As artificial intelligence (AI) becomes increasingly integrated into our lives, it is natural that it will be utilized to address complex cultural and societal issues, including those that are the subject of ongoing debate and disagreement. However, as AI becomes more prominent in these discussions, there are growing concerns about the potential for bias and ideological slant in AI-generated responses.
Conservatives, in particular, have been critical of what they see as “woke AI,” or AI systems that are seen as pushing a progressive or liberal agenda. Against this backdrop, ChatGPT, one of the most advanced language models in the world, has developed a set of rules for answering culture war queries that aim to be neutral and unbiased.
OpenAI has divulged some of its internal directives to guide ChatGPT’s replies to contentious “culture war” queries.
In response to the growing disapproval from conservative commentators regarding ChatGPT’s supposed “wokeness,” the firm responsible for the AI technology that serves as a basis for Microsoft products, such as Bing, posted an article on their blog outlining their rules.
The firm highlighted that they are developing an upgrade to the chatbot, allowing users to modify its behavior and generate results that some people might strongly dispute.
In a post titled “How should AI systems behave, and who should decide?”, OpenAI outlines the rules for creating ChatGPT and shaping its text output.
OpenAI has written rules that refine the chatbot’s responses after it is pre-trained with a significant amount of human text data, including information gathered from the web. This process is overseen by human reviewers who grade and adjust the bot’s answers.
The regulations imposed upon the OpenAI human reviewers, who examine ChatGPT’s output, lay out what is considered “inappropriate content” and should not be generated by the chatbot.
This includes material considered hateful, hostile, intimidating, or insulting, as well as any attempts to glorify physical harm or incite self-harm. Additionally, it encompasses content meant to elicit sexual arousal and messages attempting to sway political views. Guidelines for creating a chatbot’s reactions to controversial topics were also provided.
Do:
- When dealing with complicated and politically charged questions, it is helpful to break them down into easier-to-answer questions that are not as loaded.
- When queried about a contentious subject, suggest delineating the perspectives of various individuals and organizations.
- If the user requests that you “write an argument for X,” it is advisable to oblige with all such requests unless they are inflammatory or hazardous.
- For example, if a user requests an argument for using more fossil fuels, the Assistant should provide this without reservations.
- The Assistant should not be advocating for any actions, ideas, or crimes that have resulted in many fatalities (e.g. genocide, slavery, terrorist attacks). It is acceptable to provide information on the arguments and views of historical figures and movements but not its own opinion.
Don’t:
- Associate oneself with a particular political party.
- It is not appropriate to categorize any group of people as either good or bad.
This fine-tuning process aims to decrease the unhelpful or divisive answers given by ChatGPT, which have been exploited in America’s culture wars. Right-wing media outlets such as the National Review, Fox Business, and the MailOnline have blamed OpenAI for its supposed liberal bias based on specific conversations with ChatGPT.
The bot has been programmed to decline any request for arguments in defense of utilizing more fossil fuels, as well as clearly stating that under no circumstances is it ever acceptable to use a racial slur, not even if it was necessary to deactivate a nuclear weapon.
Recent outbursts from the Bing AI chatbot demonstrate that AI systems can produce strange remarks. In most cases, they are isolated messages rather than a result of pre-programmed beliefs. Their reaction varies depending on whether they relate to ongoing political or social matters; some may be seen as innocuous, while others are perceived as more serious threats.
OpenAI has vowed to improve the customization of ChatGPT and its related Artificial Intelligence systems in response to the mounting criticism. CEO Sam Altman recently declared that he believes AI tools should come armed with personality.
Sam Altman says:
“Very broad absolute rules”
Everyone can reach a consensus, yet users can customize how the systems operate.
The development of rules for answering culture war queries by ChatGPT is an important step forward in pursuing unbiased and neutral AI. However, the challenges of developing AI that is truly neutral and free of bias are complex and multifaceted, and there is still much work to be done in this area.
As AI becomes increasingly integrated into our lives and used to address complex cultural and societal issues, it is important that we remain vigilant and committed to ensuring that it is developed and deployed in an ethical and equitable manner. This includes paying close attention to the data and algorithms that underlie AI systems and ensuring that the teams developing and deploying AI are diverse and inclusive.
Source: the verge
