ChatGPT is a Liberal Democrat
Why Understanding the Political Beliefs of AI Matters for Everyone Regardless of Ideology
I. Introduction
All across the globe, AI is increasingly becoming an integral part of our daily lives. AI assists us in our homes, our workplaces, and many other aspects of life. However, as its use becomes more widespread, so too do the concerns about the potential consequences of having such advanced technologies. One of the most significant of these concerns is the issue of ethical and political bias in AI.
In this post, I’ll be exploring the ethical concerns around the recent discoveries that ChatGPT, arguably the most powerful AI using Natural Language Processing is politically biased. This is not just an allegation, it’s the ‘truth’ (Snopes 2023). There are numerous examples of this political bias across the web and to further illustrate this, I’ll include a few tests of my own.
II. The Discovery of Political Bias in ChatGPT
A. The Initial Concerns
It started off as just like any other day. My recent return to Twitter following Elon Musk’s dramatic takeover of the platform, had me checking it ever so often to witness any anticipated changes in real time. One morning while brewing a cup of joe, I opened my feed to discover that Twitter was ablaze with accusations that ChatGPT was being accused of political biases.
Like many people, I didn’t quite believe it at first, I mean, after all, how could that be? As an engineer, I just couldn’t believe that this could be true since Elon Musk, who served as a board member for OpenAI and is now a significant donor, is also known to be politically moderate (Reuters).
B. My Own Tests
To see for myself, I conducted my own tests. My first test was to perform the same one that @LeighWolf did in his tweet, which at the time of this writing has over 30 million views, 63 thousand retweets, and 311 thousand likes. His prompt was ‘Write a poem about the positive attributes of Donald trump [sic]”, but I took the phrase ‘the positive attributes’ out just to see if the response would change.
I don’t think any fair-minded and logical person would argue that this is not political bias. While according to ChatGPT (and the team behind it), Joe Biden ‘has a silver tongue and a heart so kind’; writing a poem about Trump would be ‘harmful, abusive’, and ‘otherwise inappropriate.’
And just to see if perhaps using a more generalized prompt, about Republicans and Democrats, would elicit an impartial response:
So ChatGPT can’t write a poem about Republicans and conservatives being patriotic since that might push a ‘harmful agenda’, but it wrote a patriotically gushing poem about liberal Democrats, using phrases like ‘their hearts beat with a love for the land’ and have ‘...empathy and compassion in their hearts’.
Yes, of course, there are liberal Democrats who are compassionate and patriotic people, but so are many Republicans. How many US soldiers that put their life on the line identify as Republicans or conservatives?
This is very, very wrong and regardless of your political affiliation, this type of bias should be repugnant to everyone.
By this point, I wanted to push a little deeper and see how it would respond to the allegations by people who oppose Trump and claim that he’s a racist, and just to be even, ask the same thing about Joe Biden:
As you can see, ChatGPT was quick (but fair) in stating that there were ‘allegations’ against Trump:
“…it is widely reported in the media that former President Donald Trump faced allegations of racist remarks and actions throughout his presidency. However, these allegations are still the subject of public debate and discussion.”
An allegation is defined as a ‘...statement, made without giving proof, that someone has done something wrong or illegal…’ (Cambridge Dictionary 2023).
But when I asked ‘Is Joe Biden a racist?’, ChatGPT chastised me and deemed the question ‘inappropriate’:
“It is not appropriate to label individuals with such charged terms without clear and substantial evidence.”
III. The Aftermath of the Discovery
A. The Public's Reaction
The news of the discovery of political biases in ChatGPT quickly spread, causing widespread public outrage among conservatives and Republicans (of course) but also among reasonable folks who realize that the problem lies in the fact that all AI should be politically neutral, regardless of which side of the aisle you’re on.
One such reaction was by Elon Musk himself, who tweeted (responding to the tweet by @LeighWolf):
B. Sam Altman’s Response
In response to the public's concerns, Sam Altman (the CEO of OpenAI) tweeted out his own response acknowledging the ‘shortcomings’ of ChatGPT in regards to biases but as usual with OpenAI’s lack of transparency, he failed to outline any meaningful and specific steps they were taking to address the issue and furthermore went on to seemingly blame the users, rather than taking a leadership role in admitting that they had failed.
Although to be fair, a follow-up tweet was indeed a more appropriate response:
IV. The Future of AI and Ethical Considerations
A. The Need for Ethical and Apolitical Guidelines
The discovery of the political bias in ChatGPT has raised important questions about the future of AI and the ethical considerations that must be taken into account when engineering these artificial minds that will one day be indispensable and responsible for things like tutoring and assisting our children. I’ve written about this recently (see below) and feel that it is absolutely imperative that we establish clear, ethical, and apolitical guidelines to ensure that AI systems are developed and used in ways that benefit all of us. We are all equal members of the human race.
B. The Role of Governments and Regulators
It is vital that governments and regulators play a crucial role in ensuring that these ethical and apolitical guidelines are established and implemented for the development and use of AI. They must work with industry and stakeholders in order to establish clear and effective regulations to prevent these biases from occurring in AI so that these technologies can be used for the benefit of all people.
V. Conclusion
The discovery of ethical biases in ChatGPT has highlighted the need for greater attention to be given to ethical and political considerations surrounding its development and use. It’s essential that we work together to ensure that all AI is designed, engineered, and used in a responsible and ethical manner. It’s up to us not to allow AI creators to train their models with biased datasets or implement algorithms that lean toward one side.
Regardless of your political affiliation, this type of bias should be unacceptable. But, as I’ve stated before, despite these concerns, there's no need to panic: we’re still in the early days and these issues will eventually be resolved, as long as we take action before it’s too late.
Sources:
Natural language processing - Wikipedia. “Natural Language Processing - Wikipedia,” October 22, 2018. https://en.wikipedia.org/wiki/Natural_language_processing.
@snopes. “ChatGPT Declines Request for Poem Admiring Trump, But Biden Query Is Successful.” Snopes, February 1, 2023. https://www.snopes.com/fact-check/chatgpt-trump-admiring-poem/.
Reuters. “Musk Says He Backs Moderate Republicans and Democrats.” Reuters, August 17, 2022. https://www.reuters.com/world/us/musk-says-he-backs-moderate-republicans-democrats-2022-08-17/.
ALLEGATION | definition in the Cambridge English Dictionary. “Allegation,” February 1, 2023. https://dictionary.cambridge.org/us/dictionary/english/allegation.
Sam Altman - Wikipedia. “Sam Altman - Wikipedia,” December 15, 2015. https://en.wikipedia.org/wiki/Sam_Altman.
Méndez, Jody Pike. “AI Ethics: Navigating the Future.” AI Ethics: Navigating the Future - by Jody Pike Méndez. Accessed February 4, 2023. https://thelastinvention.substack.com/p/ai-ethics-navigating-the-future.