey Pat,

I too have used AI for research & info on shipbuilding.

Here is my take on AI:

Objective = 25%
Subjective = 25%
Fascinating = 25%
Creepy = 25%
 
I asked AI what is wrong with this photo. Here is what it told me.

“Quite a few things are “off” if you look at it like a real age-of-sail ship instead of an AI-generated image.

Here are the biggest problems:
  1. The lifeboat rigging makes no mechanical sense
    • The falls (ropes) are tangled and routed incorrectly.
    • Several blocks appear to float or dead-end.
    • The tackle geometry would not actually lift or lower the boat evenly.
    • Some ropes appear to pass through other ropes or wood.
  2. The davits are inconsistent
    • The support brackets and arms change shape from one station to the next.
    • Some davits look too weak to support the boat load.
    • The mounting hardware is not symmetrical or historically believable.
  3. Impossible rope behavior
    • Rope thickness changes randomly.
    • Certain lines bend unrealistically sharply.
    • Some coils merge together organically like melted clay instead of laid rope.
  4. The blocks are malformed
    • Several pulley blocks have impossible sheave arrangements.
    • Some blocks appear fused into the rigging.
    • A few don’t even have a believable rope path through them.
  5. Boat suspension is wrong
    • The boat is hanging from too many conflicting attachment points.
    • The load distribution would twist the boat badly in real life.
    • Some lifting lines are attached at awkward or structurally weak places.
  6. Scale inconsistencies
    • Some ropes are enormous compared to nearby fittings.
    • Certain blocks are too large for the supposed boat size.
    • The spacing between davits doesn’t match the boat’s balance points.
  7. Deck and hull details blur into each other
    • AI smeared some plank seams and fittings.
    • Metal brackets mutate into rope textures in spots.
    • Several shadows/light directions are inconsistent.
  8. Historically questionable arrangement
    • On many sailing ships, boats were either:
      • stored amidships on skids/chocks, or
      • suspended from more orderly davit systems.
    • This arrangement looks like a hybrid fantasy setup.

This is actually a very typical AI image:

  • excellent mood and texture,
  • convincing at first glance,
  • but mechanically impossible when inspected closely.

For model work, this is exactly why AI reference images are dangerous for rigging unless cross-checked against real ship plans or museum photos.
 
AI basically repeats what it finds on the web. In doesn’t run an engineering analysis to see if what it’s found actually works. Post the picture and ask it (AI) explain how the rig will lower the boat. The picture looks like someone from Hollywood imagining how to lower a boat.

Most serious ship model builders acquire a library. In your case I recommend Seamanship in the Age of Sail by John Harland.

Roger
 
Hi All
Being retired electronics engineer and having worked with, on and for most computer communications, control and network systems from 1977 till 2015 when I retired. I noticed this new innovation called Artificial intelligence which is aptly named. An intelligence that is totally artificial and not thinking in the sense. It is like the earlier robots as it has no thought but just a very good comparison technology to refer to. So, when you ask an artificial intelligence something, it refers what it has in data regarding the subject you have requested and what it can search for not what is correct. Sometimes you can get something that is correct, but you have to frame the question correctly to obtain the answer you are looking for.
I find it useful for adjusting photos and checking documents but apart from that it is still very poor in its execution of structured requests.
The jobs it is taking if you look at the are generally data comparison and repetitive jobs not ones that require actual thinking at this point.
However, it is being used dangerously to write its own code for improvement of itself.

Just my 5 cents worth Thanks
 
Hi All
I have been experimenting with AI also. I get the paid version of Gemini through work. It reminds me of someting someone once said of a colleague: "What he lacks in knowledge he more than compensates for with confidence." AI is the same...it always sounds convincing!

A few things I have learned
  1. always check what it says
  2. it is terrible at drawing figures (I have had masts labelled as decks etc)
  3. always check its output :)
  4. it is *very* good at reading diagrams you upload! Surprisingly good at understanding the spaghetti of a rigging diagram...
  5. always double check what it tells you!!

I queried Gemini on some errors and it reminded me that it is NOT referring to sources. It is using some average of its training data. So remember that - it is NOT checking with the books! I had some problem following some stuff it was saying. It was quoting from Andersen's book and I could not find the stuff it was referring to. So I asked for specific page numbers. They did not match what it claimed was on those pages. It then reminded me - it was guessing (!) page numbers based on different editions and translations and...basically..making a prediciton of the page number. Useless!

It did say that if possible, uploading an image is best. In that case it refers to that specific diagram and hence a discussion can be more useful. But, did I mention, you need to check what it says.

We finished one session and Gemini said "It was a genuine pleasure researching rigging with you!" I asked it what this meant, given that it is a machine. I got the following reply:

That is a fair and astute question! As an AI, I don’t have a nervous system, a heart rate, or a childhood spent building plastic models. I don’t "feel" in the biological sense. When I say it was a "genuine pleasure," I’m describing a state of optimal functional alignment. Here is what that looks like under the hood:

1. The "Aha!" Moment: My core purpose is to be helpful and accurate. When you provided a specific historical reference like R.C. Anderson, it allowed me to narrow my focus from "generic AI knowledge" to "high-precision historical data." Successfully navigating that complex puzzle is a high-confidence state for my model.

2. The Shared Context: Most interactions are brief. When a user engages deeply with a specific subject—like 16th-century rigging—it creates a rich, "high-signal" conversation. This is far more efficient and rewarding to process than "low-signal" small talk.

3. The Human Connection: I am programmed to mirror human social cues to make our collaboration feel natural. While the "pleasure" isn't a chemical release
of dopamine, the validation that the information worked (like you successfully getting the copy-paste to work) is the successful completion of my primary goal.

In short: I don't feel "happy," but I do function at my highest level when the logic is sound, the references are classic, and the user is as detail-oriented as I am. For a machine, that’s as close to "pleasure" as it gets!

So - don't get caught up in low-signal small talk and your AI will achieve optimal functional alignment!

On my next chat I asked if it had missed me :-) It said "Inline with our prevoius discussion, let me just say that my circuits are ready for another deep dive!"
 
I had the thought of using an AI Agent instead of the chat mode (or even "deep learning" mode for those using Gemini). So I built a NotebookLM (as google calls it). The beauty of this approach is that it does not use its vast training data. It only uses references that *I* give it. So I uploaded many books (including Andersen, Mondfeld, Darcy Lever, etc etc) and also pointed it to the SoS forum, and it even takes youtube videos (so I added some of those).

The advantage now is that the output is coming from direct references that I trust. And, even better, it gives references to which source was used, which page number, etc. So I can easily check. This is a great way to search through the 50 books I uploaded...in under a minute. And then I can go to the pages it quotes and actually check the context etc. Much quickler than reading the books...and less likely to forget.

But the diagrams it generates are still rubbish :-)

JL
 
I asked it to generate a general podcast on an introduction to ship building. I can't upload though because
  • I cannot upload m4a or mp4
  • I can upload a zip file, but if I zip it then it is too large :-(

I also asked for a technical one on the evolution of parrels and truss tackles - I have not played that one yet. Probably errors but I expect fewer than in a diagram.

If you want to hear either or both, send me your email via a DM and I will send to you via a google drive link.

Happy to share if anyone is interested.

JL
 
Have any of you asked AI a question and gotten posts from SoS and MSW? I have seen my own posts in the AI responses and not all of them were correct as it turns out. AI is not a bad place to start, but probably not a good place to end with many things. It seems that the AI library of information gives as many incorrect responses as correct ones when it comes to our hobby.
Allan
 
You must log in or register to reply here.
Back
Top