Researchers measure 'people trust machines more just because they're machines' — but does that survey question really capture the belief?
What does machine heuristic actually measure when operationalized as a construct?
This explores what the 'machine heuristic' (the mental shortcut where people assume a machine's output is more objective, accurate, or unbiased because a machine produced it) actually captures once researchers turn it into something they can measure, and whether those measures track that belief or something nearby.
This explores what the 'machine heuristic' captures once researchers try to measure it. The term means the shortcut where people trust an answer more because a machine produced it, assuming machines are neutral and precise. The direct answer is that this collection does not cover the topic. None of the retrieved notes deal with the machine heuristic, with how human-computer interaction (HCI) researchers measure it, or with the survey scales and experiment designs used to capture it. The questions you would want answered here aren't covered: is it a stable trait or a reaction to the situation, is it the same thing as general trust in automation, and does it hold up when the interface changes? That gap is a useful finding in itself. The library is strong on how models behave and thin on how people perceive them.
The corpus does have a related thread that sits under your question: a measurement can look like it captures one thing while it actually captures another. In AI reasoning research this keeps happening. Longer reasoning traces look like a measure of problem difficulty. Under controlled tests, though, trace length mostly reflects how close a problem is to what the model saw in training Does longer reasoning actually mean harder problems?. Chain-of-thought gains look like evidence of logical reasoning, but deliberately illogical example reasoning works almost as well. The format of reasoning is doing the work, not its validity Does logical validity actually drive chain-of-thought gains?.
The more useful lesson for anyone studying a construct like the machine heuristic is how one study separated tangled causes. Researchers split chain-of-thought performance into three separate factors: how likely the answer is as output, memorization, and genuine but error-prone reasoning What three separate factors drive chain-of-thought performance?. A similar split is the obvious challenge for machine-heuristic research. Is the measure picking up a belief that machines are objective, general comfort with technology, how fluent and confident the output sounds, or simply the label 'AI-generated'? These can move together in a single survey score while being very different things. Agent evaluation research makes a parallel point: judging only final outcomes hides the process that produced them How should we evaluate agent behavior beyond final answers?.
If you want the human-perception side of this question, the collection can't take you there yet. The reasoning-research notes above are worth reading as a cautionary pattern. In each case a plausible measure turned out to track something else once someone controlled for the alternatives.
Sources 4 notes
Controlled A* maze experiments show trace length correlates with difficulty only in-distribution but decouples entirely out-of-distribution. Trace length primarily reflects recall of training schemas, not adaptive computation.
Illogical chain-of-thought exemplars matched valid CoT performance on BIG-Bench Hard, showing that structural properties—not logical validity—drive the gains. The model learns the form of reasoning, not genuine inference.
A shift cipher study decomposed CoT into three independent factors: output probability alone swings accuracy from 26% to 70%, memorization matches pre-training frequency patterns, and genuine reasoning exists but accumulates error with each step. This resolves the reason-or-memorize debate by showing LLMs do both simultaneously.
Evaluation of agentic systems shifts evidence from final responses to full interaction sequences, and scoring procedure from correctness alone to process quality, recoverability, coordination, and robustness. This pattern appears across multiple agent benchmarks as a coherent design move.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- When More is Less: Understanding Chain-of-Thought Length in LLMs
- Think Deep, Not Just Long: Measuring LLM Reasoning Effort via Deep-Thinking Tokens
- Deciphering the Factors Influencing the Efficacy of Chain-of-Thought: Probability, Memorization, and Noisy Reasoning
- Is Chain-of-Thought Reasoning of LLMs a Mirage? A Data Distribution Lens
- Invalid Logic, Equivalent Gains: The Bizarreness of Reasoning in Language Model Prompting
- Performative Thinking? The Brittle Correlation Between CoT Length and Problem Complexity
- Break the Chain: Large Language Models Can be Shortcut Reasoners
- Agent-as-a-Judge: Evaluate Agents with Agents