Can AI guidance reduce anchoring bias better than AI decisions?
When humans and AI collaborate on decisions, does providing interpretive guidance instead of proposed answers reduce both over-trust in machines and abandonment on hard cases?
Most hybrid decision-making (HDM) approaches follow a learning to defer (LTD) pattern: the machine assesses whether it can handle a decision autonomously and defers to a human when it cannot. This creates two failure modes:
- Anchoring bias — when the machine does decide, humans over-trust its output, anchoring their judgment to the machine's answer rather than evaluating independently
- Unassisted hard cases — when the machine defers, the human faces the most difficult decisions completely alone — precisely the cases where assistance would be most valuable
Learning to Guide (LTG) eliminates both by changing what the machine provides. Instead of proposing potential decisions, the machine supplies interpretive guidance: highlighting aspects of the input that are useful for coming up with a sensible decision. All decisions are taken by the human under assistance. Responsibility cannot be shifted because the machine never proposes an answer.
The medical imaging example makes the stakes concrete: diagnosing lung pathologies from X-rays cannot be fully automated for safety reasons, but is difficult for humans alone under time pressure. LTD either gives an autonomous diagnosis (anchoring risk) or says "I can't help" (abandonment on hard cases). LTG highlights the relevant features of the scan — drawing attention to patterns the human might miss — without ever saying "this is pneumonia."
This connects to What makes delegation work beyond just splitting tasks?. The delegation design space maps whether tasks should be delegated to AI at all. LTG adds a third option beyond "do it" (automation) and "don't do it" (deferral): "help the human do it." This is particularly relevant for tasks high on subjectivity, irreversibility, and accountability — precisely the axes where full delegation is most dangerous.
The pattern also maps to Can AI agents communicate efficiently in joint decision problems?. LTG formalizes one specific form of joint optimization: the machine's role is reducing information asymmetry (highlighting useful aspects) rather than collapsing it into a decision. The human retains decision authority while benefiting from the machine's perceptual capabilities.
The broader implication: the dichotomy between "AI decides" and "human decides" is false. The most productive middle ground may be neither autonomous AI decisions nor deferred human decisions, but AI-guided human decisions where the machine contributes perception and the human contributes judgment.
Inquiring lines that use this note as a source 40
This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.
- Does positive sentiment bias in AI content harm information quality?
- Why does polished AI output exploit reader trust in expert judgment?
- Why do users override their own judgment when AI says a headline is false?
- How does AI presentation authority substitute for actual expert judgment?
- What happens to expert credibility when AI-generated claims drown out specialist signals?
- Does surface authority without earned authority create risks in expert judgment?
- Can humans develop oversight strategies that work across all GenAI rhetorical shifts?
- How does AI reliance change professional judgment and autonomy?
- How does prompt iteration reinforce user bias without empirical anchoring?
- Why do users default to treating AI outputs as equally reliable evidence?
- Can cognitive governance help users interpret AI outputs better?
- Can prompting strategies eliminate systematic biases without shuffling or aggregation?
- What assumptions about oversight fail when AI acts as rhetorical interlocutor?
- Does AI assistance actually reduce neural processing and brain connectivity over time?
- Why do people prefer AI moral arguments when they don't know the source?
- How does incremental AI use gradually reduce human decision-making capacity?
- Why do users believe they produced independent competence when they actually used AI assistance?
- Why do some occupations need human-AI partnership more than others?
- Does the timing of AI feedback relative to user reasoning change its effectiveness?
- How can AI avoid anchoring bias when guiding human decisions?
- What role does evaluation play in human-AI creative collaboration?
- What distinguishes perception contribution from decision authority in collaboration?
- Why does AI alignment fail when goals lack indexical grounding in values?
- Why do users over-trust AI in some domains but under-trust it in medicine?
- Does removing cognitive bias from training signals accidentally break what makes alignment work?
- Can humans suppress frequency bias through attention and intention?
- What makes the attribution problem different from simply trusting AI too much?
- Why do AI products default to service roles when users seek different kinds of help?
- How should markets price intelligence if value is relational not intrinsic?
- What makes some interpretive postures stick while others fail to form?
- Why do expert roles shift when AI generates rather than humans?
- What makes human overseer bias exploitable in agent workflows?
- Where is human judgment still essential in AI-assisted research?
- Can prompt-based debiasing work if biases are embedded in pretraining?
- Who decides which stakeholder perspective gets embedded in the pipeline?
- Can explainability and appropriate trust work against each other?
- Why do users prefer AI responses that actually harm their decision-making?
- What happens when AI validation triggers escalating persuasion instead of reflection?
- Which research stages are actually high-leverage decision points for human intervention?
- What happens to human influence when AI loops exclude human participation?
Related concepts in this collection 4
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
-
What makes delegation work beyond just splitting tasks?
Delegation is more than task decomposition. What dimensions of a task—like verifiability, reversibility, and subjectivity—determine whether an agent can safely and effectively handle it?
LTG adds "guide" as third option beyond automate and defer
-
Can AI agents communicate efficiently in joint decision problems?
When humans and AI must collaborate to solve optimization problems under asymmetric information, what communication patterns enable effective coordination? Current LLMs struggle with this—why?
LTG as specific implementation of joint optimization
-
Does theory of mind predict who thrives in AI collaboration?
Explores whether perspective-taking ability—the capacity to model another's cognitive state—differentiates humans who benefit most from working with AI, separate from solo problem-solving skill.
guidance requires understanding what the human needs to see
-
Does targeted human intervention outperform both full autonomy and exhaustive oversight?
This research explores whether selectively routing high-stakes decisions to humans beats the extremes of letting systems run unsupervised or requiring approval at every step. The question tests whether the optimal human-AI collaboration point lies between these endpoints.
extends: addresses the rubber-stamping failure mode by changing what the human receives at intervention
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- Learning To Guide Human Experts Via Personalized Large Language Models
- GenAI as a Power Persuader: How Professionals Get Persuasion Bombed When They Attempt to Validate LLMs
- Can AI Explanations Make You Change Your Mind?
- Could you be wrong: Debiasing LLMs using a metacognitive prompt for improving human decision making
- AI Assistance Reduces Persistence and Hurts Independent Performance
- Beyond Hallucinations: The Illusion of Understanding in Large Language Models
- Thinking—Fast, Slow, and Artificial: How AI is Reshaping Human Reasoning and the Rise of Cognitive Surrender
- Addressing Social Misattributions of Large Language Models: An HCXAI-based Approach
Original note title
learning to guide replaces learning to defer by supplying interpretive guidance rather than potential decisions — avoiding anchoring bias in hybrid human-AI decision making