Position: The Causal Revolution Needs Scientific Pragmatism

Joshua Loftus

2024-06-04

Abstract:Causal models and methods have great promise, but their progress has been stalled. Proposals using causality get squeezed between two opposing worldviews. Scientific perfectionism--an insistence on only using "correct" models--slows the adoption of causal methods in knowledge generating applications. Pushing in the opposite direction, the academic discipline of computer science prefers algorithms with no or few assumptions, and technologies based on automation and scalability are often selected for economic and business applications. We argue that these system-centric inductive biases should be replaced with a human-centric philosophy we refer to as scientific pragmatism. The machine learning community must strike the right balance to make space for the causal revolution to prosper.

Computers and Society

What problem does this paper attempt to address?

This paper discusses the challenges of applying causal revolution in science and proposes the need to adopt a philosophy of scientific pragmatism to promote the development of causal models. The paper points out that causal models, based on predictive models, add assumptions about intervention and directionality, enabling them to answer important questions that predictive models cannot answer, such as how to change a variable. However, causal models are susceptible to criticism due to additional assumptions and, in certain fields such as computer science, tend to favor algorithms with fewer or no assumptions.

The paper presents several viewpoints:

1. For arguments against causal models, it is important to examine whether these arguments also apply to equivalent predictive models.
2. Scientists and people often desire causal conclusions, even when they are using predictive models.
3. Both predictive models and causal models should be evaluated by balanced criteria, without substituting one for the other because of lower standards for predictive models, and without excessive criticism due to additional assumptions of causal models.
4. Causal models have more falsifiability than predictive models, which is beneficial for scientific progress.
5. The preference for predictive models is sometimes a way to evade responsibility; authors don't explicitly state causal assumptions and leave causal conclusions to be inferred by readers.
6. The machine learning community has the potential to integrate causal models but also has a tendency to avoid them, favoring more general and automated algorithms.

The paper emphasizes the importance of scientific pragmatism, where theories and models should be seen as tools whose applicability depends on the purpose of use. The scientific community needs to find a balance between pursuing perfect models and practical applications to foster the prosperity of the causal revolution. The paper also mentions the diversity of values and the human-centered perspective, highlighting the central role of humans in decision-making and value judgments.