Is AI changing confidence in our own thinking?
AI can help people clarify ideas, test assumptions and feel more certain about what to do. It can also increase second-guessing, create a need for repeated confirmation and make unaided judgement feel less reliable. The Human Clarity Institute is developing an evidence-based benchmark to examine how AI use relates to confidence, self-trust and the ability to think and decide independently.
This benchmark is currently being prepared
The AI Self-Trust Benchmark is part of HCI's expanding AI Behaviour Benchmarks library. This page has been published early to document the planned scope of the benchmark while analysis of HCI's confidence, second-guessing, judgement and decision-making datasets continues.
Built from Human Clarity Institute Behavioural Datasets
The AI Self-Trust Benchmark is being developed using Human Clarity Institute population datasets examining decision confidence, second-guessing, reliance, perceived control, comfort deciding without AI and confidence in personal judgement.
These datasets make it possible to distinguish between AI that strengthens a person's thinking and AI that becomes a source of reassurance they increasingly feel unable to do without. Two people may both consult AI frequently while experiencing very different effects on confidence, independence and authorship.
These same datasets also underpin the Human Clarity Institute AI Identity & Behaviour Assessment, allowing individuals to compare their own patterns with population benchmark data.
Original datasets
HCI measures confidence, second-guessing, reliance, comfort without AI, perceived control and related behavioural dimensions.
Population benchmarks
The data is analysed to identify reference points, common patterns and differences across levels and contexts of AI use.
Personal comparison
The assessment compares an individual's responses with relevant HCI benchmark populations and AI usage groups.
What This Benchmark Will Measure
This benchmark will examine whether AI use is strengthening confidence in personal thinking or changing how much reassurance people need before they trust their own judgement.
The final benchmark will help distinguish healthy reflection and useful confirmation from patterns in which confidence becomes increasingly dependent on AI approval.
Why Self-Trust Matters
Self-trust is not the same as certainty. It is the ability to form a judgement, recognise its limits and still act without needing endless external confirmation. AI can support this process by helping people organise evidence, challenge assumptions and see alternatives.
But AI can also become a highly available source of reassurance. When every opinion, decision or interpretation can be checked instantly, people may begin to feel less comfortable relying on their own reasoning. The result may not be obvious dependence. It may appear as repeated prompting, hesitation, difficulty finalising a decision or feeling that an answer is only valid once AI agrees.
High AI use does not automatically reduce self-trust. A person may use AI intensively while remaining confident, reflective and willing to disagree. Another may use it less often but rely on it primarily for validation. The role AI plays matters more than frequency alone.
The completed benchmark will examine how common these patterns are and how they relate to reliance, trust, decision-making, verification and retained human agency.
Questions the Benchmark Will Answer
Is AI increasing second-guessing?
The completed benchmark will establish reference points for how often people revisit decisions after consulting AI.
Do people need AI confirmation?
It will examine when AI is used for useful reflection and when reassurance becomes difficult to do without.
Does AI strengthen confidence?
The benchmark will compare confidence gains with retained independence, judgement and willingness to act unaided.
Can people still disagree with AI?
It will examine whether users remain confident when their own view conflicts with an AI-generated answer.
Are heavy users less comfortable without AI?
The final analysis will compare confidence and unaided decision comfort across different levels of AI use.
What does healthy self-trust look like?
HCI will examine confidence alongside openness to evidence, verification, independence and retained human agency.
How Self-Trust Fits Within the Wider HCI Framework
Self-trust connects several behavioural dimensions at once. Trust affects how much weight people give AI. Reliance affects how quickly they seek reassurance. Verification affects whether they independently test an answer. Decision-making reveals whether confidence survives uncertainty, and human agency determines who remains the author of the final judgement.
HCI therefore examines self-trust as a connected behavioural pattern. The aim is not to encourage blind confidence, but to understand whether people retain a stable relationship with their own reasoning while using AI as a source of support.
Trust
How much authority people give AI-generated answers relative to their own knowledge and judgement.
Reliance
Whether people increasingly need AI reassurance before they feel ready to decide or act.
Verification
Whether confidence is grounded in independent checking rather than simple agreement from the system.
Decision-Making
How AI affects certainty, hesitation, second-guessing and comfort with personal choices.
Thought Partnership
Whether AI helps clarify and develop thinking without becoming the final authority on what is true or right.
Human Agency
Whether the user remains the author of the judgement and can act without repeated AI confirmation.
Explore Available Benchmarks
While the AI Self-Trust Benchmark is being prepared, the following HCI benchmarks are available now.
AI Behaviour Benchmarks
HCI's benchmark library examines distinct but connected parts of the human relationship with AI. Completed benchmarks are marked as available; the remaining benchmarks are currently being prepared.
How HCI Builds AI Behaviour Benchmarks
The completed AI Self-Trust Benchmark will be developed using HCI's three-evidence methodology. This approach is designed to distinguish original population findings from wider research while combining both into a clear behavioural reference point.
Original HCI datasets
Population research measuring confidence, second-guessing, comfort without AI, reliance, control and personal judgement.
Wider research
Relevant peer-reviewed studies, institutional reports and established research on confidence, metacognition and AI-assisted judgement.
Cross-benchmark analysis
Comparison across trust, reliance, verification, decision-making, thought partnership and human agency.
Learn more about the full process in How Human Clarity Institute Builds AI Behaviour Benchmarks.
Compare Your AI Behaviour
The Human Clarity Institute AI Identity & Behaviour Assessment uses the same behavioural datasets that underpin HCI's benchmark library.
Rather than comparing people with arbitrary scores, the assessment is designed to compare individual responses with population benchmark data across thought partnership, reliance, verification, trust, decision-making, confidence and other dimensions of human adaptation to AI.
The assessment is currently being prepared for public launch.