Line of inquiry
Inquiring lines›How do training data and architect…›How do systems prioritize meaningf…›this line of inquiry
Why do language models systematically fail at causal reasoning?
A broader line of inquiry — a family of 54 specific questions the research asks around this. Follow one into its inquiring-line page, or move sideways to a related line below.
Questions in this line of inquiry 54
Specific inquiring lines the field asks around this — ordered from the most general framing down to the most specific angle.
- How does semantic association differ from mechanistic causal reasoning?
- Can activation patching reveal which reasoning steps actually matter?
- Can external actions provide causal necessity that language models lack?
- Why do causal reasoning directions succeed while temporal reasoning directions fail?
- Do LLMs show stronger reasoning about causality than about temporal ordering?
- How might human-LLM teams reinforce each other's causal reasoning mistakes?
- What are collider structures and why do they reveal reasoning errors?
- Why do LLMs inherit causal biases from their training data?
- Can interventions on model components prove mechanism without explaining encoding?
- Can statistical correlations in LLMs function equivalently to human causal understanding?
- Can models be trained to hide causal influences in their explanations?
- Can a Reflect mechanism detect and revise failed causal predictions?
- What makes counterfactual thinking different from behavioral pattern matching?
- Do causal rules enforce robustness that statistical patterns alone cannot maintain?
- When does prediction without causal mechanism fail to support good decisions?
- Is reward propagation in RL formally dual to cause inference in memory?
- What architectural changes would help LLMs distinguish causal relationships from temporal sequences?
- Can LLMs distinguish causal mapping from superficial similarity matching?
- Does causal intervention alone explain how neural mechanisms implement representations?
- Can steering vectors prove that representations are genuinely organized?
- Can causal models be extended to include non-causal cognition?
- Can a conversational role attribution direction in representation space causally gate correction?
- Do reading vectors from activation space causally control model behavior?
- How do causal belief networks extracted from interviews enable intervention reasoning?
- What training architecture models the causal structure of partner influence?
- Why does LLM compression eliminate causal grounding in conceptual representations?
- Are traditional cognitive theories missing interaction effects between mechanisms?
- Why do causal graphs alone fail to capture human reasoning processes?
- Can causal steering change verbalization without changing internal representation?
- How does this motivational bias connect to LLMs' causal reasoning failures?
- Can activation steering causally control evaluation framing effects across downstream tasks?
- What makes causal explanations stronger anxiety predictors than counterfactuals or dissonance?
- How do humans use associative reasoning without causal connections?
- Can causal belief networks extracted from interviews predict how people respond to policy changes?
- How does causal structure avoid behaviorist limitations in LLM social simulation?
- What architectural features enable counterfactual reasoning in world models?
- Does information stored in neural networks necessarily influence generation decisions?
- What computational methods most reliably establish causal evidence in AI mechanism discovery?
- Can neural networks learn that A implies B in reverse?
- Why do steered pain directions disable consequence-weighting but preserve factual competence?
- What makes causal belief networks more auditable than prompted personas?
- How does physical proximity to harm affect causal blame assignment?
- How can extracted causal belief networks enable intervention simulation?
- How do perception and generation share timing in a single causal stream?
- Do concept connections in speech reveal anxiety better than individual word choices?
- What is the accuracy cost of enforcing temporal causality inside model parameters?
- Why does masking future experts guarantee causal validity without external verification?
- What role does prediction error play in human event segmentation?
- Why do attention circuits need causal verification beyond feature visualization?
- How does flip-event regression differ from premature thought path abandonment?
- Can belief networks from interviews simulate how people change their minds?
- Does the reversal curse stem from the same one-way commitment architecture?
- Why should low-probability severe risks trigger early intervention?
- Where do collider-type reasoning errors appear in real-world decisions?