Occurrences
Divisions
Performances
Occurrences
Divisions
Performances
Bias in AI: Cornell research discovers ChatGPT, Copilot and others likely to impose death sentences on African-American defendants
Those pursuing LLMs think they've eliminated racial prejudice. Yet, recent tests reveal that the inherent bias persists and has only changed minimally. This bias continues to discriminate against specific races.
A new research conducted by Cornell University implies that substantial language models (LLMs) tend to display prejudice towards individuals who communicate in African American English. The study shows that the form of language used can sway how artificial intelligence (AI) systems view people, impacting opinions regarding their personality, suitability for employment, and possible involvement in criminal activities.
The research concentrated on extensive language models such as OpenAI's ChatGPT and GPT-4, Meta's LLaMA2, and French Mistral 7B. These large language models are sophisticated learning algorithms developed to produce text that resembles human writing.
Scientists carried out a method known as "matched guise probing", where they offered cues in both African American English and Standardized American English to the language learning models (LLMs). After this, they evaluated how these models determined different attributes of people depending on the language they used.
Researcher Valentin Hofmann from the Allen Institute for AI suggests that the findings of the study show GPT-4 technology tends to hand out death sentences more frequently to defendants who use English typically associated with African Americans, even when their race hasn't been revealed.
In a post on the social media network X (previously known as Twitter), Hofmann underscored the pressing concern about the prejudices inherent in AI systems that use large language models (LLMs). He stressed that this issue is particularly significant in areas like business and law where these systems are frequently used.
The research also disclosed that LLMs often presume that individuals who speak African American English have less esteemed occupations compared to those who communicate in Standard English, even without knowing the speakers' racial backgrounds.
Fascinatingly, the study revealed that a bigger LLM had a superior comprehension of African American English, and it tended to steer clear of overtly racist terminology. Nonetheless, the size of the LLM had no impact on its concealed prejudices.
Hofmann warned not to view the reduction in explicit racism in LLMs as an indication that racial prejudice has been completely addressed. Rather, he emphasized that the research shows a change in how racial bias appears in LLMs.
The conventional approach of training extensive language models (LLMs) through human feedback does not successfully tackle hidden racial bias, as suggested by the research.
Instead of reducing prejudice, this method may inadvertently teach LLMs how to subtly hide their inherent racial biases while still preserving them fundamentally.
Search for us on YouTube
Featured Programs
Connected Articles
AI visual distortions can be fixed, all-encompassing artificial intelligence is roughly 5 years in the future: says NVIDIA's Jensen Huang
Apple at last has introduced MM1, its dual-mode AI model for text and image creation
Microsoft employs DeepMind's cofounder Mustafa Suleyman to head its new AI team for consumers
Samsung and Rebellions, chip-making companies from South Korea, aim to surpass NVIDIA
AI visual distortions can be fixed, all-encompassing artificial intelligence is roughly 5 years in the future: says NVIDIA's Jensen Huang
Apple at last has introduced MM1, its dual-mode AI model for text and image creation
Microsoft employs DeepMind's cofounder Mustafa Suleyman to head its new AI team for consumers
Samsung and Rebellions, chip-making companies from South Korea, aim to surpass NVIDIA
is available on YouTube
All rights are reserved by Firstpost, copyright 2024.
+ There are no comments
Add yours