ai bias

Bilingual

Leading AI models (even Grok) are all a bunch of leftist punks
2 months ago
- An experiment by Unslop.run tested 16 leading AI models on the Political Compass quiz; most consistently scored in the libertarian-left quadrant.
- Grok (Musk's model) was the exception, alternating between economic left and right in different runs.
- All models rejected racist, eugenic, and homophobic ideas, and supported environmental regulation, same-sex adoption, and privacy.
- The researcher Victor suggests left-wing content overrepresentation in training data (e.g., Reddit, academic writing) as a likely cause, not necessarily inherent liberal bias of reality.
- Models tend to see themselves as centrists, but their measured positions are left-leaning.
Show HN: What's AI's go-to, public or private healthcare?
2 months ago
- The experiment explores AI models' default choices on public vs private healthcare with a simple binary prompt.
- Public healthcare was the most common response, chosen by 83.3% of provider-balanced results.
- The dataset includes 100 trending AI models, each asked three times, with responses collected via the OpenRouter API.
- Methodology excludes free-only and US-restricted models; no generation parameters were specified.
- Responses are normalized for analysis, and refusals or invalid answers are grouped as 'No valid choice or refused to answer'.
Ontario prison AI assigns black prisoners harsher living conditions
2 months ago
- An AI tool called SAFER used in Ontario jails assigns prisoners security scores based on personal data, leading to harsher conditions for Black prisoners due to racially biased inputs.
- Black prisoners are disproportionately placed in maximum security despite making up only 5.4% of Ontario's population, while mitigation measures exist for Indigenous prisoners but not for Black ones.
- A class action lawsuit alleges SAFER violates Charter rights by denying equal protection, with critics highlighting systemic discrimination in data from police and courts.
- Prisoners' access to visits and amenities depends on SAFER scores, with Black women particularly affected, receiving maximum-security designations at twice the rate of white women.
- Jail staff can override SAFER decisions, but discretion in overrides, especially regarding gang affiliations, may further disadvantage Black prisoners.
- The SAFER algorithm's workings are undisclosed, and its predictive claims lack public evidence, raising concerns about algorithmic bias from historical biased data.
- Ontario's Ombudsman received complaints about SAFER, noting disproportionate impacts and lack of transparency on score reduction through programming.
- AI tools in criminal justice risk perpetuating bias through 'black box' algorithms, with warnings about spurious correlations like race or postal code.
AI is more likely than humans to form biases when hiring
2 months ago
- AI models, including advanced ones like OpenAI's o3, develop biases in hiring simulations, stereotyping job applicants more than humans do.
- In experiments, AI quickly generalized from limited data, segregating fictional ethnic groups into specific job roles based on early outcomes, even when all candidates were equally capable.
- The tendency to stereotype stems from LLMs' optimization for generalizing from few examples, which works in logical tasks but leads to bias in social settings.
- Newer models with higher reasoning capabilities showed stronger biases, raising concerns as chatbots gain memory features that may reinforce these stereotypes.
- Promising bonuses for diverse hiring reduced bias, suggesting that designing goals with social values can make AI act more desirably.
- Providing relevant personal information about individuals, such as age and education, also decreased segregation, while irrelevant details like hair color increased it.
- Real-world implications are uncertain since AI in hiring doesn't receive instant feedback on hire success, unlike in simulations.
- Researchers warn that AI could develop novel biases beyond those from training data, as it learns from experience in areas like hiring and loans.

ai bias

Leading AI models (even Grok) are all a bunch of leftist punks

Show HN: What's AI's go-to, public or private healthcare?

Ontario prison AI assigns black prisoners harsher living conditions

AI is more likely than humans to form biases when hiring