Personal experience and domain knowledge

We started this chapter by talking about how babies perform experiments to build causal world models. In this section, we’ll look into an adult’s approach to refining and building such models.

Imagine a rainy chilly afternoon somewhere in Northern Europe. You stand at a bus stop near your favorite park. There’s a large puddle on the street in front of you. You notice a car approaching from the left, driving close to the sidewalk. It seems that it will drive straight into the puddle. As the car approaches the puddle without slowing down, you instinctively jump behind the bus stop’s shelter. Water hits the shelter’s glass right next to your face, but fortunately, you are safe on the other side.

Your reaction was likely a result of many different factors, including a fast instinctive response to a threatening stimulus (splashing muddy water), but it was likely not entirely instinctive. You noticed a car...

Causal structure learning

The last source of causal knowledge that we will discuss in this chapter is causal structure learning. Causal structure learning (sometimes used interchangeably with causal discovery) is a set of methods aiming at recovering the structure of the data-generating process from the data generated by this process. Traditional causal discovery focused on recovering the causal structure from observational data only.

Some more recent methods allow for encoding expert knowledge into the graph or learning from interventional data (with known or unknown interventions).

Causal structure learning might be much cheaper and faster than running an experiment, but it often turns out to be challenging in practice.

Many causal structure learning methods require no hidden confounding – a condition difficult to guarantee in numerous real-world scenarios. Some causal discovery methods try to overcome this limitation with some success.

Another challenge is scalability...

Wrapping it up

In this chapter, we discussed three broad sources of causal knowledge: scientific insights, personal experiences and domain knowledge, and causal structure learning.

We saw that humans start to work on building world models very early in development; yet not all world models that we build are accurate. Heuristics that we use introduce biases that can skew our models on an individual, organizational, or cultural level.

Scientific experiments are an attempt to structure the process of obtaining knowledge so that we can exclude or minimize unwanted interferences and sources of distortion.

Unfortunately, experiments are not always available and have their own limitations. Causal structure learning methods can be cheaper and faster than running experiments, but they might rely on assumptions difficult to meet in certain scenarios.

Hybrid methods that combine causal structure learning, domain expertise, and efficient experimentation are a new exciting field of...

References

Gopnik, A. (2009). The philosophical baby: What children’s minds tell us about truth, love, and the meaning of life. Farrar, Straus and Giroux.

Hall N. S. (2007). R. A. Fisher and his advocacy of randomization. Journal of the History of Biology, 40(2), 295–325.

Harrell, F. (2023, February 14). Randomized Clinical Trials Do Not Mimic Clinical Practice, Thank Goodness. Statistical Thinking. https://www.fharrell.com/post/rct-mimic/

Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.

Kostis, J. B., & Dobrzynski, J. M. (2020). Limitations of Randomized Clinical Trials. The American Journal of Cardiology, 129, 109–115.

Kuhn, T. S. (1962). The structure of scientific revolutions. University of Chicago Press.

Martín, F. M. (2009). The thermodynamics of human reaction times. arXiv, abs/0908.3170.

Muenssinger, J., Matuz, T., Schleger, F., Kiefer-Schmidt, I., Goelz, R., Wacker-Gussmann, A., Birbaumer, N., &...

Causal structure learning

The last source of causal knowledge that we will discuss in this chapter is causal structure learning. Causal structure learning (sometimes used interchangeably with causal discovery) is a set of methods aiming at recovering the structure of the data-generating process from the data generated by this process. Traditional causal discovery focused on recovering the causal structure from observational data only.

Some more recent methods allow for encoding expert knowledge into the graph or learning from interventional data (with known or unknown interventions).

Causal structure learning might be much cheaper and faster than running an experiment, but it often turns out to be challenging in practice.

Many causal structure learning methods require no hidden confounding – a condition difficult to guarantee in numerous real-world scenarios. Some causal discovery methods try to overcome this limitation with some success.

Another challenge is scalability...

Wrapping it up

In this chapter, we discussed three broad sources of causal knowledge: scientific insights, personal experiences and domain knowledge, and causal structure learning.

We saw that humans start to work on building world models very early in development; yet not all world models that we build are accurate. Heuristics that we use introduce biases that can skew our models on an individual, organizational, or cultural level.

Scientific experiments are an attempt to structure the process of obtaining knowledge so that we can exclude or minimize unwanted interferences and sources of distortion.

Unfortunately, experiments are not always available and have their own limitations. Causal structure learning methods can be cheaper and faster than running experiments, but they might rely on assumptions difficult to meet in certain scenarios.

Hybrid methods that combine causal structure learning, domain expertise, and efficient experimentation are a new exciting field of...

References

Gopnik, A. (2009). The philosophical baby: What children’s minds tell us about truth, love, and the meaning of life. Farrar, Straus and Giroux.

Hall N. S. (2007). R. A. Fisher and his advocacy of randomization. Journal of the History of Biology, 40(2), 295–325.

Harrell, F. (2023, February 14). Randomized Clinical Trials Do Not Mimic Clinical Practice, Thank Goodness. Statistical Thinking. https://www.fharrell.com/post/rct-mimic/

Kahneman, D. (2011). Thinking, fast and slow. Farrar, Straus and Giroux.

Kostis, J. B., & Dobrzynski, J. M. (2020). Limitations of Randomized Clinical Trials. The American Journal of Cardiology, 129, 109–115.

Kuhn, T. S. (1962). The structure of scientific revolutions. University of Chicago Press.

Martín, F. M. (2009). The thermodynamics of human reaction times. arXiv, abs/0908.3170.

Muenssinger, J., Matuz, T., Schleger, F., Kiefer-Schmidt, I., Goelz, R., Wacker-Gussmann, A., Birbaumer, N., &...

Join our book's Discord space

Join our Discord community to meet like-minded people and learn alongside more than 2000 members at: https://packt.link/infer