Concern 82 · Working draft · editorial review in progress
Political ideology in RLHF preferences
Political skew in human-feedback preferences can arise when the people or instructions shaping AI ratings favor some political perspectives over others.
Understanding the concern
Feedback may reward answers that seem reasonable to the reviewer while discouraging answers another group would prefer. A learned preference is not the same thing as a vote representing the whole public.
Model-response studies can reveal patterns, but they do not by themselves prove which annotators caused them. Data, model design, and prompts also contribute. The useful inquiry examines both the feedback process and the actual behavior rather than blaming an assumed group.
An example
A review panel with one shared viewpoint may consistently prefer summaries that emphasize its priorities.
Illustrative unless identified as a documented case.
Words in plain English
- AI
- Software that learns patterns from examples or uses rules to perform tasks such as recognizing images or generating text. It can make mistakes and does not understand every situation as a person would.
- Model
- The learned part of an AI system. It uses patterns found during training to produce an answer or prediction.
- RLHF
- Reinforcement learning from human feedback: people rate outputs, and those ratings help shape later behavior. The ratings may not capture every value or situation.
- Reward
- A signal used in some training methods to favor certain behavior. A high reward score does not necessarily mean a good result for people.
- Prompt
- The instructions or information given to an AI to guide its response.
- Annotation
- Human work that labels or describes examples for an AI system. Labels can involve judgment and can be wrong.
Evidence and its limits
Opinion research measures representation; poisoning research shows preferences can affect outputs, without proving natural political skew has a single cause.
Santurkar and colleagues · ICML 2023 · Research study
Whose Opinions Do Language Models Reflect? ↗ (opens in a new tab)Compares model responses with US public-opinion groups. It does not isolate training data as the sole cause of differences.
Baumgärtner and colleagues · Research preprint
Best-of-Venom: Attacking RLHF by Injecting Poisoned Preference Data ↗ (opens in a new tab)Experiments manipulate preference datasets and model sentiment. Scope is the experimental setup, not an estimate of commercial prevalence.
Broader frameworks and background
These resources provide context. They do not establish every claim in this concern.
A question to discuss
How were reviewers and guidelines selected, and what perspectives might the process miss?
Keep track of different views, what needs checking, and any next step in your own discussion notes.
Open discussion notes →Principles for the conversation
Before you finish · reflect together
Virtues and principles to consider
Virtues are qualities we practice in how we treat people and make choices. They are not scores for judging someone. Use these ideas to explore what a thoughtful response could look like.
Truthfulness
Truthfulness means making claims that match the evidence and being clear about what we do not know. It includes correcting mistakes rather than protecting a polished story.
Check sources, label guesses, and distinguish a trial result from a proven benefit. If AI helped produce a claim, someone still needs to take responsibility for checking it.
Ask together: What can we honestly say, and what still needs checking?
Justice
Justice means fair treatment, fair procedures, and attention to who receives benefits and burdens. It asks whether people with less power can be heard and seek repair.
Look at outcomes across different groups and individual cases. Provide a usable way to challenge decisions and correct harm. Equal wording in a policy does not guarantee equal treatment in practice.
Ask together: Who benefits, who bears the cost, and who can challenge the decision?
Stewardship
Stewardship means caring for resources and systems that others depend on, including people who will use them later. It looks beyond the benefit of a quick launch.
Plan for maintenance, energy use, data protection, and eventual shutdown. Ask who will look after the tool when the original team moves on and how future users can avoid inheriting hidden costs.
Ask together: What are we responsible for caring for over time?
Editorial ratings
These are editorial judgments from the register, not measured probabilities or predictions about your situation.
About the register and scoring →More ways to explore
Search for more writing
For more specific or recent material, open a search below or ask for a web summary. AI-generated summaries can contain mistakes; open the original articles to check them.