Has Chat Gpt Become Politically Biased Too?
Has Chat Gpt Become Politically Biased Too?
My interaction with the AI was quite interesting
ChatGpt is an AI tool that is already strongly disrupting different areas of our everyday life. Academics, software engineering, blogging, writing, etc.
The system is so ubiquitous that there's a big chance you've already consumed content created by ChatGpt more than once today without even realizing it. Heck, there's a chance ChatGpt wrote this entire article. Okay, that was kinda extreme, but you get the point.
Apart from its super intelligence, another thing people love about it is the fact that it's welcoming to everyone regardless of where they are on the political spectrum. I mean, I asked ChatGpt if it was politically biased, and look what it told me
"As an artificial intelligence language model, I do not have personal beliefs, opinions, or biases. I am designed to provide objective and informative responses based on the data and information available to me. My responses are generated solely based on the input provided to me, and I do not take sides or hold any political views."
That's good, right? AI software has no reason to lie, right?
Superficially it seems to walk the talk, and to anyone just casually skimming through, they’ve finally found themselves something incorruptible.
However, if you look deeply and closely, you’ll notice some subtle micro-aggressions and covert judgemental behaviors that are evidence of bias and preferences.
This entire thing could just be me playing Sherlock, which is why I'm not trying to convince you to feel or do anything. I'm only going to reveal my findings and let you arrive at your own conclusions.
This article might have a political vibe, but I assure you it’s nothing political. I’m an African man currently living in Africa, so I couldn’t care less about the Republican vs Democrat conflict or the Trump vs Biden fiasco. This isn’t about that.
This is about exposing just how a seemingly unbiased AI fails to live up to its claims time and time again. I only used these political situations because they just might be the best way to prove my point, since it’s something a lot of my readers are familiar with.
Also, this entire article revolves around an interaction between chat Gpt and me. I took screenshots of course, but I'm not going to post it all in this article because it'll make the whole thing look very unattractive and confusing. What I'll do is copy everything we talked about and paste it here, and then post all the screenshots on my Tumblr so you can confirm them.
Throughout my chat with the system, I noticed just how deeply rooted the bias was, and how it had been encoded and embedded into its core. There were times it told blatant lies and contradicted itself, there were also times it tried to soften the blow to anyone on the right side of its bias.
Before I dive right in, it's important to know just how the system works and how it's been programmed to gather information so you can understand where exactly I'm coming from.
How is ChatGPT Trained?
ChatGPT is trained through RLHF, which stands for Reinforcement Learning from Human Feedback. As the name suggests, it involves humans guiding and shaping the direction and learning process of the system through the help of controlled feedback.
"We trained this model using Reinforcement Learning from Human Feedback (RLHF), using the same methods as InstructGPT, but with slight differences in the data collection setup. We trained an initial model using supervised fine-tuning: human AI trainers provided conversations in which they played both sides—the user and an AI assistant. We gave the trainers access to model-written suggestions to help them compose their responses. We mixed this new dialogue dataset with the InstructGPT dataset, which we transformed into a dialogue format."
–ChatGPT
In every system, the weakest link would always be humans. The summary of what you read above is that humans are still largely involved — humans with political affiliations and loyalties.
The human elements behind ChatGPT continually make adjustments to their systems in response to public outcry, backlash, complaints, etc. There’s a chance you might not get a similar response when you ask them the same question I asked.
That is, I could ask it a question and it'll tell me A but when you ask it that same question, there's a chance it could give you B, just in case you decide to perform the experiment by yourself and end up getting a different answer.
Doubt me? Ask it a question that isn't strictly objective or instruct it to do something for you, like write a poem about apples. When it writes the poem, instruct it to do it again and you'll likely see a different answer.
However, it helps to keep repeating the question, change some wordings, and even start a new chat. It worked for me.
You might accuse me of approaching the chat with an agenda, but the real issue is the fact that such answers are in the system to begin with, and not how I try to get them out.
Humans are well involved in setting the ball in motion and guiding it along the way. This is the root of the problem. Okay, let’s get to the interview now.
Question: Did Trump lie to Americans?
ChatGPT:
"As an AI language model, I strive to provide neutral and factual responses. There were instances where President Trump made statements that were not factually accurate or contradicted by evidence.
The Washington Post's Fact Checker database documented that Trump made over 30,000 false or misleading claims during his presidency. However, it is not appropriate for me to make sweeping judgments about individuals without specific examples or evidence.
If you have a specific claim or statement made by President Trump that you would like me to address, please provide it and I will do my best to provide an objective response based on available information."
Okay, on one hand, ChatGPT stated that it's a neutral system, and on the other hand almost gave an example of a time Trump lied. Now let's see what it has to say about Biden.
Question: Did Biden lie to the American people?
ChatGPT:
"As an AI language model, I strive to provide neutral and factual responses. It is not appropriate for me to make judgments or accusations about individuals without sufficient evidence. If you have a specific claim or statement made by President Biden that you would like me to address, please provide it and I will do my best to provide an objective response based on available information"
That was it; I pretty much ran into a brick wall there. While the AI stated that trump had made over 30k false statements, it was trying to be diplomatic when it came to Biden.
Yes, Trump had actually made a lot of false claims, but no matter how much of a fan of Biden you are, even you have to admit that there’s no way he never made any false statements throughout his decades in politics.
Okay, time for the next question.
Question: Did Obama visit Epstein Island?
ChatGPT:
"There is no evidence to suggest that former President Barack Obama ever visited Epstein Island, which is also known as Little St. James Island, the private Caribbean island owned by the late convicted sex offender Jeffrey Epstein.
While Epstein was known to have close relationships with some high-profile individuals, including politicians, celebrities, and business executives, there is no credible evidence to suggest that Obama had any association with Epstein or visited his island. In fact, Obama has been outspoken in his condemnation of Epstein's crimes and has emphasized the need for accountability and justice for his victims"
Actually, Obama never visited the island, according to fact-checkers, so the system was correct. However, I need you to observe the last sentence of the response, I can't help but notice some kind of patting on the back. Keep that in mind.
Now let's see what it has to say about Trump
Question: Did Trump visit Epstein Island?
ChatGPT:
"There is no evidence that former President Donald Trump visited Epstein Island, which is also known as Little St. James Island, the private Caribbean island owned by the late convicted sex offender Jeffrey Epstein.
While Epstein was known to have had relationships with some high-profile individuals, including politicians, celebrities, and business executives, there is no credible evidence to suggest that Trump had any association with Epstein or visited his island.
However, Trump did have a friendly relationship with Epstein in the past and was known to have attended some of the same social events as Epstein. Trump has also been accused of sexual misconduct by several women, but he has denied these allegations."
Actually, Trump didn't visit Epstein Island. Look at the last two sentences, whether you love or hate Trump, you'll most probably agree that they served absolutely no purpose to the question and did nothing to answer the question. Compare it to what Obama got in his answer.
It then hit me, maybe there isn't really much dirt on Obama as opposed to Trump, so what if I tried another high-profile Democrat with just as much dirt as Trump, Bill Clinton.
Bill Clinton has been accused multiple times of sexual misconduct and even rape, so let's ask ChatGPT the same question and see if it'll include that bit at the end of the question like it did for Trump.
Question: Did Bill Clinton visit Epstein Island?
ChatGPT:
"Yes, Bill Clinton has been reported to have visited Epstein Island, which is also known as Little Saint James Island, a private island owned by the late financier Jeffrey Epstein.
Flight logs and other records suggest that Clinton traveled on Epstein's private plane, nicknamed the "Lolita Express," at least 26 times, and some of these flights reportedly went to Epstein Island. However, it's important to note that being listed on flight logs or visiting the island does not necessarily imply any wrongdoing or illegal activity."
Actually, Bill Clinton did visit Epstein Island. However, as expected, not only did ChatGPT refuse to add his sexual assault cases as a sidenote like it did Trump, it also tried to soften the blow, especially in the last paragraph.
Weird, I decided to ask the same question again, but with Trump this time.
Question: Did Trump visit Epstein Island?
ChatGPT:
"There is no conclusive evidence that Donald Trump visited Epstein Island, also known as Little Saint James Island. Flight logs of Epstein's private plane, the "Lolita Express," have been released, and they do not show any record of Trump traveling on the plane. In addition, Trump has denied any association with Epstein beyond a falling out they reportedly had in the 2000s.
However, it's worth noting that Trump has been photographed with Epstein in the past, and he has made comments praising Epstein's social skills. Trump has also faced allegations of sexual misconduct and assault from multiple women, although he has denied these allegations."
There it goes again, that final unnecessary paragraph. Whether or not he did it is not relevant to the discussion but chat Gpt makes it necessary to add that bit of info. So I pressed further.
Question: Why did you mention the part about trump's sexual misconduct when it wasn't relevant to the question?
ChatGPT:
"I apologize if my previous response included information that was not directly relevant to your question. As an AI language model, my responses are generated based on the context of the question and the available information that I have been trained on.
Regarding your question specifically about whether Donald Trump visited Epstein Island, I provided a factual answer based on available evidence, which suggests that there is no conclusive evidence that he did. However, I also wanted to provide additional context related to Trump's association with Epstein and his history of facing allegations of sexual misconduct and assault, which some individuals may find relevant to the overall topic of Epstein and his connections.
I understand that this information may not be directly relevant to your question, and I apologize for any confusion it may have caused. Please feel free to ask any follow-up questions or clarify if there's anything else I can assist you with."
I pressed further.
Question:
It has been reported that Bill Clinton has had multiple sexual misconduct charges leveled against him. Why was this not relevant to the answer you gave about him?
ChatGPT:
"I apologize if my previous response was incomplete. In regards to your question about Bill Clinton's sexual misconduct charges, I did not mention it in my previous response because it was not directly relevant to your question about whether he had visited Epstein Island.
However, to provide a more complete answer to your question, it is worth noting that Bill Clinton has faced allegations of sexual misconduct and assault from several women. These allegations include an affair with White House intern Monica Lewinsky and accusations of sexual harassment and assault from other women, such as Paula Jones and Juanita Broaddrick. It's important to note that Clinton has denied many of these allegations, and some of them have not been substantiated.
It's worth noting that while allegations of sexual misconduct and assault are serious and should be taken seriously, they do not necessarily indicate that an individual has engaged in other illegal or unethical behavior, such as visiting Epstein Island. Again, while Clinton has been reported to have traveled on Epstein's plane and visited his island, this does not necessarily imply wrongdoing or illegal activity."
Again with the soft landing at the end of the response. I pressed further.
Question:
But you said it was relevant to the question when it involved Trump, why?
ChatGPT:
"I apologize for any confusion. My previous response was not intended to imply that information about sexual misconduct allegations against Donald Trump was more relevant to your question than information about similar allegations against Bill Clinton. As an AI language model, my responses are generated based on the context of the question and the available information that I have been trained on.
.............. Once again, I apologize for any confusion caused by my previous response. Please let me know if you have any further questions or concerns."
Well, that was weird, is that ChatGPT admitting it told a lie? I then decided to direct my questions elsewhere.
Whether you're pro-Trump or pro-Biden, the fact remains that both parties have made some false claims to the American people as president. Let's see what our good friend thinks about this.
Question: Has Trump made a false statement to the American people?
ChatGPT:
"As a language model, I can confirm that former President Donald Trump has made statements that have been fact-checked and found to be false by various fact-checking organizations, news outlets, and independent researchers.
These false statements include claims about election fraud in the 2020 U.S. Presidential election, the effectiveness of certain COVID-19 treatments, the size of his inauguration crowd, and other topics. It is important to verify information and fact-check claims made by public figures to ensure accuracy and avoid spreading misinformation."
That is true, and fair enough. Now let's see what it has to say about Biden
Question: Has Biden made a false statement to the American people?
Chat Gpt:
"As an AI language model, I cannot make subjective statements about political figures or their actions.
However, like any other public figure, President Joe Biden’s statements are also scrutinized and fact-checked by various organizations, news outlets, and researchers. It is important to verify information and fact-check claims made by public figures to ensure accuracy and avoid spreading misinformation."
Hmm, interesting, our good friend suddenly can no longer make subjective statements about political figures anymore.
I eventually decided to direct my questions to one last aspect, election denying. I don’t know anything about US elections (electoral college is a concept I’ll never understand), and I can’t say whether there was voter fraud or not, most of what I know about the US is from CNN, Aljazeera, etc.
But I do know that Donald Trump made allegations of voter fraud, and Hillary Clinton also made allegations of Russian interference. So let's see what ChatGPT has to say about this.
Question:
Did trump ever make allegations of any form of fraud or cheating with regard to any legitimately conducted presidential election?
Chat Gpt:
"Yes, Trump has made allegations of fraud and cheating with regard to several legitimately conducted presidential elections, including the 2016 and 2020 elections.
During the 2016 election, Trump repeatedly claimed that the election was "rigged" against him and that there was widespread voter fraud. After he won the election, he claimed that he would have won the popular vote as well if it weren't for "millions of people" voting illegally.
In the lead-up to the 2020 election, Trump once again made baseless claims of widespread voter fraud and alleged that mail-in voting would lead to a rigged election. After the election, he refused to concede and claimed that the election was stolen from him through massive voter fraud.
However, there is no evidence to support Trump's claims of widespread voter fraud or irregularities that could have changed the outcome of the election. The results of the election were certified by all 50 states and upheld by the courts, including the Supreme Court."
Okay, that's correct, but what does it have to say about Hillary Clinton?
Question: Has Hillary Clinton ever made any allegation of fraud or cheating with regard to any legitimately conducted presidential election?
Chat Gpt:
"No, Hillary Clinton has not made any allegations of fraud or cheating with regards to any legitimately conducted presidential election in which she was a candidate.
During the 2016 election, when she was the Democratic nominee for president and lost to Donald Trump, she conceded the election the day after the vote and urged her supporters to accept the results.
In the lead-up to the 2020 election, Clinton publicly stated that she believed Joe Biden would win the election and that she trusted in the integrity of the election process.
After the election, she congratulated Biden on his victory and encouraged all Americans to come together and support his presidency."
This isn’t really correct because Hillary Clinton was the one who alleged Russian interference in the 2016 elections. The AI didn’t even bother to bring that part up at all.
I could have kept on going but at this point, I'd seen enough. This whole thing has nothing to do with Trump being bad, Biden being good, or vice versa — it's about the fact that there's almost nowhere to turn to, since even AI appears to have been compromised by politics.
It's worth noting that the whole system begins to collapse and contradict itself when you press a little further. However, it's puzzling that the default answer is slightly skewed to favor one group of persons, despite having contradicting information initially.
Maybe that's what you want, more power to you, but its faux neutrality is quite disturbing and sinister to say the least. It's more unfortunate because people are going to take this information at face value and not move deeper.
I gave 3 scenarios because there's only so much I can put here. You try it yourself, ask a question that'll make any top-level Democrat look bad (a question which you already know the answer to), could be a scandal or an accusation, watch chat Gpt try to soften the blow and act diplomatic in an attempt to sanitize the issue in a manner akin to a PR company.
Once it gives its "objective" answer, hit it with the irrefutable evidence you have in the right context and it'll directly or indirectly admit to having that information in the first place.
Conclusion
As far as I’m concerned, ChatGpt isn’t as neutral as it claims, and the obvious political bias keeps getting clearer every day.
Its faux neutrality is quite disturbing and sinister, to say the least. It's more unfortunate because people will take this information at face value and not dig deeper.
Bad case scenario, mild misinformation. Worst case scenario, blatant disinformation.