MIT AI Risk Repository · Risk Sub-Category · 73.04.01
Harms of Representation and Other Biases
Category: LLM-Systems Can Be Untrustworthy
Description
"A pretrained LLM generally has many of the stereotypical biases commonly present in the human society (Touvron et al., 2023). This makes it difficult for users to trust that LLMs will work well for them and not produce unfair or biased responses. Appropriate finetuning can effectively limit the bias displayed in LLM outputs in a variety of situations, e.g. when models are explicitly prompted with stereotypes (Wang et al., 2023k), but it does not ‘solve’ the problem. Even after finetuning, biases often resurface when deliberately elicited (Wang et al., 2023k), or under novel scenarios, e.g. in
From Foundational Challenges in Assuring Alignment and Safety of Large Language Models (Anwar2024), as extracted by the MIT AI Risk Repository (CC BY 4.0).
Classification
- Causal entity
- AI
- Intent
- Unintentional
- Timing
- Post-deployment
Subdomain definition: Unequal treatment of individuals or groups by AI, often based on race, gender, or other sensitive characteristics, resulting in unfair outcomes and representation of those groups.
Real-world incidents in this subdomain
- DOGE Reportedly Relied on Unvetted ChatGPT Outputs in Canceling National Endowment for the Humanities Grants
- Sora Video Generator Has Reportedly Been Creating Biased Human Representations Across Race, Gender, and Disability
- Meta AI Characters Allegedly Exhibited Racism, Fabricated Identities, and Exploited User Trust
- Alleged AI-Generated Photo Alteration Leads to Inappropriate Modifications in Speaker's Conference Picture
- Algorithmic Bias in French Welfare System Allegedly Discriminates Against Marginalized Groups
- Department for Work and Pensions (DWP) AI Systems Allegedly Discriminate Against Single Mothers
How other frameworks describe this risk
Other entries from Anwar2024
- Agentic LLMs Pose Novel Risks
- Natural Language Underspecifies Goals
- Goal-Directedness Incentivizes Undesirable Behaviors
- Safety Risks from Affordances Provided to LLM-agents
- Multi-Agent Safety Is Not Assured by Single-Agent Safety
- Foundationality May Cause Correlated Failures
- Groups of LLM-Agents May Show Emergent Functionality
- Collusion between LLM-Agents
- Dual-Use Capabilities Enable Malicious Use and Misuse of LLMs
- Misinformation and Manipulation
- Cybersecurity
- Cybersecurity