Personality and Attitudes Toward AI: Research, Trust and Use Boundaries

That is why arguments about AI are rarely only technical arguments. Many disagreements sound like “you are optimistic and I am cautious,” but underneath they are usually about the order of three questions: Can it help me finish faster? Do I still own the final decision? Is it quietly weakening my originality, responsibility, or long-term capability?

FermatMind is less interested in whether you “like AI” and more interested in whether you know when you are treating it as a tool, when you are using it for companionship, and when you have quietly placed it in the role of judge. Personality measures may be associated with some attitudes in particular studies, but they do not determine these uses or justify assigning access, responsibility or risk from a type. For a particular task, ask what matters to you: finishing a repeated process, protecting an original idea, retaining responsibility, or obtaining a response without immediate social pressure. These are questions to answer from the situation, not patterns that a personality score establishes.

Stein and colleagues (2024) developed and examined the ATTARI-12 general AI-attitude scale in two studies, then studied associations in 298 US MTurk participants. In that study’s multivariable analysis, agreeableness was associated with more positive general AI attitudes, while conspiracy beliefs were associated with more negative attitudes. This sample-specific association does not establish causation, correct AI use, or how a particular person will behave in a particular task.

Separate five questions before naming an attitude

Acceptance means willingness to try or adopt a tool; trust concerns reliance on a particular output; dependence concerns whether use displaces essential checking, action or support; anxiety concerns distress about use or consequences; verification means checking against reliable evidence. These can vary independently. Frequent use is not proof of trust or dependence, and caution is not incapacity. Record the task, evidence checked and outcome instead of assigning a type.

2. The FermatMind Framework: Four Variables Behind Your AI Attitude

Instead of dividing people into “tech optimists” and “tech skeptics,” it is more useful to look at four default settings. These are editorial questions to ask about a specific task, not a validated personality model or an assignment of who should use AI.

Many conflicts around AI come from the fact that these variables have never been named. You may be talking about efficiency while someone else is talking about responsibility. You may mean exploration while someone else fears substitution. You may mean relief while someone else sees the long-term hollowing out of ability.

After naming the four variables, an “AI attitude” becomes a trust configuration: where am I willing to let AI enter, and where will I not let it enter?

3. Five AI User Profiles: Which One Sounds Like You?

These are not diagnoses or personality labels. They are editorial scenarios. A person can move between them as tasks, access, experience and consequences change; no validated dominant-profile scoring is claimed.

In learning, if you often ask AI to organize notes or create review outlines, you are probably close to the productivity or structure types. If you pour career hesitation into it and ask it to name emotions and options, you are closer to the emotional mirror. If you often ask it to choose for you, you are near the edge of decision substitution.

The key question is not which type is best. It is which type resembles you and also maps onto your most vulnerable point.

Before a prompt: protect inputs and define a stop

Use the minimum information needed. Remove names, contact details, identifiers and private organizational material; do not paste another person’s answers, records, messages or sensitive health, financial or relationship details without appropriate authorization. Describe a fictional or de-identified situation where possible, and check the tool’s current data settings.

Name the task, allowed input, consequence of an error, a human reviewer and a stop condition. For example: “Help outline a public presentation using these non-confidential points; I will verify every claim before sharing.” Stop when a response is distressing, pressures you toward an unsafe action, requests unnecessary sensitive data or becomes your sole basis for a consequential decision. You can stop or seek support without having a direct conversation with someone who threatens you.

4. The Same AI Amplifies Different Problems in Learning, Work, Writing, and Relationships

In Learning

AI is useful for compressing messy information into workable steps. It can list priorities, group themes, and provide first explanations. But if you only receive organized answers and rarely search, compare, or recombine information yourself, learning turns into transported understanding rather than generated understanding.

A simple test is whether you still raise your own questions, find counterexamples, and rewrite definitions. If not, AI is not saving you low-value labor; it is removing the cognitive actions you most need to practice.

In Work

At work, AI is strong as a pre-processing layer: gathering clues, listing comparisons, breaking down tasks, drafting copy, and summarizing meetings. But your career ceiling still depends on direction-setting, value ranking, trade-offs, and relationship handling. Those all involve consequences, and consequences cannot be outsourced.

If you start using AI to decide whether to switch roles, accept an offer, resign, or change direction, do not rush to accept its conclusion. Ask instead: what information did it add, what blind spots did it expose, and what must be verified in the real world?

In Writing

Writing is where AI most easily creates the illusion that a lot has already been done. It can produce a decent first draft quickly, but not necessarily your stance, your choices, or your voice. Many people do not lose writing ability all at once; they slowly lose the muscle for asking, “What am I actually trying to say?”

If a passage is smooth, complete, and correctly formatted, but you read it and feel “this does not sound like me,” that is often not successful polishing. It is partial replacement of authorship.

In Relationships

AI can help rehearse difficult words, sort out conflict, and provide continuous response when you do not want to be interrupted. Those are real benefits. But the core of a relationship is not making words rounder; it is whether you are willing to carry consequences back into real interaction.

If you increasingly talk to AI first and turn real interaction into the secondary version, ask whether you are using it to avoid uncertainty and vulnerability in actual relationships.

5. Judgments You Should Not Outsource to AI

• Do not outsource value ranking. What matters more to you and what deserves long-term investment must remain your definition.
• Do not outsource relational commitment. AI can help you sort through whether to break up, forgive, or begin, but it cannot commit for you.
• Do not outsource irreversible career decisions. Resigning, changing fields, moving, starting a company, or choosing a major requires real information and human feedback.
• Do not outsource high-risk professional judgment. Medical, legal, financial, and compliance questions require qualified professionals; AI can only prepare and organize.
• Do not outsource “who you are.” AI can name traits, but it cannot define your identity, life direction, or long-term mission.

A blunt version: AI is best for candidate answers, not final interpretive authority. It can shorten the path from confusion to clarity, but it should not become your only basis for moving from clarity to decision.

6. Four Common Misreadings

• Mistaking smoothness for correctness. A polished reply is not the same as reliable judgment.
• Mistaking frequent use for good fit. Using it often does not mean it belongs in that part of the process.
• Mistaking response for understanding. AI can generate contextual comfort without paying the relational cost of understanding.
• Mistaking efficiency gain for capability growth. You may deliver faster while not necessarily thinking better.

The problem is often not that AI “fooled” you, but that you gave up the middle step too early: I check, I rewrite, I decide. Mature use keeps that step.

How teams can discuss differences in AI attitudes

Calling cautious colleagues “resistant to change” and active users “tech people” quickly turns a risk discussion into a clash of identities. A better approach is to ask everyone about the same task: what data may be entered, what an error would cost, who reviews the output, when work must stop, and how an adopted suggestion will be recorded.

For example, a content team may allow AI to suggest headlines while checking citations and product facts against authoritative sources. A hiring team may use it to organize public job materials, but should not replace human judgment with personality labels or automated scores. A support team may draft responses, but must use controlled systems for accounts, payments, and personal results. These are task boundaries, not judgments about whose personality is more advanced.

Managers should not assign “AI owner” and “reviewer only” roles from personality tests. Digital literacy can be taught, and accountability for risk needs clear rules. Records of actual errors, rework, time saved, and user effects are more useful than comparing employees' attitudes toward AI.

The same person's view can change over time. A major error may lower trust; learning how to check results may ease anxiety. A change in attitude does not require a change in personality.

7. The FermatMind Boundary Template: Put AI Back in the Right Place

Add the most important question: after using AI this time, am I clearer, or only more relieved? Clarity is judgment upgrade. Relief may be temporary pain control. Both can matter, but they must not be confused.

8. This Week’s Action Card: Run an AI Use Audit

• Review five recent AI interactions from the last seven days. Write the scene, purpose, result, and whether uncertainty was actually reduced.
• Mark one moment when you most clearly treated AI as a substitute. Write what real-world verification you would need if you made the decision yourself.
• Take one question you would normally ask AI to answer, and change it to: “Please list what information I am still missing.”
• Remove one high-frequency, low-value AI habit, and use the saved time for one real-world validation step.

FermatMind is not trying to make you use AI less. The point is to preserve the main chain of judgment, responsibility, and long-term capability while you use it.

What to Do Next in FermatMind Tests

• If you already have Big Five or MBTI-style results, use them to ask questions about observed preferences. They do not assess digital literacy, establish a cognitive-ability profile, or determine AI access, responsibility or risk.
• If you already use AI for career decisions, stop asking “which is best” and check the real job environment, growth path, and risk constraints in the career library.
• If you rely on AI for emotional sorting more and more, build a real-person support list instead of expanding AI’s role in your relationship system.

Research Notes and References

The sources below discuss attitudes, advice or disclosure in their own study settings. They do not validate these editorial user patterns, assign risk by MBTI type, establish a disclosure quota, or prove the effectiveness of this action plan. Check each study’s task, population and limits before applying it.

1. 【1】Kaya, F., Aydın, F., Schepman, A., Rodway, P., Yetişensoy, O., & Demir Kaya, M. (2024). The Roles of Personality Traits, AI Anxiety, and Demographic Factors in Attitudes toward Artificial Intelligence. International Journal of Human-Computer Interaction. DOI: 10.1080/10447318.2022.2151730.

2. 【2】Logg, J. M., Minson, J. A., & Moore, D. A. (2019). Algorithm Appreciation: People Prefer Algorithmic to Human Judgment. Organizational Behavior and Human Decision Processes, 151, 90-103. DOI: 10.1016/j.obhdp.2018.12.005.

3. 【3】Bogert, E., Lauharatanahirun, N., & Schecter, A. (2022). Human Preferences toward Algorithmic Advice in a Word Association Task. Scientific Reports, 12, 14501. DOI: 10.1038/s41598-022-18638-2.

4. 【4】Ho, A., Hancock, J., & Miner, A. S. (2018). Psychological, Relational, and Emotional Effects of Self-Disclosure After Conversations With a Chatbot. Journal of Communication, 68(4), 712-733. DOI: 10.1093/joc/jqy026.

5. 【5】Papneja, H., & Yadav, N. (2025). Self-disclosure to Conversational AI: A Literature Review, Emergent Framework, and Directions for Future Research. Personal and Ubiquitous Computing, 29, 119-151. DOI: 10.1007/s00779-024-01823-7.