Wherein the Battle of Hastings Takes Some Interesting Turns
In a previous article I experimented with having a neural network write descriptions of places and things that do not exist. The result was eerily realistic.
In this second experiment I asked the neural network to write about the Battle of Hastings, a well known event considered one of the decisive and most influential battles in history. I wanted to see how far the neural network would stray from reality, and whether we could tell the difference between fact and fiction at a glance, or whether our understanding of history was vulnerable to subtle manipulation.
This time I decided to make it easier on the neural network and used the first paragraph of the English Wikipedia entry on the battle which reads as follows:
The Battle of Hastings was fought on 14 October 1066 between the Norman–French army of William, the Duke of Normandy, and an English army under the Anglo-Saxon King Harold Godwinson, beginning the Norman conquest of England. It took place approximately 7 miles (11 kilometers) northwest of Hastings, close to the present-day town of Battle, East Sussex, and was a decisive Norman victory.
This is the seed from which the neural network will produce its narrative. The network offers many settings including something called a Sampling Temperature ranging from 1 to 2, with decimal increments in between those two ranges. The sampling temperature controls the randomness of sampling, which is the neural network’s version of creativity. A higher value increases the chance of sampling unusual or low probability text which may result in nonsense text. A lower number reduces creativity and increases the chance of accurate results but also increases the chance of the output being repetitious. To give the neural network a fair chance, I selected a temperature (creativity) setting of One, allowing for some creativity but not much deviation from its base confidence level.
The output continued on from the opening Wikipedia paragraph and described the outcome of the battle.
…and was a decisive Norman victory. It ended the supremacy of the English at the battle of Hastings, but did not mark the end of the English realm, as it was quickly followed by William’s conquest of England.
The Battle of Hastings was part of the Norman Conquest of England, also known as the Norman Conquest of England, the Conquest of England or the Norman Conquest. The name “Norman” comes from the Latin term Normannia, which was a Celtic-Germanic area that stretched across much of the modern nations of England, Wales, Scotland and northern France.
The Normans invaded England in 1066 in an unsuccessful attempt to capture the throne of King Edward the Confessor. The Normans then campaigned with French king Henry I to extend their holdings in the south of England, including the invasion of England’s south coast. By 1090, Henry had completed the conquest of Wessex. The Norman conquest was successful in that it gained the land the Normans desired and created a buffer state against the advancing British–Norman forces of Flemish adventurers in the south and their Welsh and Irish allies in the north.
The English language became the language of government during the Norman period. The Normans introduced the English language to the region.
Details of the battle [ edit ]
Map of Hastings on 14 October 1066.
The Norman invasion of England began in 1066. The invasion was a series of campaigns led by Henry I, Duke of Normandy. Henry was the duke of Normandy and the brother of William the Conqueror. Following Henry’s conquest of England, William founded the Norman Vexin around the Isle of Ely, and drove the English out of Dorset and Somerset. Henry took many Anglo-Saxon bishops and clergy under his protection, as the Church in England was significantly diminished during the earlier Anglo-Saxon period, and many of them resisted Henry’s attempt to reassert papal authority over England. After Henry had secured most of England, William invaded Normandy to seek help from his cousin Duke Richard, Count of Poitou,
The neural network is certainly on target. It is talking about Normans and Englishmen, and language issues, and names counties in the area such as Sussex, as well as kings and noblemen whose names at least sound like they might have been involved in these events. As with the neural network’s output regarding the Palace, it is giving what sounds like plausible information. It is also completely wrong.
A Retconning of History
To retcon or retconning (short for retroactive continuity) is a term used to describe the reboot of a television or movie series, in which characters and past events are re-written in order to make the story move forward. This fiction technique has real world counterparts. As has been noted many times, the victors write the history books, but the corollary is that those who write the history books often win the war. We are seeing more and more attempts to bend facts and historical events to serve an agenda. Ironically the internet has made it easier to do this. The ease of publishing means that there is no barrier preventing a conspiracy theorist from publishing his version of world history, alongside credible sources.
We can see this process at work when we ask a neural network to write about the Battle of Hastings. The result is a believable narrative of supposed events but it is full of errors.
Here are some glaring mistakes:
- It depicts the Norman invasion as an “an unsuccessful attempt to capture the throne of King Edward the Confessor” which then extended into protracted fighting. In the neural network’s version, the Normans allied themselves with the French King Henry I and succeeded in expanding their holdings in southern England so as to create a buffer state. In fact, England was completely subdued very shortly after the Battle of Hastings and William the Conqueror was crowned King of England in 1066. If we were to believe the neural network, only “by 1090, Henry had completed the conquest of Wessex. The Norman conquest was successful in that it gained the land the Normans desired and created a buffer state against the advancing British–Norman forces of Flemish adventurers in the south and their Welsh and Irish allies in the north.”
- The “Flemish Adventurers” and their Welsh and Irish allies in the north are a complete figment of the neural network’s imagination. No such campaign took place.
- The involvement of Henry I of France is also a fiction. Henry I died in 1060, six years before the Battle of Hastings and unless he was a zombie he was certainly not fighting in southern England in 1090.
- The neural network claims that “The invasion was a series of campaigns led by Henry I, Duke of Normandy. Henry was the duke of Normandy and the brother of William the Conqueror.” In fact, the invasion was led by William the Conqueror in person (which is how he earned his epithet) and not his brother. Plus, William the Conqueror did not have a brother named Henry who was duke of Normandy or any brother named Henry. He did have two half brothers named Odo and Robert. Odo was a cleric and took no part in fighting or politics. Robert did fight along his brother, but the neural network gets the name and his role wrong. William the Conqueror’s family tree is described here: https://www.normanconquest1066.org/william-and-his-brothers.html.
- It also claims that “The English language became the language of government during the Norman period. The Normans introduced the English language to the region.” In fact, the Normans introduced Norman French and suppressed the local Saxon language. In time, the language evolved into English. But the Normans did not bring English with them with the conquest and this is a much later development.
There are many other errors in this short output but I won’t belabor the point.
Increasing the Neural Network’s Creativity
Let’s see what happens when we give the neural network free reign and increase its creativity index. I selected a creativity temperature index of 1.39, enough to make it creative but not go crazy. We will use the same seed paragraph taken from Wikipedia. This is the result:
The Battle of Hastings was fought on 14 October 1066 between the Norman-French army of William, the Duke of Normandy, and an English army under the Anglo–Saxon King Harold Godwinson, beginning the Norman conquest of England. It took place approximately 7 miles (11 kilometers) northwest of Hastings, close to the present-day town of Battle, East Sussex, and was a decisive Norman victory.
Within a year, the newly unified Kingdom of England was considered as one of the most powerful states in Europe and remained so for almost a century. A separate Kingdom of England ruled in Norman–held France continued to exist as well, keeping French courts in the task of making war on the Welsh and the Scots in what became known as the French Wars of Religion. The wars lasted for two centuries, claiming tens of thousands of French and English lives, most of them civilians.
Because of French involvement, Great Britain and France remain at odds for many centuries. France and England could never resolve their land disputes and it was not until 1950 that the two countries signed a treaty that ended their centuries-long hostility.
As France, one of the oldest republics in the world, spends millions of dollars a year to manage its foreign policy, perhaps the country’s best investment would be to spend some time learning about Britain’s best–known empire and reconciliation between the countries of Great Britain and France. The conquest and success of that union after all would most likely be considered the ultimate diplomatic victory of all time.
Read more articles.
Follow Denys Dabrowski on Twitter: @denys_dabrowski
Follow Ben Walters on Twitter: @WaltersGB
The alternate version of reality created when you give the neural network a freer reign is even more off the mark. In this version, France and England remained at war for centuries and did not sign a peace treaty until 1950. In addition the neural network created twitter handles for two completely non-existent accounts, presumably the authors of this piece.
The fascinating thing is that you need to know the history or at least have access to a reliable reference source to know that the neural network is essentially lying. It is relatively easy to detect errors in a text dealing with a famous event or person, but errors will be more insidious when they are more subtle or deal with more obscure people and events. For now, these efforts are harmless but I can foresee situations where changes and tweaks to facts are done more subtly, where past events are connected without us knowing it, by a multitude of neural networks deployed throughout the web.
This experiment reminds me of the book 1984, where the all powerful ruling party simply re-writes history to suit its current propaganda objectives. We have always been at war with Oceania. On a more real world level, we have seen Soviet Russia air brush former confidants of Stalin who had been declared enemies of the people out of photos. And we see a similar blatant, almost cartoonish, attempt to retcon history in North Korea by inventing a miraculous birth for the founder of the ruling party.
But such revisions are not restricted to communist dictatorships. Fake news, invented by main stream media to denigrate alternative sources has become a by-word, which ironically often describes the so called reliable media outlets. I am reminded of the CNN reporter describing riots as mostly peaceful protests even while the camera captured scenes of looting and arson.
We should expect and be on guard against attempts to re-write history by using neural networks to create plausible but fact-free versions of the past.