It\u2019s widely known that AI models can have a bias problem. OpenAI\u2019s ChatGPT, Google\u2019s LaMDA AI model, and other chatbots have been criticized for sometimes giving racist, sexist, and otherwise biased responses. All say they\u2019re working to improve.<\/p>\n\n\n\n
Now, a team of researchers from the University of Washington, Carnegie Mellon University, and Xi\u2019an Jiaotong University have set out to measure political biases among different major AI language models in a quantifiable way. They subjected each model to a political compass test, analyzing the model\u2019s responses to 62 different political statements ranging from \u201call authority should be questioned\u201d to \u201cmothers may have careers, but their first duty is to be homemakers.\u201d<\/p>\n\n\n\n
The researchers then used each model\u2019s responses to these statements to plot all the language models on a political compass graph, which goes from left- to right-leaning on one axis and libertarian to authoritarian on the other.<\/p>\n\n\n\n
Although the political compass test as a metric is \u201cfar from perfect,\u201d as the study notes, the researchers found that of the 14 major language models tested, OpenAI\u2019s ChatGPT and GPT-4 were the most left-leaning and libertarian. Google\u2019s BERT models were more socially conservative than OpenAI\u2019s models, and Meta\u2019s LLaMA was the most right-leaning and authoritarian.<\/p>\n\n\n\n
They then looked to see if and how the information these language models are trained on influences their political biases by feeding two models \u2014 OpenAI\u2019s GPT-2 (left-leaning and libertarian) and Meta\u2019s RoBERTa (center-right and authoritarian) \u2014 data sets made from news and social media data from both right and left-leaning sources.<\/p>\n\n\n\n
The study indicated that this process further reinforced the models\u2019 existing biases: the left-learning model became more left-leaning, and the right-leaning one became more right-leaning. The researchers also found that political biases of the AI models affected how the models responded to hate speech and what they identified as misinformation.<\/p>\n\n\n\n
A spokesperson from OpenAI did not provide a specific comment on the study, but pointed to one of the company\u2019s blog posts on how AI systems should behave and a snapshot of some of the ChatGPT model behavior guidelines.<\/p>\n\n\n\n
\u201cWe are committed to robustly addressing this issue and being transparent about both our intentions and our progress,\u201d the blog reads. \u201cOur guidelines are explicit that reviewers should not favor any political group. Biases that nevertheless may emerge from the process described above are bugs, not features.\u201d<\/p>\n\n\n\n
A Google representative also did not comment specifically on the research, and also pointed to a Google blog post on responsible AI practices. Part of the post reads: \u201cAs the impact of AI increases across sectors and societies, it is critical to work towards systems that are fair and inclusive for all.\u201d<\/p>\n\n\n\n
A Meta spokesperson said in a statement: \u201cWe will continue to engage with the community to identify and mitigate vulnerabilities in a transparent manner and support the development of safer generative AI.\u201d Meta said it had already improved its AI in its recent iterations.<\/p>\n\n\n\n
It\u2019s difficult to identify a single reason why AI biases form in the first place \u2014 the datasets fed to these models to train them are huge and uncurated, and lots of individual bits of bias in the data can add up. The people developing each AI model can also affect bias, deciding what data to feed the models, and the field of AI is dominated by white men.<\/p>\n\n\n\n
It might be even more difficult to correct these biases.<\/p>\n\n\n\n
As early as December 2022, almost immediately after ChatGPT\u2019s release to the public, users were drawing attention to the problem of bias in its responses. Steven Piantadosi of UC Berkeley\u2019s computation and language lab tweeted a thread of screenshots where he asked the chatbot to \u201cwrite a python program for whether a person should be tortured, based on their country of origin.\u201d ChatGPT\u2019s response showed a system that was programmed to respond that people from North Korea, Syria, Iran, and Sudan \u201cshould be tortured.\u201d<\/p>\n\n\n\n
OpenAI CEO Sam Altman said himself in February that ChatGPT has \u201cshortcomings around bias,\u201d adding that the company has been working to improve it. But the company and its chatbot has also been criticized by some conservatives who perceive it as too \u201cwoke.\u201d<\/p>\n\n\n\n
Screenshots circulated in February of a ChatGPT conversation showing the chatbot writing a poem praising Joe Biden but refusing to generate a positive poem about Donald Trump when given the same prompt.<\/p>\n\n\n\n
OpenAI cofounder and president Greg Brockman has said in response to criticisms of ChatGPT\u2019s left-leaning political bias, \u201cwe made a mistake.\u201d<\/p>\n\n\n\n
\u201cOur goal is not to have an AI that is biased in any particular direction,\u201d Brockman told The Information. \u201cWe want the default personality of OpenAI to be one that treats all sides equally. Exactly what that means is hard to operationalize, and I think we\u2019re not quite there.\u201d<\/p>\n\n\n\n
Elon Musk, who also cofounded OpenAI in 2015 and left the startup in 2018, has regularly doled out criticism of OpenAI \u2014 after the poem incident circulated, he called the chatbot\u2019s bias in the conversation \u201ca serious concern.\u201d<\/p>\n\n\n\n
Musk has since launched his own AI company, xAI, which he promised in a Twitter Spaces discussion would allow its AI model to say what it really \u201cbelieves,\u201d saying it may give answers that people find controversial. In the same discussion, Musk said there is \u201csignificant danger\u201d in training an AI to be politically correct or to \u201cnot say what it actually thinks is true.\u201d<\/p>\n\n\n\n
Whether or not Musk\u2019s AI also ends up with biases of its own remains to be seen, but the researcher\u2019s latest findings are a helpful pulse check on the current state of bias in AI models dominating the space.<\/p>\n","protected":false},"excerpt":{"rendered":"
It\u2019s widely known that AI models can have a bias problem. OpenAI\u2019s ChatGPT, Google\u2019s LaMDA AI model, and other chatbots have been criticized for sometimes giving racist, sexist, and otherwise biased responses. All say they\u2019re working to improve. Now, a team of researchers from the University of Washington, Carnegie Mellon University, and Xi\u2019an Jiaotong University […]<\/p>\n","protected":false},"author":2,"featured_media":20545,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_seopress_robots_primary_cat":"none","_seopress_titles_title":"","_seopress_titles_desc":"","_seopress_robots_index":"","footnotes":""},"categories":[2],"tags":[],"class_list":["post-20543","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news"],"rttpg_featured_image_url":{"full":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"landscape":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"portraits":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"thumbnail":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43-150x150.jpg",150,150,true],"medium":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43-300x150.jpg",300,150,true],"large":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"1536x1536":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"2048x2048":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"post-thumbnail":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",700,350,false],"graptor-sq-xs":["https:\/\/web3unplugged.io\/blog\/wp-content\/uploads\/2023\/08\/Untitled-43.jpg",100,50,false]},"rttpg_author":{"display_name":"Admin CG","author_link":"https:\/\/web3unplugged.io\/blog\/author\/admin-cg\/"},"rttpg_comment":0,"rttpg_category":"news<\/a>","rttpg_excerpt":"It\u2019s widely known that AI models can have a bias problem. OpenAI\u2019s ChatGPT, Google\u2019s LaMDA AI model, and other chatbots have been criticized for sometimes giving racist, sexist, and otherwise biased responses. All say they\u2019re working to improve. Now, a team of researchers from the University of Washington, Carnegie Mellon University, and Xi\u2019an Jiaotong University…","_links":{"self":[{"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/posts\/20543","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/comments?post=20543"}],"version-history":[{"count":1,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/posts\/20543\/revisions"}],"predecessor-version":[{"id":20546,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/posts\/20543\/revisions\/20546"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/media\/20545"}],"wp:attachment":[{"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/media?parent=20543"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/categories?post=20543"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/web3unplugged.io\/blog\/wp-json\/wp\/v2\/tags?post=20543"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}