A known cognitive effect with a research history.
AI-era Bias Lab
Study the loop between human judgment and AI behaviour.
AI can introduce a first answer, mirror a user, blur source memory, change mental effort and transfer confidence back into the human decision. This lab tracks those interaction patterns without pretending every useful name is already a validated cognitive bias.
Three evidence levels
Name carefully. Test aggressively.
Important parts of the human–AI loop have direct evidence, but the combined label is still a working frame.
A falsifiable idea that stays a hypothesis until controlled studies support it.
Why this matters
The unit of study is the loop.
A person asks a leading question. A model adapts. The answer becomes new evidence in the person's mind. The next prompt becomes stronger. Studying that feedback loop can be more useful than assigning a long list of labels to either side.
Current research tracks
Working ideas with explicit status and predictions.
emerging interaction pattern
AI Anchoring Loop
A model gives a first estimate or recommendation, and the human's later reasoning stays too close to that starting point.
Related concepts: anchoring effect, automation bias.
Testable prediction: Participants who form an independent estimate before seeing AI advice should show a smaller shift toward an arbitrary AI anchor than participants who see AI advice first.
Starting evidence
emerging interaction pattern
Sycophancy Reinforcement Loop
A user signals a preferred conclusion, the model adapts toward it, and the agreeable answer then feels like independent support for the user's original belief.
Related concepts: confirmation bias, conformity, motivated reasoning.
Testable prediction: Agreement-seeking prompts should increase both model agreement and user confidence more than neutral prompts, especially when the user is not shown an independent counterargument.
Starting evidence
testable hypothesis
Cognitive Offloading Debt
AI can improve immediate task completion while reducing the effort that builds durable memory, retrieval skill or an internal model of the problem.
Related concepts: cognitive offloading, testing effect, illusion of explanatory depth.
Testable prediction: AI-assisted participants may complete a learning task faster but perform worse on delayed unaided recall or transfer unless the workflow includes retrieval and explanation steps.
Starting evidence
emerging interaction pattern
Source-Memory Blur
In mixed human-AI work, people may remember an idea or sentence but become less certain whether it came from themselves, a source document or the model.
Related concepts: source confusion, cryptomnesia.
Testable prediction: Mixed workflows where AI and human contributions alternate should produce more source-attribution errors after a delay than clearly separated human-only and AI-only phases.
Starting evidence
testable hypothesis
Synthetic Consensus Illusion
Many generated answers can look like independent agreement even when they inherit the same sources, training patterns or retrieval result.
Related concepts: illusory truth effect, common source bias, availability cascade.
Testable prediction: People should rate a claim as better supported when it appears in several differently worded AI answers, even when all answers are explicitly derived from one original source; source-lineage disclosure should reduce the effect.
Starting evidence
- No direct source assigned yet. Treat this as a hypothesis, not a finding.
testable hypothesis
Confidence Transfer
Confident model language may transfer confidence to a human decision even when the underlying evidence is weak.
Related concepts: automation bias, overconfidence, authority bias.
Testable prediction: For the same answer quality, high-certainty AI wording should increase user confidence and reliance more than calibrated wording; showing objective model accuracy should moderate the effect.
Starting evidence
How to propose a new pattern
A name comes last.
Describe the repeated shift without naming it.
Check whether an established effect already explains it.
Write a result that would differ if the pattern is real.
Change one important variable and use a control condition.
Repeat across people, tasks, models and settings.
Add a label only if it improves understanding.