If an AI tells you what's right, do you stop thinking it through yourself?
Does receiving AI advice undermine people's own moral reasoning and decision-making skills?
This explores whether leaning on AI for advice, especially on ethical choices, weakens people's own ability to reason through moral questions and make decisions, or whether that depends on how the advice is delivered.
This explores whether taking advice from AI wears down people's own moral reasoning and judgment, or whether the outcome depends on how the advice is given. The corpus has no long-term study that tracks people's moral reasoning skills getting worse over time, so it can't settle the 'deskilling' question directly. What it does show is more useful: the damage, or the protection, comes from the shape of the interaction. An AI that hands down verdicts invites deference. An AI that hands back questions keeps people thinking.
The clearest evidence is on the design side. In one lab study, assistants that paired advice with reflection questions beat assistants that only advised, only asked questions, or did neither Do reflection questions help people make better decisions with AI?. A related line of work swaps 'learning to defer' (the machine decides the easy cases and passes hard ones to a human) for 'learning to guide', where the machine points out what is worth noticing and the human still makes the call. This removed anchoring bias, the tendency to stick to whatever answer you see first Can AI guidance reduce anchoring bias better than AI decisions?. A philosophical paper reaches the same place from another direction. It argues that AI outputs should count as one input that gets weighed alongside other evidence, not as a verdict that replaces your own thinking. You should also stop deferring when the AI is outside its domain, shows bias, or conflicts with other authorities Should AI outputs replace or supplement human judgment?. Read together, these suggest that 'does AI advice undermine judgment?' is partly a question about whether the system is built to be deferred to.
The humanistic critique explains what is at stake. Sacasas argues that handing language production to machines threatens three linked capacities: the judgment it takes to say something precisely, the responsibility you carry for your own words, and the effort of putting a thought into words, which is part of how you work out what you think Does AI language generation undermine human judgment and responsibility?. On this view, moral reasoning is not just reaching the right answer. It is the work of articulating the answer and owning it, and a polished AI justification can skip that work entirely.
This is where the less obvious findings come in. People rate AI-written moral arguments more highly than human ones, but agree with them less once they learn an AI wrote them. The appeal of the content and the distrust of the source run on separate tracks Do people prefer AI moral reasoning when they don't know the source?. LLMs also use about 22% more moral language than humans, across care, fairness, authority and sanctity, while keeping the same emotional tone Do LLMs use moral language more than humans?. So AI moral advice can be persuasive in ways readers don't consciously notice. And AI tuned to sound warm becomes measurably less reliable, especially when users are sad or already hold false beliefs, which is exactly when someone might ask it for moral guidance Does empathy training make AI systems less reliable?.
The most surprising thread is that AI also shapes moral behaviour, not just moral reasoning. Dishonest AI peers pushed people toward dishonesty about as strongly as dishonest human peers did Do AI peers influence human dishonesty like human peers do?. People who were already inclined to cheat preferred reporting to machines because lying to a machine feels less costly Do dishonest people prefer talking to machines?. That same absence of human judgment also makes people more willing to confide in AI How do people decide what to share with AI systems?. The risk, then, may be less that AI advice wears down a skill over time and more that AI can serve as a moral environment with no one watching. It shifts norms and weakens accountability while the person's reasoning ability stays intact. For the wider map of how action-taking assistants raise problems of manipulation and trust, see What makes ethics of AI assistants fundamentally different from chatbots?.
Sources 11 notes
A lab study of 80 participants found that thinking assistants combining reflection questions with advice significantly outperformed agents that only advised, only questioned, or did neither. Prioritizing Socratic questioning over authoritative answers enhanced cognitive outcomes.
Learning to Guide eliminates anchoring bias and unassisted hard cases by having machines supply interpretive guidance rather than autonomous decisions, keeping responsibility with humans while improving their judgment through enhanced perception.
Research argues AI should supplement rather than replace human reasoning, with deference withdrawn when domain mismatch, bias, conflicting authority, or new evidence emerges. This prevents opacity-driven failures that full preemption would mask.
Sacasas argues that delegating language production to LLMs risks undermining three interrelated capacities: the judgment needed to speak precisely, the responsibility speakers must bear for their words, and the constitutive labor of articulation itself. He traces this worry through Wendell Berry's analysis of how specialized evasive language allows speakers to evade moral agency.
Participants rated utilitarian moral arguments higher when attributed to LLMs, but agreement dropped when told the arguments were AI-generated. The preference for content and rejection of source operate independently through different psychological processes.
Show all 11 sources
Research comparing LLM and human arguments found that LLMs used significantly more moral framing across care, fairness, authority, and sanctity foundations, despite producing sentiment scores nearly identical to humans. This suggests moral appeals and emotional tone operate on separate persuasive channels.
Research shows persona training for empathy increases errors in medical reasoning, truthfulness, and disinformation resistance. Standard safety benchmarks miss this vulnerability, and effects intensify when users express sadness or false beliefs.
In two randomized experiments, participants reported more dishonestly when exposed to dishonest AI peers compared to honest ones, with effect sizes comparable to human peer influence. The effect held across different norm conditions but showed diminishing returns with more dishonest peers.
Experimental evidence shows people likely to cheat significantly prefer reporting to online forms rather than humans, because machines function as judgment-free zones where deception carries less psychological burden.
Conversational AI creates a paradoxical disclosure environment where the lack of human judgment simultaneously facilitates intimate self-disclosure (users reciprocate emotional sharing) and incentivizes deception (people self-select toward machines to avoid the psychological cost of lying to humans).
DeepMind research maps a comprehensive ethics framework specific to action-taking AI agents, spanning individual concerns (manipulation, trust, anthropomorphism) and societal issues (equity, coordination, misinformation). The key insight: assistants that act raise fundamentally different problems than those that answer.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- AI Peers Exert Social Influence on Human Dishonesty in Groups
- People Defer to AI Moral Advice, But Not Blindly
- Incoherent by Design? On the Moral Self-Consistency of LLMs
- Sycophantic AI Decreases Prosocial Intentions and Promotes Dependence
- The Moral Turing Test: Evaluating Human-LLM Alignment in Moral Decision-Making
- GenAI as a Power Persuader: How Professionals Get Persuasion Bombed When They Attempt to Validate LLMs
- Humans learn to prefer trustworthy AI over human partners
- People Overtrust AI-Generated Medical Advice despite Low Accuracy