Epistemic uncertainty
Uncertainty from a lack of knowledge or data, which could be reduced with more information.
Epistemic uncertainty means a model doesn't know something because it hasn't seen enough examples or data. It's like a student who hasn't studied a particular topic for a test. The more they study, the more confident they become.
For example, if you ask a language model about a very recent event it wasn't trained on, it might not know the answer. This is epistemic uncertainty. If it were trained on more recent data, it might be able to answer.
Epistemic uncertainty stems from a deficiency in the model's knowledge or the training data. This type of uncertainty is theoretically reducible; acquiring more data or improving the model's ability to learn from existing data can decrease it. It contrasts with aleatoric uncertainty, which is inherent noise or randomness in the data itself and cannot be reduced by more data.
Frontier AI labs are focusing on making models more transparent about what they don't know. This allows users to better gauge the reliability of AI-generated information, especially in critical applications. Teaching models to express epistemic uncertainty is a step towards more trustworthy AI systems.
- Teaching models to express their uncertainty in wordsOpenAI · OpenAI News ·
- OpenAI ChatGPT
- OpenAI has explored methods for models like GPT-3 to verbalize their confidence, which relates to expressing epistemic uncertainty.
- Aleatoric uncertainty
- This is uncertainty inherent in the data itself, like ran
- Model confidence
- This is a general term for how sure a model is, which can enco
The concept of epistemic uncertainty originates from Bayesian probability theory and has been widely applied in machine learning, particularly in the context of probabilistic models and neural networks. Source ↗
Ask a language model a question about a very niche or extremely recent event. Observe if it states it doesn't know or provides a speculative answer. Then, ask a more common question and see if its confidence level or phrasing changes.
The Playbook: reusable how-tos, prompts, configs and tips pulled out of the news.
- PromptRole-play a student struggling with math for AI tutor simulationSimulate a student's math struggles and uncertainty for AI tutor evaluation. · 5 min
- TemplateReusable workflow templates for healthcare teamsTeams can create and share templates for common workflows like documentation or prior authorizations. · 15 min
- TemplateTemplate for building an AI tool tailored to your workA structured approach to defining and building a custom AI tool for specific business needs. · 15 min
- TemplateEvals CSV prompt list for agent skillsA CSV file template for listing prompts used in evaluating agent skills. · 20 min
- HeuristicDouble coding agent task completion with docs linksProviding coding agents with a documentation link significantly increases their task completion rate. · 2 min
Questions people ask
- How is epistemic uncertainty different from a model just being wrong?
- A model can be wrong due to epistemic uncertainty when it lacks the necessary knowledge. It's different from a factual error made when the model has sufficient knowledge but misapplies it.
- Can models learn to reduce epistemic uncertainty?
- Yes, models can reduce epistemic uncertainty by being trained on more diverse and comprehensive data, or by using techniques that allow them to better infer knowledge gaps.
First seen on Subvolts May 28, 2022. Updated Sep 15, 2026. Written by an AI model from the items above and reviewed against them weekly; report an error.