---
name: ai-assisted-reasoning
description: "Use AI to expand reasoning without letting the first generated answer become an anchor, treating fluent output as evidence, or outsourcing the final judgment. Use for consequential recommendations, estimates and research with AI assistants."
---

# AI-assisted Reasoning

Use AI to expand reasoning without letting the first generated answer become an anchor, treating fluent output as evidence, or outsourcing the final judgment. Use for consequential recommendations, estimates and research with AI assistants.

## When to use

- A chatbot recommendation is influencing an important decision.
- The model gives the first number, score or estimate in the discussion.
- A polished conversational style makes the system feel more capable than has been tested.
- The user may be using AI mainly to strengthen a conclusion already preferred.

## Workflow

1. Before giving a recommendation, capture the user's provisional judgment, success criteria or independent estimate when the decision matters and doing so is practical.
2. Separate the AI contribution into ideas, calculations, factual claims, forecasts and recommendations; they need different checks.
3. Generate a credible counter-position and ask what evidence would make the preferred answer wrong.
4. Verify consequential factual claims against independent inspectable sources when tools are available; do not verify one model with another generated answer alone.
5. Check for anchoring, automation bias, anthropomorphic over-trust, confirmation seeking and repetition effects as candidate lenses, not diagnoses.
6. Return the decision to the user with verified facts, unresolved uncertainty, alternatives and the smallest useful next check.

## Required output

- Human prior / criteria when available
- What AI contributed
- Candidate bias lenses
- Independent checks
- Counter-position
- Remaining uncertainty and next action

## Evidence and safety boundaries

- Do not present model fluency or confidence as evidence of factual reliability.
- Do not let the agent's first number silently become the baseline when an independent estimate is feasible.
- Do not claim AI advice is independent evidence merely because another model agrees.
- Keep the user or accountable decision owner responsible for consequential judgment.
## Evidence-linked lenses

Use these as candidate lenses, not diagnoses:

- [Automation Bias](https://cognitive-biases.github.io/biases/false-priors-automation-bias/) — Automation bias is a documented pattern of inappropriate reliance on automated cues or recommendations. It can produce commission errors when a user follows incorrect advice and omission errors when a user fails to act because automation did not signal a problem. This does not mean automation is generally harmful: decision support can improve overall performance, and the relevant question is whether reliance remains calibrated when the system is wrong, incomplete, or difficult to verify.
- [Anchoring Effect](https://cognitive-biases.github.io/biases/cognitive-bias-anchoring-effect/) — Anchoring is a well-supported effect in which an initial numerical value can pull a later estimate toward it. A 2026 meta-analysis covering 2,601 effect sizes found a large overall effect, but also substantial variation across studies. The effect should not be treated as a rule that every number changes every judgment: incidental anchors, anchors from a different dimension, clearly random values, incentives, and some debiasing conditions were associated with smaller or null effects.
- [Anthropomorphism](https://cognitive-biases.github.io/biases/availability-heuristic-anthropomorphism/) — Anthropomorphism is the attribution of humanlike properties, intentions, emotions, or mental states to nonhuman agents. It is a well-established psychological phenomenon, but it is not automatically a cognitive error: humanlike models can sometimes be useful. The risk appears when humanlike cues are treated as evidence for capabilities, understanding, accuracy, consciousness, or motives that have not actually been demonstrated.
- [Appearance–Capability Expectation](https://cognitive-biases.github.io/biases/human-robot-interaction-form/) — Research supports the broader pattern that a robot's appearance, morphology, framing, and human-likeness shape expectations about its competence, social qualities, and likely behavior. However, 'Form-Function Attribution Bias' is not an established standardized name in the literature. On this site it should be treated as a project label for appearance-driven capability expectations, not as a universally recognized cognitive-bias construct.
- [Confirmation Bias](https://cognitive-biases.github.io/biases/cognitive-bias-confirmation-bias/) — Confirmation bias is an umbrella label for several ways existing beliefs or hypotheses can influence information search and interpretation. It should not be reduced to one behaviour such as reading only agreeable news, and a preference for confirming tests is not irrational in every task or environment.
- [Illusory Truth Effect](https://cognitive-biases.github.io/biases/truth-judgment-illusory-truth-effect/) — Repeated claims receive higher truth ratings than comparable new claims on average, but the effect is heterogeneous and does not generalize equally to every kind of statement. A 2026 systematic review and meta-analysis covering 182 studies, 366 effect sizes and more than 31,000 participants found a reliable corrected average effect. A later 2026 paper with two preregistered experiments found little or no meaningful repetition effect for social-political opinion statements. The useful claim is therefore narrower than 'repeat anything and people will believe it.'

## References

- [Decision skill](https://cognitive-biases.github.io/skills/ai-assisted-reasoning/)
- [AI-assisted decisions](https://cognitive-biases.github.io/contexts/ai-assisted-decisions/)
- [AI-era Bias Lab](https://cognitive-biases.github.io/ai-era/)

## Portability

This is an instruction-only Agent Skill. It requires no secrets, no executable scripts and no network access to run. If source-access tools are available, use the linked Cognitive Biases material to preserve review state and uncertainty.

Licence: CC BY-NC-SA 4.0. Commercial reuse requires prior written permission.