Thought experiments are used to explore chains of causes and consequences. They are useful to create mental models, especially when analyzing events.
To understand each type of thought experiment, it is important to consider whether you are exploring events in the past or the future, as well as the direction in which you are conducting your exploration (yellow arrows on the image). Are you starting in the past and going to the present or are you starting in the present and going back to the past?
The types of temporal thought experiments. The yellow arrow displays the direction of your reasoning.
Here are the seven types of temporal thought experiments:
Prediction: What will happen?
Predictions are a large family of thought experiments focused on the future. You are neither manipulating the past nor the future, but foreseeing what will happen if all mechanisms at play follow their natural course. When carrying out a prediction, you are thinking of future events in a sequential manner. Examples include:
Nowcasting a specific Twitter hashtag trend in 2 hours.
Forecasting next Sunday’s weather.
Predicting a global change in world politics.
Retrodiction: What happened?
Retrodictions are a large family of thought experiments concerned with the past. Just like in a prediction, you are not manipulating any event. Here, you are focused on understanding what has led to a given situation. And for that, you start your reasoning from the current point in time. A retroduction is, therefore, the equivalent of prediction, but with a focus on the past.
Reverse engineering follows the concept of retrodiction: You have a situation and you are working out the steps that led to such an outcome.
Hindcasting: Test idea on the past
When hindcasting, you create a model and you verify how well it predicts the past. Essentially, you are starting in the past with a specific model in mind. As you move along the timeline, you will see whether you arrive at the same present. Once again, there is no manipulation of the past, just a descriptive model. If you did manipulate the past, that would become a counterfactual.
"I made a model that predicts conflict, but when I tested it with past data, it missed WWI and WWII."
Backcasting: Retroplanning
Backcasting is fundamentally a retroplanning: You have a situation in the future and you ask yourself what needs to happen for you to arrive there.
You start from an end state, then gradually go back to where you currently are. By doing that, you discover, step-by-step, what is required to get to the result.
Prefactual: If-then
“IF A, THEN B will happen”. In a prefactual, you manipulate or create an event, and you see how this conditions the future. You can think of a prefactual as a counterfactual focused on the future:
"If prices keep rising, people will start to protest."
Semifactual: Even if
A semifactual focuses on what outcome would have been the same, given a change in the past. In essence, it is a counterfactual that emphasizes not what has changed but what remains the same:
“Even if he had left earlier, he still would have been caught up in traffic.”
Counterfactual: What if
As the big star of thought experiments, counterfactuals are getting a lot of press right now, especially due to research in interpretable AI. Counterfactuals explore the potential outcomes of a change in the past and ask "what would have happened if A happened instead of B?". They are indeed a thought experiment that sounds particularly natural to humans when searching for an explanation: If you change an event in the past, what would have happened instead?
"What would be the state of the European economy today, had the covid-19 pandemic never happened?"
"During my last time travel, I stepped on a flower. When I came back to the present, Apes were ruling the world."
The latter is actually a negative example of how to use a counterfactual in political science, due to the concept of causa proxima. But we will save that for another post.