Why might the finding that people treat computers politely fail to replicate — for reasons that have nothing to do with computers just feeling less new?

This explores why a replication of the 'Computers Are Social Actors' (CASA) finding might fail for reasons other than computers simply no longer being new to people. The corpus doesn't describe the failed replication itself, so it can't say what went wrong in that study. What it does offer is a sharp reason to doubt that novelty explains anything at all. The usual account says people are polite to computers 'mindlessly': on autopilot, out of social habit. Lee shows that most CASA research never measures mindlessness directly. It infers mindlessness from the very social behavior it claims to explain Can mindlessness in CASA effects be measured independently?. That loop matters for your question. The novelty story usually says people were once caught off guard by machines and now aren't, so the autopilot no longer fires. Because the autopilot was never measured, nobody can tell whether it faded, whether it was ever the cause, or whether something else changed.

That opens up several explanations novelty can't rule out. One is that the original effect may not have been driven by mindlessness at all. Without a separate measure of that mechanism, a replication can't tell whether it was testing the same cause in different conditions or a cause that was never there. Another is a measurement problem. Lee's critique implies the study design can't separate the cause from the outcome, so when results fail to replicate, it's unclear whether the effect changed or the instrument just can't detect it. A third is that 'novelty' bundles several things together: familiarity, expectations about how machines behave, and what counts as a 'computer' today.

The corpus offers one useful analogy from elsewhere, and it is only an analogy. In AI-text detection, classifiers that caught almost every rewrite just before a model release caught only 3.8% afterward. The detectors didn't get worse; the thing they were measuring changed underneath them Why do AI detectors suddenly fail when models update?. CASA may face something similar. A 1990s desktop and a modern conversational AI are different stimuli, so a 'replication' may unknowingly test a different object. That isn't the same as people getting used to computers. It's a mismatch between the original setup and the new one. A separate lesson from AI training research also applies: two effects that look like one thing can sit at different levels of measurement and come apart without contradicting each other Can genuine reasoning activation coexist with contaminated benchmarks?. Social responses to machines and the presumed mental cause behind them may be just such a pair.

The surprising takeaway is that a failed CASA replication may say less about how people have changed than about the theory itself. If mindlessness was never independently testable, the replication couldn't really disconfirm it, and the original findings never really confirmed it either. For a direct account of the specific replication, you'd need sources this corpus doesn't hold.