Lesson overview | Previous part | Lesson overview

Structural Causal Models: Part 6: ML and LLM Applications to References

6. ML and LLM Applications

ML and LLM Applications develops the part of structural causal models specified by the approved Chapter 22 table of contents. The treatment is causal, not merely predictive: the central objects are mechanisms, interventions, assumptions, and counterfactuals.

6.1 causal representation learning

Causal representation learning belongs to the canonical scope of Structural Causal Models. The central move in causal inference is to distinguish a statistical relation from a claim about what would happen under an intervention.

For this subsection, the working scope is structural assignments, causal graphs, d-separation, interventions, Markovian assumptions, and SCM links to robust ML. The mathematical objects are variables, mechanisms, graphs, interventions, and assumptions. A causal claim is incomplete until all five are visible.

$$V_i=f_i(\operatorname{pa}_i,U_i).$$

The formula gives a compact handle on causal representation learning. It should not be read as a purely algebraic identity. In causal inference, equations encode assumptions about mechanisms, missing variables, and which parts of the world remain stable under intervention.

Three examples of causal representation learning:

1. A recommender team wants the causal effect of ranking a document higher, not merely the correlation between rank and clicks.
2. An LLM platform changes a safety policy and wants to estimate whether refusals changed because of the policy or because user prompts shifted.
3. A fairness auditor asks whether a proxy feature transmits an impermissible causal path into a model decision.

Two non-examples expose the boundary:

1. A high predictive coefficient is not a causal effect unless the graph and intervention assumptions justify it.
2. A plausible narrative produced by a language model is not a counterfactual unless it is grounded in a causal model.

The proof habit for causal representation learning is to name the graph operation. Conditioning restricts a distribution. Intervention replaces a mechanism. Counterfactual reasoning updates exogenous uncertainty from evidence, changes a mechanism, then predicts.

observed association:      P(Y | X=x)
intervention question:     P(Y | do(X=x))
counterfactual question:   P(Y_x | E=e)
discovery question:        which G could have generated P(V)?

In machine learning, causal representation learning is valuable because models are often deployed under interventions: ranking changes, policy changes, safety filters, tool-use gates, data collection changes, and human feedback loops. Prediction alone does not tell us which change caused which downstream behavior.

Notebook implementation will use synthetic SCMs and small graphs. This keeps the examples executable while preserving the conceptual split between identification and estimation.

Checklist for using causal representation learning responsibly:

• State the causal question before choosing a method.
• Draw or describe the assumed causal graph.
• Mark observed, latent, treatment, outcome, and adjustment variables.
• Separate intervention notation from conditioning notation.
• Decide whether the query is identifiable before estimating it.
• Report assumptions that cannot be tested from the observed data alone.
• Use ML as an estimation aid, not as a substitute for causal design.

This chapter follows the boundary set by Chapter 21. Statistical learning theory controls prediction error under distributional assumptions. Causal inference asks what happens when the distribution changes because something is done.

Modern AI systems make this distinction unavoidable. A foundation model can predict which action historically followed a context, but a decision system needs to know what would happen if it took a different action in that context.

Thus, causal representation learning is not an abstract philosophical add-on. It is a production and research tool for deciding which model, prompt, policy, feature, or intervention actually changed an outcome.

A final diagnostic question is whether the claim would survive a policy change. If the answer depends only on a historical correlation, it belongs in predictive modeling. If the answer depends on what mechanism is replaced and which paths remain active, it belongs in causal inference.

6.2 robust prediction under shift

Robust prediction under shift belongs to the canonical scope of Structural Causal Models. The central move in causal inference is to distinguish a statistical relation from a claim about what would happen under an intervention.

For this subsection, the working scope is structural assignments, causal graphs, d-separation, interventions, Markovian assumptions, and SCM links to robust ML. The mathematical objects are variables, mechanisms, graphs, interventions, and assumptions. A causal claim is incomplete until all five are visible.

$$P(\mathbf{v})=\prod_{i=1}^{d}P(v_i \mid \operatorname{pa}_i).$$

The formula gives a compact handle on robust prediction under shift. It should not be read as a purely algebraic identity. In causal inference, equations encode assumptions about mechanisms, missing variables, and which parts of the world remain stable under intervention.

Three examples of robust prediction under shift:

1. A recommender team wants the causal effect of ranking a document higher, not merely the correlation between rank and clicks.
2. An LLM platform changes a safety policy and wants to estimate whether refusals changed because of the policy or because user prompts shifted.
3. A fairness auditor asks whether a proxy feature transmits an impermissible causal path into a model decision.

Two non-examples expose the boundary:

1. A high predictive coefficient is not a causal effect unless the graph and intervention assumptions justify it.
2. A plausible narrative produced by a language model is not a counterfactual unless it is grounded in a causal model.

The proof habit for robust prediction under shift is to name the graph operation. Conditioning restricts a distribution. Intervention replaces a mechanism. Counterfactual reasoning updates exogenous uncertainty from evidence, changes a mechanism, then predicts.

observed association:      P(Y | X=x)
intervention question:     P(Y | do(X=x))
counterfactual question:   P(Y_x | E=e)
discovery question:        which G could have generated P(V)?

In machine learning, robust prediction under shift is valuable because models are often deployed under interventions: ranking changes, policy changes, safety filters, tool-use gates, data collection changes, and human feedback loops. Prediction alone does not tell us which change caused which downstream behavior.

Notebook implementation will use synthetic SCMs and small graphs. This keeps the examples executable while preserving the conceptual split between identification and estimation.

Checklist for using robust prediction under shift responsibly:

• State the causal question before choosing a method.
• Draw or describe the assumed causal graph.
• Mark observed, latent, treatment, outcome, and adjustment variables.
• Separate intervention notation from conditioning notation.
• Decide whether the query is identifiable before estimating it.
• Report assumptions that cannot be tested from the observed data alone.
• Use ML as an estimation aid, not as a substitute for causal design.

This chapter follows the boundary set by Chapter 21. Statistical learning theory controls prediction error under distributional assumptions. Causal inference asks what happens when the distribution changes because something is done.

Modern AI systems make this distinction unavoidable. A foundation model can predict which action historically followed a context, but a decision system needs to know what would happen if it took a different action in that context.

Thus, robust prediction under shift is not an abstract philosophical add-on. It is a production and research tool for deciding which model, prompt, policy, feature, or intervention actually changed an outcome.

A final diagnostic question is whether the claim would survive a policy change. If the answer depends only on a historical correlation, it belongs in predictive modeling. If the answer depends on what mechanism is replaced and which paths remain active, it belongs in causal inference.

6.3 recommender interventions

Recommender interventions belongs to the canonical scope of Structural Causal Models. The central move in causal inference is to distinguish a statistical relation from a claim about what would happen under an intervention.

For this subsection, the working scope is structural assignments, causal graphs, d-separation, interventions, Markovian assumptions, and SCM links to robust ML. The mathematical objects are variables, mechanisms, graphs, interventions, and assumptions. A causal claim is incomplete until all five are visible.

$$P_M(Y \mid \operatorname{do}(X=x))=P_{M_x}(Y).$$

The formula gives a compact handle on recommender interventions. It should not be read as a purely algebraic identity. In causal inference, equations encode assumptions about mechanisms, missing variables, and which parts of the world remain stable under intervention.

Three examples of recommender interventions:

1. A recommender team wants the causal effect of ranking a document higher, not merely the correlation between rank and clicks.
2. An LLM platform changes a safety policy and wants to estimate whether refusals changed because of the policy or because user prompts shifted.
3. A fairness auditor asks whether a proxy feature transmits an impermissible causal path into a model decision.

Two non-examples expose the boundary:

1. A high predictive coefficient is not a causal effect unless the graph and intervention assumptions justify it.
2. A plausible narrative produced by a language model is not a counterfactual unless it is grounded in a causal model.

The proof habit for recommender interventions is to name the graph operation. Conditioning restricts a distribution. Intervention replaces a mechanism. Counterfactual reasoning updates exogenous uncertainty from evidence, changes a mechanism, then predicts.

observed association:      P(Y | X=x)
intervention question:     P(Y | do(X=x))
counterfactual question:   P(Y_x | E=e)
discovery question:        which G could have generated P(V)?

In machine learning, recommender interventions is valuable because models are often deployed under interventions: ranking changes, policy changes, safety filters, tool-use gates, data collection changes, and human feedback loops. Prediction alone does not tell us which change caused which downstream behavior.

Notebook implementation will use synthetic SCMs and small graphs. This keeps the examples executable while preserving the conceptual split between identification and estimation.

Checklist for using recommender interventions responsibly:

• State the causal question before choosing a method.
• Draw or describe the assumed causal graph.
• Mark observed, latent, treatment, outcome, and adjustment variables.
• Separate intervention notation from conditioning notation.
• Decide whether the query is identifiable before estimating it.
• Report assumptions that cannot be tested from the observed data alone.
• Use ML as an estimation aid, not as a substitute for causal design.

This chapter follows the boundary set by Chapter 21. Statistical learning theory controls prediction error under distributional assumptions. Causal inference asks what happens when the distribution changes because something is done.

Modern AI systems make this distinction unavoidable. A foundation model can predict which action historically followed a context, but a decision system needs to know what would happen if it took a different action in that context.

Thus, recommender interventions is not an abstract philosophical add-on. It is a production and research tool for deciding which model, prompt, policy, feature, or intervention actually changed an outcome.

A final diagnostic question is whether the claim would survive a policy change. If the answer depends only on a historical correlation, it belongs in predictive modeling. If the answer depends on what mechanism is replaced and which paths remain active, it belongs in causal inference.

6.4 fairness and proxy variables

Fairness and proxy variables belongs to the canonical scope of Structural Causal Models. The central move in causal inference is to distinguish a statistical relation from a claim about what would happen under an intervention.

For this subsection, the working scope is structural assignments, causal graphs, d-separation, interventions, Markovian assumptions, and SCM links to robust ML. The mathematical objects are variables, mechanisms, graphs, interventions, and assumptions. A causal claim is incomplete until all five are visible.

$$M=(\mathbf{U},\mathbf{V},\mathbf{F},P(\mathbf{U})).$$

The formula gives a compact handle on fairness and proxy variables. It should not be read as a purely algebraic identity. In causal inference, equations encode assumptions about mechanisms, missing variables, and which parts of the world remain stable under intervention.

Three examples of fairness and proxy variables:

1. A recommender team wants the causal effect of ranking a document higher, not merely the correlation between rank and clicks.
2. An LLM platform changes a safety policy and wants to estimate whether refusals changed because of the policy or because user prompts shifted.
3. A fairness auditor asks whether a proxy feature transmits an impermissible causal path into a model decision.

Two non-examples expose the boundary:

1. A high predictive coefficient is not a causal effect unless the graph and intervention assumptions justify it.
2. A plausible narrative produced by a language model is not a counterfactual unless it is grounded in a causal model.

The proof habit for fairness and proxy variables is to name the graph operation. Conditioning restricts a distribution. Intervention replaces a mechanism. Counterfactual reasoning updates exogenous uncertainty from evidence, changes a mechanism, then predicts.

observed association:      P(Y | X=x)
intervention question:     P(Y | do(X=x))
counterfactual question:   P(Y_x | E=e)
discovery question:        which G could have generated P(V)?

In machine learning, fairness and proxy variables is valuable because models are often deployed under interventions: ranking changes, policy changes, safety filters, tool-use gates, data collection changes, and human feedback loops. Prediction alone does not tell us which change caused which downstream behavior.

Notebook implementation will use synthetic SCMs and small graphs. This keeps the examples executable while preserving the conceptual split between identification and estimation.

Checklist for using fairness and proxy variables responsibly:

• State the causal question before choosing a method.
• Draw or describe the assumed causal graph.
• Mark observed, latent, treatment, outcome, and adjustment variables.
• Separate intervention notation from conditioning notation.
• Decide whether the query is identifiable before estimating it.
• Report assumptions that cannot be tested from the observed data alone.
• Use ML as an estimation aid, not as a substitute for causal design.

This chapter follows the boundary set by Chapter 21. Statistical learning theory controls prediction error under distributional assumptions. Causal inference asks what happens when the distribution changes because something is done.

Modern AI systems make this distinction unavoidable. A foundation model can predict which action historically followed a context, but a decision system needs to know what would happen if it took a different action in that context.

Thus, fairness and proxy variables is not an abstract philosophical add-on. It is a production and research tool for deciding which model, prompt, policy, feature, or intervention actually changed an outcome.

A final diagnostic question is whether the claim would survive a policy change. If the answer depends only on a historical correlation, it belongs in predictive modeling. If the answer depends on what mechanism is replaced and which paths remain active, it belongs in causal inference.

6.5 causal logging for LLM agents

Causal logging for llm agents belongs to the canonical scope of Structural Causal Models. The central move in causal inference is to distinguish a statistical relation from a claim about what would happen under an intervention.

For this subsection, the working scope is structural assignments, causal graphs, d-separation, interventions, Markovian assumptions, and SCM links to robust ML. The mathematical objects are variables, mechanisms, graphs, interventions, and assumptions. A causal claim is incomplete until all five are visible.

$$V_i=f_i(\operatorname{pa}_i,U_i).$$

The formula gives a compact handle on causal logging for llm agents. It should not be read as a purely algebraic identity. In causal inference, equations encode assumptions about mechanisms, missing variables, and which parts of the world remain stable under intervention.

Three examples of causal logging for llm agents:

1. A recommender team wants the causal effect of ranking a document higher, not merely the correlation between rank and clicks.
2. An LLM platform changes a safety policy and wants to estimate whether refusals changed because of the policy or because user prompts shifted.
3. A fairness auditor asks whether a proxy feature transmits an impermissible causal path into a model decision.

Two non-examples expose the boundary:

1. A high predictive coefficient is not a causal effect unless the graph and intervention assumptions justify it.
2. A plausible narrative produced by a language model is not a counterfactual unless it is grounded in a causal model.

The proof habit for causal logging for llm agents is to name the graph operation. Conditioning restricts a distribution. Intervention replaces a mechanism. Counterfactual reasoning updates exogenous uncertainty from evidence, changes a mechanism, then predicts.

observed association:      P(Y | X=x)
intervention question:     P(Y | do(X=x))
counterfactual question:   P(Y_x | E=e)
discovery question:        which G could have generated P(V)?

In machine learning, causal logging for llm agents is valuable because models are often deployed under interventions: ranking changes, policy changes, safety filters, tool-use gates, data collection changes, and human feedback loops. Prediction alone does not tell us which change caused which downstream behavior.

Notebook implementation will use synthetic SCMs and small graphs. This keeps the examples executable while preserving the conceptual split between identification and estimation.

Checklist for using causal logging for llm agents responsibly:

• State the causal question before choosing a method.
• Draw or describe the assumed causal graph.
• Mark observed, latent, treatment, outcome, and adjustment variables.
• Separate intervention notation from conditioning notation.
• Decide whether the query is identifiable before estimating it.
• Report assumptions that cannot be tested from the observed data alone.
• Use ML as an estimation aid, not as a substitute for causal design.

This chapter follows the boundary set by Chapter 21. Statistical learning theory controls prediction error under distributional assumptions. Causal inference asks what happens when the distribution changes because something is done.

Modern AI systems make this distinction unavoidable. A foundation model can predict which action historically followed a context, but a decision system needs to know what would happen if it took a different action in that context.

Thus, causal logging for llm agents is not an abstract philosophical add-on. It is a production and research tool for deciding which model, prompt, policy, feature, or intervention actually changed an outcome.

A final diagnostic question is whether the claim would survive a policy change. If the answer depends only on a historical correlation, it belongs in predictive modeling. If the answer depends on what mechanism is replaced and which paths remain active, it belongs in causal inference.

7. Common Mistakes

8. Exercises

1. (*) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

2. (*) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

3. (*) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

4. (**) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

5. (**) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

6. (**) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

7. (***) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

8. (***) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

9. (***) Work through a causal-inference task for structural causal models.

   • (a) State the causal query using intervention or counterfactual notation.
   • (b) Draw or describe the relevant graph and assumptions.
   • (c) Decide whether the estimand is identifiable from the available data.
   • (d) Give an estimator or diagnostic only after identification is clear.
   • (e) Explain the AI or LLM system implication.

10. (***) Work through a causal-inference task for structural causal models.

• (a) State the causal query using intervention or counterfactual notation.
• (b) Draw or describe the relevant graph and assumptions.
• (c) Decide whether the estimand is identifiable from the available data.
• (d) Give an estimator or diagnostic only after identification is clear.
• (e) Explain the AI or LLM system implication.

9. Why This Matters for AI

10. Conceptual Bridge

Structural Causal Models follows statistical learning theory because learning theory explains how observed samples support future prediction claims. Causal inference asks a different question: what happens when an action changes the system that generated those samples?

The backward bridge is risk and uncertainty. Chapter 21 provides language for finite-sample generalization. Chapter 22 adds intervention semantics, graph assumptions, and counterfactual worlds. A causal claim is not just a better prediction; it is a claim about a modified data-generating mechanism.

The forward bridge is game theory. Once multiple agents adapt to interventions, the causal question becomes strategic: actions change incentives, incentives change behavior, and behavior changes the causal system. Chapter 23 will study that interaction explicitly.

+--------------------------------------------------------------+
| Chapter 21: prediction under finite samples                  |
| Chapter 22: intervention, counterfactuals, causal discovery  |
| Chapter 23: strategic interaction and adversarial systems    |
+--------------------------------------------------------------+

References

• Pearl. Causality: Models, Reasoning, and Inference. https://www.cambridge.org/core/books/causality/6836DD2F4FD4A767DE97BBECDD1655F5
• Pearl. Causal inference in statistics: An overview. https://projecteuclid.org/journals/statistics-surveys/volume-3/issue-none/Causal-inference-in-statistics-An-overview/10.1214/09-SS057.pdf
• Peters, Janzing, and Scholkopf. Elements of Causal Inference. https://tile.loc.gov/storage-services/master/gdc/gdcebookspublic/20/20/71/97/58/2020719758/2020719758.pdf
• Pearl, Glymour, and Jewell. Causal Inference in Statistics: A Primer. https://www.wiley.com/en-us/Causal+Inference+in+Statistics%3A+A+Primer-p-9781119186847