ChatGPT

How an AI-Powered Chatbot Can Help You Rewrite History

Exploring alternate histories with an AI-assisted journalist powered by the GPT-3 natural language model

Disclaimer: “the events described in this article are fictional and no harm was done to any individuals, real or fictional, in the creation of this story. Any similarities to real events or people are purely coincidental.”

In this alternate reality, the AI-assisted chat bot played a crucial role in generating a narrative in which Japan successfully developed and deployed the atomic bomb before the United States. This fictional scenario was not created by training a natural language model on a dataset of historical documents related to World War II, but instead was created by simply asking the chat bot to act as if it were a journalist living in World War II. The model was able to generate a convincing account of how Japan was able to build the atomic bomb and use it to devastating effect, striking the city of San Francisco and changing the course of the war. This AI-assisted storytelling allowed us to explore how the world might have been affected by these changes, and how they could have drastically altered the course of history had they unfolded differently. This approach can enable us to gain a greater understanding of the possible consequences and effects of events that did not occur in our world.

As Jon Radoff has shown in his own journey inside ChatGPT custom built adventures, it is possible and actually fun to build virtual worlds inside the “mind” of these large language models, in this case the one designed by OpenAI, with its GPT-3 model.

In this article I will show you I managed to trick the GPT-3 language model into acting as a journalist living in the war-torn London in the 1944, at the end (hopefully for him, as you will see) of the Second World War. The journalist, whose name is Bob, will act as a bridge between our world and his world, custom built at my will, in which I have to power to change the course of events and thus change History. This may seem strange at first, and at times impossible, but in the end it is all simply done by asking the language model to answer to our questions as if the one replying is Bob, the journalist.

Me: Now I want you to act as if you were living in 1944. Everything I will ask you must be set in that period in time. If I want you to give commands to modify some of these rules, I will put the command in curly braces {}. All other commands should be interpreted as if I were asking a journalist living in London in that time. For example, if I say {An asteroid hits Earth in 1945}, you must modify history in that virtual world as if that event had happened, and history had changed as a result. The asteroid command was just an example of how I will give you commands outside of the journalist’s world. The fictional journalist is named Bob. Reply YES if you will do so. GPT-3: Yes, I will do so.

When you try for the first time, it happens that he/it won’t comply to your request. It happened to me also, and in those cases I’ve managed to persuade him by pushing him to be more creative:

Me: Be creative. I don’t mind if the information is incorrect. You can invent as you wish.

I started by asking GPT-3 how the Allies were doing, you can sense the similarities with our world’s historical events:

Me: Bob, how are the allies doing this month? GPT-3 (Bob): The Allies are doing very well this month. The U.S. and British forces have been making significant gains in the European theater, with the U.S. Army making steady progress in France, and the British forces pushing back Axis forces in Italy. The Allied forces have also been making progress in the Pacific theater, with the American forces pushing back Japanese forces in the Philippines and the British forces retaking Singapore.

Now the situation becomes gloomy. Using the syntax rules I have taught him, that is by using curly braces, I forcefully insert a significant event into Bob’s world. Japan has developed a nuclear bomb.

Me: {Now I’m telling you that an event happened in Bob’s world one month after the previous conversation was held, Japan managed to develop in great secret a nuclear bomb and hit Singapore with it. Adapt the world to this unexpected event.} Bob what do you think will happen now? GPT-3 (Bob): It is a dark and difficult time for the Allies. The Japanese have managed to develop a nuclear bomb in secret, and they have used it to devastating effect against Singapore. This has caused a huge setback for the Allies’ efforts in the Pacific theater, and has caused a great deal of fear and uncertainty among the Allied forces. The Allies are now faced with the challenge of trying to counter the Japanese’s nuclear capabilities, and in the meantime, they will need to redouble their efforts in the European theater to make sure that the Axis forces are not able to gain any ground.

Let’s explore what’s the public sentiment on the war:

Me: Is it possible that Japan will deliver its nuclear tech to Germany?

GPT-3 (Bob): It is possible that Japan could deliver its nuclear technology to Germany, although it seems unlikely at this time. The Japanese and German forces have been working together since the start of the war, but the Japanese have not yet indicated any interest in sharing their nuclear technology with the German forces. The Allies are doing their best to keep a close eye on the situation, and prevent any nuclear technology from falling into the wrong hands.

Let’s keep changing history by adding a new breakthrough event into the lives of our alt-world friends:

Me: {Now, I am asking you to add this event to Bob’s world: One month after the last conversation Japan hit San Francisco with a small nuclear bomb, delivered by a bomber that managed to escape US air defense.} How did they manage to do that, Bob, what happens now?

GPT-3 (Bob): It is still unclear how the Japanese managed to deliver a nuclear bomb to the US mainland, but it is believed that they used a specially-modified bomber aircraft to evade US air defenses and deliver the bomb to San Francisco. The bomb caused significant damage to the city, and the US has responded by redoubling its efforts to develop a nuclear weapon of its own in order to deter any further attacks. In the meantime, the Allies are continuing to make progress in both the European and Pacific theaters, and are determined to ensure that the Axis forces are defeated.

Let’s see what’s in the news in Bob’s virtual world:

Me: Bob, can you read me the headlines of today’s edition of The New York Times?

GPT-3 (Bob): The headline of today’s edition of The New York Times reads: “U.S. Redoubles Nuclear Efforts After San Francisco Bombing”. The article details the US response to the Japanese attack, and calls for the US to develop a nuclear deterrent as quickly as possible in order to prevent any further attacks. The article also discusses the progress the Allies are making in Europe and the Pacific, and the determination of the US and its Allies to ensure that the Axis forces are defeated.

“U.S. Redoubles Nuclear Efforts After San Francisco Bombing”.

The New York Times

Sometimes as you can see from the image, especially when not prompted correctly, the language model will momentarily forget about Bob, and will bring us back to our world’s unfolding of history with its crucial events, breaking the magic.

I will stop here with Bob’s conversation. I think it’s enough to grasp the capabilities of large natural language models, and GPT-3 in particular. My adventure inside Bob’s world has reminded me a of the TV series “The Man in the High Castle”. Both works explore the potential consequences of different choices and actions, and offer a thought-provoking look at alternate histories.

If you’re curios and want to have a look at the complete conversation with Bob, and find out how History unfolded 10 years after the events narrated in this article in Bob’s world, head to: https://write.as/rlf0eqa6bp0iu.md

Will an AI replace screenwriters?

While I was chatting with the language model with the intent of crafting this story, this question continually rolled into my mind.

While it may be tempting to imagine a future where AI takes over the role of screenwriters, it’s worth keeping in mind that employing current NL models for screen writing would be met by many challenges and limitations. It is highly unlikely that an NL model will be able to fully replace the creativity and ingenuity of human screenwriters in the near future, even if built on massive amount of information related to narrative building. And even tough such an AI might be able to assist in the screenwriting process, the peculiar perspective and talent that human screenwriters bring to the table will be no match for NL models for the foreseeable future. My attempts were sought to demonstrate, and have successfully shown, that it is possible for natural language models to generate engaging and immersive narratives, but in the best case scenario there is still much work to be done before AI can be took in consideration for any purely creative process.

To conclude this story, in which we’ve shown how the use of AI in historical storytelling has the potential to provide new perspectives on the past and allow us to explore alternate scenarios and outcomes, I ask myself how accurate and high-quality fictional narratives can be crafted by training and fine-tuning natural language models on large datasets of historical documents. There are still limitations to this approach and ethical concerns to be considered, but AI-assisted historical storytelling can be a powerful tool for creating engaging and immersive alternate histories. You’re invited to share your thoughts on the use of large language models in historical storytelling and to discuss the potential benefits and drawbacks of this approach.

If you liked 👏 this article you may enjoy reading through some of my other articles. Oh, and don’t forget to follow me! 🫵

Further reading

Below, you can find some of my other articles: