I’m a big fan of history and it is always interesting to think of the ‘what if?’ scenarios. In fact there are plenty of tabletop war games and a lot of video games dedicated to playing out alternative scenarios and also some great books on the subject. ChatGPT offers another way of exploring alternative scenarios and well.. re-writing history. At the moment, it isn’t too sophisticated.
Changing history also has political implications – describing events in a slightly different way can give a different slant. It is rarely objective and is almost always based on an incomplete version of the exact facts. Outcomes are mostly undeniable however.
I got ChatGPT to run through a couple of scenarios during the American War of Independence or the Revolutionary War as it is also known. One pivotal moment was the Battle of Trenton on the 26th of December 1776. The Hessians (German auxiliaries to the British forces) were defeated in a surprise attack by the Continental Army which re-invigorated morale and inspired others to join the cause. It also inspired this famous painting of George Washington on his way to the battle, crossing the icy Delaware:
Washington Crossing the Delaware, by Emanuel Luetz, Metropolitan Museum of Art (wikimedia commons)
The story goes that the attack began at daybreak and the Hessian garrison were caught by surprise and overwhelmed by Washington’s Continental Army. Some say they were drunk after Christmas festivities but this seems to have been misinterpretation. There is even a webpage dedicated to myth-busting things about the battle.
Changing the outcome of the Battle of Trenton
So back to ChatGPT – would a different result in this battle have resulted in a different outcome to the War of Independence? I asked it to create a new wikipedia page based on the Hessians winning at Trenton. It is interesting what tack is taken – in the hypothetical ‘win’ scenario Colonel Rall on the Hessian side receives intelligence about Washington’s attack. Note that this was a factor in Washington’s real victory – he had a spy in the Hessian camp that had played down the state of the Continental Army (this was a small factor and it was also true that the Continental Army soldiers barely had shoes, even with the freezing conditions). ChatGPT does clearly present it as an alternate scenario (although I asked it do this – my prompt was: “Create a wikipedia page about the battle of trenton, but in a scenario where the hessians win”). The text broadly sticks to the facts of what happened, with the alternate twist being that the Hessians are not surprised by the attack and are ready and waiting (disciplined as they were).
Interestingly ChatGPT also states that it would not have changed the outcome, so the timeline reverts back to the real world outcome by the end of its text. This is an interesting position to take because of course the path of the war after that is non-deterministic, but the body of real-world text brings the world back to one where the Americans achieved their independence.
Here’s the real wikipedia page about the Battle of Trenton and below is my generated one:
“Battle of Trenton (Hessian Victory)
The Battle of Trenton was a pivotal engagement that took place during the American Revolutionary War on December 26, 1776, in Trenton, New Jersey. In this alternate scenario, the battle resulted in a resounding victory for the Hessian forces under the command of Colonel Johann Rall, instead of the historical outcome where the American forces emerged triumphant.
After several early defeats in the war, General George Washington, commander-in-chief of the Continental Army, sought to rally his demoralized troops by planning a daring attack on the Hessian garrison in Trenton. The Hessian mercenaries, led by Colonel Rall, were stationed in the town and played a crucial role in supporting British forces in the region.
Deployment of Forces
In this alternative scenario, Colonel Rall had received intelligence about Washington’s planned attack. Recognizing the threat, he ordered his troops to adopt a more cautious and defensive position, fortifying key positions within Trenton. Rall made use of the town’s natural defenses, such as the Assunpink Creek and the houses along the streets, to create strong defensive positions for his forces.
On the American side, Washington’s plan to launch a surprise attack relied heavily on the element of surprise. However, in this scenario, the Hessian defenses, combined with their knowledge of the American plan, made the task much more challenging.
As dawn broke on December 26, 1776, the American forces, numbering around 2,400 troops, attacked the Hessian positions in Trenton. However, they were met with a stubborn and well-prepared defense from the Hessian forces, which numbered around 1,500 soldiers.
The American assault faced fierce resistance as they attempted to breach the Hessian defenses. The Hessian soldiers, well-disciplined and fighting from fortified positions, repelled multiple American charges. The Continental Army suffered heavy casualties in their attempts to break through the Hessian lines.
Despite their initial setbacks, Washington’s troops continued to press the attack, but the Hessian defenders held their ground. Colonel Rall skillfully directed his troops, countering American maneuvers and maintaining the cohesion of his forces. The Hessian soldiers fought with determination, inflicting heavy losses on the American forces and frustrating Washington’s attempts to gain a decisive advantage.
After hours of intense fighting, the American forces, depleted and demoralized, were forced to withdraw from Trenton. Colonel Rall’s strategic defenses and the resilience of his troops secured a resounding victory for the Hessian forces.
The Hessian victory at the Battle of Trenton had significant implications for the course of the American Revolutionary War. The loss dealt a severe blow to the morale of the Continental Army and dampened the spirits of American supporters. The successful defense of Trenton bolstered British and Hessian confidence and consolidated their control over the region, hampering American efforts to gain momentum in the war.
Following the Hessian victory at Trenton, the British and Hessian forces expanded their control over New Jersey, further stabilizing their hold on the region. The loss at Trenton forced General Washington to reevaluate his strategies and consider alternative approaches to challenging British and Hessian dominance.
Although the Hessian victory at Trenton temporarily halted the American momentum, it did not ultimately alter the outcome of the Revolutionary War. The conflict would continue for several more years, with the American forces rebounding and achieving significant victories in subsequent battles, eventually leading to their independence from Great Britain.
The Battle of Trenton, in this alternative scenario, stands as a testament to the effectiveness of Hessian defenses and the formidable resistance they put up against the Continental Army. It remains a noteworthy event in the annals of military history, showcasing the complexities and uncertainties of warfare.”
The great thing about ChatGPT is that having established this context in the conversation you can ask follow-up questions. I asked it what Washington’s next steps would have been and it generated what I would say was a relatively generic (classic GPT style) response. I think that highlights the limitations of where were now with this kind of AI, but if supplied with additional context it does provide very quick analysis – much faster than a human being could perform. However, it does require the human to understand what actually was going on at the time to help correct it. If you do understand some of the facts on the ground, then you can use these as modifiers to help play out other sub-scenarios. Here’s another time I was looking at this – I asked ChatGPT what it thought about the decisions at the time (there was a huge blizzard). The answers are pretty much what you or I would come up with if we thought about it, so in that sense there is the magic of the clairvoyant at play if you don’t think too hard. I’ve constrained the options significantly (both for those originally on the ground and for the AI):
So the benefit is a rapid playing out of different scenarios but again you do need to correct it. Here there is some ‘AI hallucination’ – it confidently states that the outcome of the battle was determined only by strategic factors and the actions on the ground, not the fact that it was Christmas. For context here – there is no clear answer to why they were caught unprepared – it was probably a combination of things including Christmas festivities and crucially the Hessians had been warned that there may be an attack. When I pointed this out (and this has happened a few times when I have tried this), it feels like there is a lot of back-tracking (of course this is just the reader’s perception):
ChatGPT can be useful in thinking about things where you may not have explored them yourself (and quickly). For example, the Hessians didn’t use their artillery in the battle, so I sought ChatGPT’s opinion on why. Again the caveat is that it will not necessarily consider ‘every’ factor available in history books – but you can certainly imagine a future where it will:
Finally – we can jump around the theatre of war to other points – for example – the Caribbean islands were very strategic when it came to French re-supplies:
There are many different scenarios that can be played out that could have helped the British which they clearly didn’t pay enough attention to at the time – of course very easy with the benefit of hindsight. We’re also not considering the huge number of almost uncontrollable factors that were playing out at the same time – whether it be dealing with the French, the Dutch merchants or the opinion of parliamentarians.
Polluting the datasets or erasing history…
So in summary, ChatGPT is quite a nice tool for exploring alternative scenarios in history, but it has limitations right now. It is not difficult to imagine modern-day commanders utilising AI in their decision-making, but just as we see above, context and accuracy are crucial and that’s all without an active adversary that may just be polluting the data with other alternative facts or even changing history… It will be ever-more important that original historical sources are preserved and protected from change or destruction.
For some further reading on the Revolutionary War, I recommend the excellent books ‘The First Salute‘ by Barbara W. Tuchman and 1776 by David McCullough.
David, Excellent reflections on the emerging “fun” with AI – Things to be explored and considered as we all move onwards. For now and for the sake of preserving a clear view of US history I’ll hang onto my paper copies of David Hackett Fischer’s many evidence based historical analyses including “Washington’s Crossing”. More valuable I like to think is a comprehensive “masterpiece” of his written in 1970, to justify the need for working from a solid evidence base. A book that I have used over the years when I have been pondering the distortions delivered in all fields by people of ‘untrustable’ intentions – and in anticipation of progress being made with AI: “Historians’ Fallacies – Toward a Logic of Historical Thought”. I will (as ever) go out on a limb and suggest that this book with its many categories of fallacies is an excellent lens to observe the weaknesses and dangers of current AI approaches – and, if my colleagues had won the necessary funding, would be a structure for considering issues and potential remedies. 🙂