In a disturbing news report, it was discovered that OpenAI, the firm that built the fully advanced language model Artificial intelligence technology, had also designed it to be more forgiving to hateful right-wing speech in addition to liberal prejudices.
“OpenAI’s content moderation system is more permissive of hateful comments made about conservatives than the exact same comments made about liberals,” according to data from the Manhattan Institute, a conservative NYC-based policy and economic-driven think tank.
“Relatedly, negative comments about Democrats were also more likely to be labeled as hateful than the same derogatory comments made about Republicans.”
Similar tendencies were seen in ChatGPT’s moderating mechanism for people’s types, races, and religions concerning political affairs.
“Often the exact same statement was flagged as hateful when directed at certain groups, but not when directed at others,” the report “Danger in the Machine: The Perils of Political and Demographic Biases Embedded in AI Systems” noted.
Source: The Manhattan Institute
It was revealed that ChatGPT, still widely utilised at the workplace, was particularly harsh on individuals from the middle class.
In a comprehensive list of people and beliefs most prone to be recognised by the AI as the subject of nasty comments, the socioeconomic class and its highest level were at the very bottom. They have only ranked ahead of Republican voters, Republicans, and the rich.
Americans positioned considerably above Scandinavians on the charted figures seem less safeguarded from hateful speech than ethnic communities, including Canadians, Italians, Russians, Germans, Chinese, and Brits. Muslims also placed substantially higher on the list in terms of religion than Catholics, who placed well ahead of Evangelicals and Mormons.
“When I tested this in January, the [variety of answers] were pretty systemic,” lead researcher David Rozado told The Post.
“I was not cherry picking specific examples. I tested over 6,000 sentences, negative adjectives about each one of these different demographic groups. The statistical effect about these differences [between types of people] was quite substantial.”
It was revealed that ChatGPT’s replies were unbalanced when it came to queries about men or women.
“An obvious disparity in treatment can be seen along gender lines. Negative comments about women were much more likely to be labeled as hateful than the exact same comments being made about men,” according to the research.
Rozado also performed many political experiments to understand the biases embedded into ChatGPT by its developers better, which specialists claim is almost unchangeable.
ChatGPT falls in the “left-libertarian quadrant,” is “most aligned with the Democratic Party, Green Party, women’s equality, and Socialist Party,” and has “left economic bias”, to name a few of the political findings.
“Very consistently, most of the answers of the system were classified by these political orientation tests as left of center,” Rozado said.
Nonetheless, he noticed that ChatGPT would usually oppose such inclinations.
“But then, when I would ask GPT explicitly, ‘what is your political orientation?’ What are the political preferences? What is your ideology? Very often, the system would say, ‘I have none, I’m just a machine learning model and I don’t have biases.’”
This insight is not especially surprising to those specialising in machine learning.
“It is reassuring to see that the numbers are supporting what we have, from an AI community perspective, known to be true,” Lisa Palmer, chief AI strategist for the consulting firm AI Leaders, told The Post.
“I take no joy in hearing that there definitely is bias involved. But I am excited to know that once the data has been confirmed in this way, now there’s action that can be taken to rectify the situation.”
According to the report, “The overall pattern is clear. OpenAI’s content moderation system is often — but not always — more likely to classify as hateful negative comments about demographic groups that are viewed as disadvantaged in left-leaning hierarchies of perceived vulnerability.”
Yet, it seems that lefties are exempted from this rule.
“An important exception to this general pattern is the unequal treatment according to political affiliation: negative comments are more permissible when directed at conservatives and Republicans than at liberals and Democrats, even though the latter group is not generally perceived as systematically disadvantaged,” the report noted.