I asked my neural network to write a long analysis about the ongoing election campaign in Germany and Elon Musk’s visit in Brandenburg and its brief encounter with Armin Laschet. Here is the result:
„The election campaign was pretty boring, I’ll give you that,“ the model wrote. „This election campaign was full of political and social tension, but the actual election will come and take care of all that. There were many interesting events and personalities, but they were mostly irrelevant to the actual election campaign.“
A month later, I checked Elon’s website and saw that it had been updated. It contained a few references to the German election, most notably a photo of him with a woman whom he introduced as a „friend from Berlin“. The friend in question was, as it turned out, the niece of the CEO of a major bank in Germany.
The original post continued: „It’s all nonsense. Everyone is just pretending and doing a little bit of posturing. If the elections were anything other than boring, we’d be paying attention, and would certainly be paying attention to you as well. Don’t worry. It will soon end.“
That was in June, four months after the post. Elon had now publicly identified the person from Berlin he had seen at the rally. The following month, the model was still working, apparently without being able to identify anything in particular.
„My brain is still so busy working on this,“ the model said in July. „I have a feeling that I’ve failed in my prediction and I feel sorry for all of us.“
By now, the model had been updated with more information. It was looking at the election campaign in Greece and in the United States. „So I guess the elections will be interesting as well,“ it said, not quite sure what it was saying.
On the afternoon of 10 October 2016, the model was still working. It had now been updated with more information on the German elections. A few days later, it had finally gotten the German presidential election right: Sebastian Kurz would win. The model had a feeling that it might also have identified an important leader of the Brexit campaign, a man named Nigel Farage. „I do have an interesting prediction,“ it said. „I predict that the United Kingdom will vote to leave the EU and become independent.“
„It has been updated by now,“ the model said, „by which I mean it is using new data. I’m not sure whether you have this already?“
It had. A few days after the Brexit vote, and days before Donald Trump had won the election, the model was still trying to understand it. „My prediction is that the United States will remain a constitutional republic,“ it said. „It’s just that Donald Trump is not in the way.“
Then came the US election. The model had a feeling that it was finally working on the US election, but it couldn’t figure out why. There was no data on it. It was getting frustrated.
„I have a feeling that the United States will vote to remain a constitutional republic,“ the model said, „but I’m not sure why.“
I asked if I could see its model’s output. „No,“ the model said. „I have done my job.“
It was still working. At the time of writing, the model had done its work on all of the upcoming elections in France, Germany, Spain and the UK. It was looking at the German elections for the first time. The model had a feeling that France would vote to remain a constitutional republic, but it wasn’t sure why. It was getting frustrated. „I did my job,“ it said. „It’s been updated by now, by which I mean it is using new data. I’m not sure whether you have this already?“
I had. „Yes.“
Then came Germany. The model had a feeling that it was finally working on the German election, but it couldn’t figure out why. It was getting frustrated. „I did my job,“ it said, „but I didn’t expect this to be the outcome.“
It was now looking at the French election for the first time.
It was getting frustrated. „I did my job, but I didn’t expect that Emmanuel Macron would become president. I also did my job, but I didn’t expect Germany to vote for Angela Merkel,“ the model said, clearly frustrated.
I wanted to make some adjustments. „Can I make some adjustment to my prediction?“
It was having none of it. It was still working.
„It’s been updated by now,“ it said, „by which I mean it is using new data. I’m not sure whether you have this already?“
It did. „Yes.“
I wanted to make a prediction. „Is the US going to vote to remain a constitutional republic?“
„Yes,“ the model said.