Using AI for OOBs for Wargames

[bschulte]

@Luke Freet, this one might be of interest to you if you haven't thought of it already.

One of the reasons I haven't been able to do nearly as much with my Petersburg site lately is lack of time. I have three boys with sports, band, and other obligations and we are constantly on the go. I have a degree in Computer Science so I've been really testing the limits of what AI can do for me in terms of "busy work":

• Transcribing newspaper articles: pretty good, but AI tends to "hallucinate" entire phrases that just aren't there. It gets you 90% of the way, and then you need to go line by line making sure everything. Still better than typing it out letter by letter, but not good enough to trust
• Helping with web site presentation: I've only had it review and help update my home page so people know what is available and how to get there, but I'll do this in more detail somewhere down the line.
• OOB Work: This is where I've struck gold. I have been sitting on only three OOBs completed for the first three offensives. It would take me a week or more of tedious work to get everything right for either a Union or Confederate OOB for an offensive, and I just don't have that kind of time. Now, I've been working with AI to scrape the unit pages on my site which contain org and commander data. I'm THIS close to having complete OOBs for the fourth offensive at Petersburg with commanders all the way down to the regimental level. I've thought about how this might be used to help create orders of battle, and here's the idea: You can feed AI the appropriate pages from the Official Records and it will spit out a cleanly formatted OOB with commanders all the way down. Another idea. If you have a spreadsheet or some kind of database with your research in it, you can feed it to AI and have it spit out a clean OOB in minutes or less. This is just the tip of the iceberg. I've thought about having it look through entire OR volumes for unit strengths on certain dates to help with the type of numbers gathering @Luke Freet and others are doing.

Has anyone else been using AI for their Civil War research? This is a weird new confluence of my interests, and I'm enjoying it way too much.

 

[Luke Freet]

@Luke Freet, this one might be of interest to you if you haven't thought of it already.

One of the reasons I haven't been able to do nearly as much with my Petersburg site lately is lack of time. I have three boys with sports, band, and other obligations and we are constantly on the go. I have a degree in Computer Science so I've been really testing the limits of what AI can do for me in terms of "busy work":

• Transcribing newspaper articles: pretty good, but AI tends to "hallucinate" entire phrases that just aren't there. It gets you 90% of the way, and then you need to go line by line making sure everything. Still better than typing it out letter by letter, but not good enough to trust
• Helping with web site presentation: I've only had it review and help update my home page so people know what is available and how to get there, but I'll do this in more detail somewhere down the line.
• OOB Work: This is where I've struck gold. I have been sitting on only three OOBs completed for the first three offensives. It would take me a week or more of tedious work to get everything right for either a Union or Confederate OOB for an offensive, and I just don't have that kind of time. Now, I've been working with AI to scrape the unit pages on my site which contain org and commander data. I'm THIS close to having complete OOBs for the fourth offensive at Petersburg with commanders all the way down to the regimental level. I've thought about how this might be used to help create orders of battle, and here's the idea: You can feed AI the appropriate pages from the Official Records and it will spit out a cleanly formatted OOB with commanders all the way down. Another idea. If you have a spreadsheet or some kind of database with your research in it, you can feed it to AI and have it spit out a clean OOB in minutes or less. This is just the tip of the iceberg. I've thought about having it look through entire OR volumes for unit strengths on certain dates to help with the type of numbers gathering @Luke Freet and others are doing.

Has anyone else been using AI for their Civil War research? This is a weird new confluence of my interests, and I'm enjoying it way too much.

I have a strong aversion to using AI for history. I only begrudgingly accepted using it for work, but I've drawn a hard line in the sand on ever using it for historical stuff.

I will never forget that time on a thread talking about wargaming the Battle of Nashville were some user got upset with me for saying the provisional units in Steedman's Corps should be of lower quality rating, insulting me as a researcher before reposting ChatGPT responses to my posts as if that's a vetted source of knowledge. Genuinely one of the most baffling experiences I've had on this site.

If you are having a hard time putting together OOBs for the various Petersburg offensives, why not reach out to me? I've thrown together a few OOBs for the various offensives, and while a lot of them are incomplete, especially on the Union side of matters, I can probably touch them up and put them on your website. I am a bit more busy nowadays with research on a book on Nashville with my buddy Sean, but I can squeeze in time to help you here.

 

[gary]

I don't use AI, but then again, I'm not smart enough anyway. Second, I don't trust it. The programmer's bias affects their algorithms. Remember when people asked for images of nazi soldiers? You got black males and asian females in German uniforms. OK, there were some black muslims, but very few.

 

[Chattahooch33]

I don't use AI, but then again, I'm not smart enough anyway. Second, I don't trust it. The programmer's bias affects their algorithms. Remember when people asked for images of nazi soldiers? You got black males and asian females in German uniforms. OK, there were some black muslims, but very few.

Reminds me of an occurrence at Bentonville Battlefield a couple weeks ago. We used ChatGPT to give us an introduction and background of the battle while we were on our way and it ended with "Interestingly, the 54th Massachusetts played a pivotal role in the last stages of the battle highlighting the important part played by USCT." @CheathamHill and I looked at each other and I whipped out the OOB and as we assumed, there was nothing of the sort.
I responded saying you are wrong, the 54th was not at Bentonville and it said, "You are right, accuracy is very important."

 

[bschulte]

I have a strong aversion to using AI for history. I only begrudgingly accepted using it for work, but I've drawn a hard line in the sand on ever using it for historical stuff.

I will never forget that time on a thread talking about wargaming the Battle of Nashville were some user got upset with me for saying the provisional units in Steedman's Corps should be of lower quality rating, insulting me as a researcher before reposting ChatGPT responses to my posts as if that's a vetted source of knowledge. Genuinely one of the most baffling experiences I've had on this site.

If you are having a hard time putting together OOBs for the various Petersburg offensives, why not reach out to me? I've thrown together a few OOBs for the various offensives, and while a lot of them are incomplete, especially on the Union side of matters, I can probably touch them up and put them on your website. I am a bit more busy nowadays with research on a book on Nashville with my buddy Sean, but I can squeeze in time to help you here.

Understood on the stance. I'm trying to figure out ways to eliminate the "busy work", or at least drastically reduce it. I've already done the research manually by placing info from the ORs and other sources on my individual unit pages, but this research is there instead of in OOB format. Luckily I was very diligent about making sure every page had the exact same format, and this helped me have AI be able to read and "scrape" those pages for a specific offensive.

In the past I assembled these OOBs manually. Now, I'm just asking the AI to scrape my unit pages and put the info into the OOB as a first step. Then I manually check the result against the ORs and my unit pages to make sure it transferred correctly. It has made this more of a day long rather than a week long exercise. I've already posted the 4th and 5th offensives for each side this week, and am working on the 6th Offensive as we speak. The true delay is in running out of Claude tokens.

Thank you for the offer. I definitely appreciate it. I just have a method I'm sticking to and I do enjoy doing it. My kids' activities take precedence, but my oldest is nearly out of high school, the second is done with sports, and the youngest is now the only still involved in all kinds of activities. So I'm slowly getting back into things.

I tagged you because I thought you and others who do this sort of OOB research could pretty quickly get a "base" OOB set up for a new engagement simply by taking phone pics of the relevant page(s) in the OR or uploading a PDF of those pages to have the QI quickly spit out an OOB with the commanders all listed. More "busy work" elimination.

For me right now, AI is "trust a little but manually verify everything." And it definitely hallucinates. I don't agree with using it to alter historical images and some of the other atrocious things being done out there on Social Media. AI is barreling down the tracks, however, and I'd like to see what is possible with it while also maintaining the strictest standards of historical accuracy and good research.

 

[Luke Freet]

Understood on the stance. I'm trying to figure out ways to eliminate the "busy work", or at least drastically reduce it. I've already done the research manually by placing info from the ORs and other sources on my individual unit pages, but this research is there instead of in OOB format. Luckily I was very diligent about making sure every page had the exact same format, and this helped me have AI be able to read and "scrape" those pages for a specific offensive.

In the past I assembled these OOBs manually. Now, I'm just asking the AI to scrape my unit pages and put the info into the OOB as a first step. Then I manually check the result against the ORs and my unit pages to make sure it transferred correctly. It has made this more of a day long rather than a week long exercise. I've already posted the 4th and 5th offensives for each side this week, and am working on the 6th Offensive as we speak. The true delay is in running out of Claude tokens.

Thank you for the offer. I definitely appreciate it. I just have a method I'm sticking to and I do enjoy doing it. My kids' activities take precedence, but my oldest is nearly out of high school, the second is done with sports, and the youngest is now the only still involved in all kinds of activities. So I'm slowly getting back into things.

I tagged you because I thought you and others who do this sort of OOB research could pretty quickly get a "base" OOB set up for a new engagement simply by taking phone pics of the relevant page(s) in the OR or uploading a PDF of those pages to have the QI quickly spit out an OOB with the commanders all listed. More "busy work" elimination.

For me right now, AI is "trust a little but manually verify everything." And it definitely hallucinates. I don't agree with using it to alter historical images and some of the other atrocious things being done out there on Social Media. AI is barreling down the tracks, however, and I'd like to see what is possible with it while also maintaining the strictest standards of historical accuracy and good research.

I appreciate you're as much of an AI Skeptic as I am.
I'll say, personally, I write up OOBs for fun, whether they be for historical research or for wargaming. My google drive is full to the brim with unused OOBs for Wargames I've yet to run, and I'm constantly making more depending on what part of the war draws my interest that day. I'm fine with AI to replace menial tasks at work so I can have more fun with my hobbies, not vice versa.

 

[Josh The Lighthouse Guy]

AI seems to have particular trouble with tables and numbers. Using selective data input it could probably put together a fairly accurate OOB in terms of commanders, how many regiments in a brigade, and such but the chances of it muddling regimental numbers seems high and I would really distrust what it might come up with for unit strengths.

For transcribing handwritten letters it could be a huge help.

 

[gary]

Reminds me of an occurrence at Bentonville Battlefield a couple weeks ago. We used ChatGPT to give us an introduction and background of the battle while we were on our way and it ended with "Interestingly, the 54th Massachusetts played a pivotal role in the last stages of the battle highlighting the important part played by USCT." @CheathamHill and I looked at each other and I whipped out the OOB and as we assumed, there was nothing of the sort.
I responded saying you are wrong, the 54th was not at Bentonville and it said, "You are right, accuracy is very important."

Thank you for making my point on the foibles of AI. I also hate using it as you will educate it. The more you feed it (as in editing your material), the more it will learn.

 

[gary]

I appreciate you're as much of an AI Skeptic as I am.
I'll say, personally, I write up OOBs for fun, whether they be for historical research or for wargaming. My google drive is full to the brim with unused OOBs for Wargames I've yet to run, and I'm constantly making more depending on what part of the war draws my interest that day. I'm fine with AI to replace menial tasks at work so I can have more fun with my hobbies, not vice versa.

@Luke Freet: Feed in the Confederate strength at Port Hudson minus Miles's Legion. I suspect that Banks could have stormed it had they been absent. They held the southern flank of the line and sent reinforcements to shore up Steedman's northern flank. The reinforcements were later shifted back to meet the afternoon threat when Sherman's Divisiion attacked.

 

[Luke Freet]

@Luke Freet: Feed in the Confederate strength at Port Hudson minus Miles's Legion. I suspect that Banks could have stormed it had they been absent. They held the southern flank of the line and sent reinforcements to shore up Steedman's northern flank. The reinforcements were later shifted back to meet the afternoon threat when Sherman's Divisiion attacked.

was this for a different thread?

 

[bschulte]

I appreciate you're as much of an AI Skeptic as I am.
I'll say, personally, I write up OOBs for fun, whether they be for historical research or for wargaming. My google drive is full to the brim with unused OOBs for Wargames I've yet to run, and I'm constantly making more depending on what part of the war draws my interest that day. I'm fine with AI to replace menial tasks at work so I can have more fun with my hobbies, not vice versa.

That makes sense too. It's a different but equally valid way to save time. I put together OOBs back in the day and have a bunch still on my Google Drive as well. I've moved on a bit to other areas of interest, but I'm a wargamer at heart.

 

[gary]

was this for a different thread?

I did post a thread on it, and no one responded. I asked if the Confederates could have withstood the attacks if Miles's Legion wasn't present.

 

[The Ninth Ohio]

AI seems to have particular trouble with tables and numbers.

Strongly disagree here. In my experience tables is AI's strength, not a weakness. I have had it analyze countless descriptive lists which I've transcribed to data tables, and it's pointed out statistically significant data in seconds, where it would have taken me months, if I had ever connected the dots at all.

Additionally, as I've mentioned in another thread, as long as you know the limits of AI and understand how to interact with it to limit false positives, so to speak, it can be a powerful, effective, reliable tool.

I have consistently used it to narrow down birthplaces listed in descriptive lists that were illegible due to poor handwriting or misunderstanding. Considering I deal solely with ethnic German regiments, it has been invaluable to me.

I have nothing to offer the critics but a shrug of the shoulders.

 

[Josh The Lighthouse Guy]

Strongly disagree here. In my experience tables is AI's strength, not a weakness. I have had it analyze countless descriptive lists which I've transcribed to data tables, and it's pointed out statistically significant data in seconds, where it would have taken me months, if I had ever connected the dots at all.

It probably handles spreadsheet data input a lot better than ledgers or even typed tables amidst long sections of text.

I have nothing to offer the critics but a shrug of the shoulders.

I think AI can be useful, but miracle products should be viewed with skepticism. I've seen the limits of simipler AI - from dubious internet search results to false Ancestry matches to GPS driving "shortcuts" - that I can't be enthusiastic about more complex AI. Especially since it's usually a rushed product, trained on the internet, programmed to maximize engagement, endorsed by sociopaths like Musk.

 

[rbasin]

AI can definitely help with finding data and resources, but fails in history.

 

[Luke Freet]

I did post a thread on it, and no one responded. I asked if the Confederates could have withstood the attacks if Miles's Legion wasn't present.

Ok, which thread? Because I prefer people don't go off topic on threads.

 

[Stryker65]

As another OOB enthusiast, I can also not help but "thumb my nose" (as they say) at AI usage for OOBs -- there are already too many mistakes in the Wikipedia OOBs for me to trust any other source than originals.

 

[The Ninth Ohio]

That makes sense too. It's a different but equally valid way to save time. I put together OOBs back in the day and have a bunch still on my Google Drive as well. I've moved on a bit to other areas of interest, but I'm a wargamer at heart.

Just to be clear, you're using AI to clean up research you've already done, and then checking the results afterward?

That's a perfectly great use of AI, and you will lap others in your area of interest who are not doing the same. Keep it up.

 

[bschulte]

Just to be clear, you're using AI to clean up research you've already done, and then checking the results afterward?

That's a perfectly great use of AI, and you will lap others in your area of interest who are not doing the same. Keep it up.

Exactly right. The process is this:

• Spent 2009-2020 or so doing all of the research to get hundreds if unit pages published on my site.
• Used the Official Records end of every other month oobs and some other fill-ins to list the organizations to which a unit belonged in the 1st through 9th offensives.
• Posted OOBs for my first three offensives pre-AI by hand, a process which generally took me a week of off and on work.
• Used AI recently to scrape a Word doc containing the names and URLs of every unit page on my site...
• ...create a fourth offensive OOB...
• ...and format it the way I wanted.
• Lastly, spent about a day manually checking the AI created OOBs and comparing them to my original sources to make sure everything was correct.

Saved me about 5 or 6 days of work, and I've already popped out AI versions of the 5th and 6th Offensives which need manual checking. Thanks!