Methodology for ICT Writeup

The Book of Why

Fiction, Science Fiction, and Historical Fiction are a way of exploring counterfactuals

Science as Falsification

Labs implicitly judge research through the sense of dollars per paper

An Institute for Creative Technology - Riffing on ARPA

# Counterfactuals are hard
People have always loved to find causal relationships between things in the world. Causal relationships are also incredibly important for changing and improving the world! You need to have confidence that “if I do X, Y will happen.” 

However, 

In situations where you have a manipulable model, you can interrogate it directly -- “if I change these things, what will happen?” 

Any time we assert a causal relationship between two things, we implicitly assert a counterfactual at the same time. However, when that counterfactual is made explicit, it often doesn’t hold up to scrutiny. People will say “you’re fat because you eat chips” but won’t agree that if you changed nothing but your chip eating habit, you would be skinny. 

In the real world it is basically impossible to hold everything constant 

One upshot is that [[How do you find counterfactuals?]] is perhaps a more important question than we treat it. 

### Related
- [[Fiction, Science Fiction, and Historical Fiction are a way of exploring counterfactuals]]
- There is some economics work on how to account for counterfactuals in accounting. [[Shapley values - Better than counterfactuals]] - when you’re
- [[Forecasting Counterfactuals in Uncontrolled Settings (FOCUS)]]
- [[The Book of Why]]

[Web URL for this note](http://notes.benjaminreinhardt.com/Counterfactuals_are_hard)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Counterfactuals_are_hard)






Counterfactuals are hard

# Fiction, Science Fiction, and Historical Fiction are a way of exploring counterfactuals
[[Counterfactuals are hard]]. They’re basically a white-space for you to fill with your biases. On a personal level counterfactuals are daunting. What would have happened if you had asked that person out? Taken that job? Moved to that city? Now take all those things that are impossible to know and amplify them to a societal level.  What would have happened in 2008 with a larger bailout? A smaller one? If Clinton had been elected instead of trump? And those are just discrete events. Now consider counterfactuals to long-lasting policies - what would have happened without the Bayh-Dole act? With a much smaller NSF? 

Economists and historians try to illustrate counterfactuals with data.  Natural experiments can show what happens when a policy is implemented in one place vs. another, trends that change suddenly can suggest what would have happened, experiments in labs hint at what human behavior would have led to in a different situation. But the world is so complex that everybody can basically pick their own data-driven explanation. As far as I can tell after a good amount of time looking, there are no rigorous tools for thinking about counterfactuals beyond telling stories disguised as causal explanations.

Here’s a different proposal: embrace the story-telling aspect of counterfactual analysis. Instead of pretending you’re not telling a story, explicitly tell a story and use it to explore the complexity of the real world. Good historians who explore the rich complexity of the past do this to some extent. Now change something and do the same rigorous analysis. “But this is just historical fiction!” Yes, but in the same way [[You can use science fiction to explore a future world in a way that nonfiction cannot]], my hunch is that you could unlock something special by writing historical fiction with the goal of understanding instead of just entertaining.

[[Good science fiction is self-consistent]] - economists and historians could use that self-consistency requirement to play around with different scenarios. 

Fiction-based counterfactuals would be much less satisfying and atomic than tight academic explanations. However, they might end up being more thought-generating and truth-capturing. 

One possible way of answering [[How do you find counterfactuals?]]

## Questions
## Related
- [[Scientists, Engineers, and Economists should write more science fiction]]
- [[Counterfactuals are hard]]

## References
- [[Forecasting Counterfactuals in Uncontrolled Settings (FOCUS)]]
- [[Financial Historical Fiction]]



[Web URL for this note](http://notes.benjaminreinhardt.com/Fiction,_Science_Fiction,_and_Historical_Fiction_are_a_way_of_exploring_counterfactuals)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Fiction,_Science_Fiction,_and_Historical_Fiction_are_a_way_of_exploring_counterfactuals)




Shapley values - Better than counterfactuals

# How do you find counterfactuals?



[Web URL for this note](http://notes.benjaminreinhardt.com/How_do_you_find_counterfactuals)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/How_do_you_find_counterfactuals)




# Shapley values - Better than counterfactuals
Published: October 2019
Counterfactual analysis can overcount impact 
(If something is decided by one vote, each voter provides that much value) 
::Doesn’t that just mean it’s not actually a good counterfactual accounting?::
[Shapley value - Wikipedia](https://en.wikipedia.org/wiki/Shapley_value)

Related: [The counterfactual impact of agents acting in concert - EA Forum](https://forum.effectivealtruism.org/posts/EP8x3vHRQJP57TjFL/the-counterfactual-impact-of-agents-acting-in-concert)

### These properties:
* Property 1: Sum of the values adds up to the total value (Efficiency)
* Property 2: Equal agents have equal value (Symmetry)
* Property 3: Order indifference: it doesn’t matter which order you go in (Linearity). Or, in other words, if there are two steps, Value(Step1 + Step2) = Value(Step1) + Value(Step2).
### And an extra property:
* Property 4: Null-player (if in every world, adding a person to the world has no impact, the person has no impact). You can either take this as an axiom, or derive it from the first three properties.

### Related
- [[Counterfactuals]]
- [[How do you find counterfactuals?]]
### Link
[Shapley values: Better than counterfactuals - EA Forum](https://forum.effectivealtruism.org/posts/XHZJ9i7QBtAJZ6byW/shapley-values-better-than-counterfactuals#:~:text=who%20comes%20first.-,The%20Shapley%20value%20is%20uniquely%20determined%20by%20simple%20properties.,to%20the%20total%20value%20(Efficiency))


[Web URL for this note](http://notes.benjaminreinhardt.com/Shapley_values_-_Better_than_counterfactuals)

[Comment on this note](http://via.hypothes.is/http://notes.benjaminreinhardt.com/Shapley_values_-_Better_than_counterfactuals)




Ben's working notes

How do you find counterfactuals?

Links to this note