Hannah Liu

Research at the intersections

Each area draws on several fields at once. In the graph, these are the gold nodes.

• Decentralised AI governance

  Frameworks that hold up across countries and cultures, without a single central authority.

  PhilosophyPolicy

  • My research focus
  • ROAR
  • Research engineer at RAI

• Fairness for underrepresented groups

  How AI systems can serve people who are usually missing from training data and policy rooms.

  PhilosophyPolicyPsychology

  • My research focus
  • Bias in clinical prediction models
  • Bias in LLM image generation

• Human and machine cognition

  Using what we know about minds to ask better questions about models.

  Cognitive scienceNeuroscienceComputation

  • BA in Cognitive Science
  • Traits of successful forecasters

• Measuring model bias

  Evaluating bias both in what models say and in how they represent people internally.

  ComputationPsychology

  • Dissertation: fine-tuning and bias guardrails
  • Visiting researcher at Cambridge
  • Bias in LLM image generation
  • Bias in clinical prediction models

• Fine-tuning and safety guardrails

  How much fine-tuning it takes to wear down the guardrails built into language models.

  PolicyComputation

  • Dissertation: fine-tuning and bias guardrails
  • Visiting researcher at Cambridge

• Causal inference and RLHF

  Using causal methods to find where bias comes from and correct it, including a causal RLHF pipeline.

  PhilosophyComputation

  • Bias in clinical prediction models
  • Academic interests

• AI in healthcare

  Clinical models that help without creating feedback loops that harm patients.

  PolicyNeuroscienceComputation

  • Bias in clinical prediction models
  • Predicting dyspnea with machine learning

• Forecasting and human judgement

  What makes some people better at predicting the future, and what that says about judgement.

  Cognitive sciencePsychology

  • Traits of successful forecasters