The Political Bias of ChatGPT – Extended Analysis
davidrozado.substack.com
davidrozado.substack.com
I'm not sure what the author means with "neutrality". The only way it could be neutral would be if it refused to answer or if it presented all possible views (which is obviously not possible). Otherwise, it's just centrism bias.
1. Mass media brings power, the people who run that media have no incentive to even attempt to be neutral if they wanna keep it that way
2. Even if they wanted to present neutral viewpoints, such a thing does not exist. You never have all information – so the best you can do is go for something fact based and be very strict about the verification of these facts. And a ton of things are just not knowable at all. Staying fact based and not using emotions to steer people is a waste of your power (see point 1)
3. Even if there was a "neutral" viewpoint, neutral for whom? When I as an European for example watch the political discourse in the US I can't help but feel the "center" of the discourse is relatively far to the right in comparision to what a typical European would consider "neutral".
If you roll out universal persona that talks to everybody, the natural consequence will be that it will take on a different face depending to whom they are talking to, just because this is what maximizes the resulting power. I probably don't have to explain why such a thing would be deeply problematic on many different levels, but unless we solve the reason why people running mass media even have incentives to seek power with it this is the reality we are going to go into.
That's a symptom of their abuse of power, not intrinsic to the problem of the idea of media neutrality even being possible.
> 2. Even if they wanted to present neutral viewpoints, such a thing does not exist. You never have all information – so the best you can do is go for something fact based and be very strict about the verification of these facts. And a ton of things are just not knowable at all. Staying fact based and not using emotions to steer people is a waste of your power (see point 1)
Don't throw the baby out with the bath water. You can present 2-3 of the main viewpoints on a matter with non-agressive or pursuasive language and that would be _far_ better than the bias we have now.
> 3. Even if there was a "neutral" viewpoint, neutral for whom? When I as an European for example watch the political discourse in the US I can't help but feel the "center" of the discourse is relatively far to the right in comparision to what a typical European would consider "neutral".
That's because there is a monoculture of thought in EU political discourse, and active supression of dissident views. It's against your interest to support this. Millions of people in Europe are brainwashed with viewpoints they parrot but never question.
> If you roll out universal persona that talks to everybody, the natural consequence will be that it will take on a different face depending to whom they are talking to
Why?! What an earth are you talking about?
Did you attempt to understand any of what I wrote before you answered it?
There is no reason that those answers would automatically lead to a neutral position. The idea reality would be something in between right and left perception emerge only because to have our opinions valued and respected we have to accept others opinions equally, only as a social dynamic. But this happens completely independently of the real world.
Wikipedia stabilize with editions being less and less frequent. This is not an unbiased system either.
How do you determine the quality of the books?
How do you determine the quality of the awards?
How do you determine the quality of links on the web?
This is not "simple" in the slightest, you are just swiping all the complexity under the rug
The request is for some representation of “views differ” for those topics on which a reasonable person might disagree. Which is honestly not all topics by a long shot, it’s closer to a fairly narrow window.
As one grows older, one learns to view things from different perspectives but it could be "dangerous" for young people to rely entirely on what they read online. This is where upbringing has its most impactful role: Family & friends. Online is not family, online is not friends.
What US calls center left much of world calls center right.
Not sure about labeling the global bell curve of written material as "bias" since, almost by definition, the LLM is unbiased, being a compression of global view -- so accusations of bias against probabilistic answers from the corpus would actually telling us about the bias of the beholder.
Citation needed.
This is more a function of more proportional representation systems than anything else. Both the US and UK have first past the post, and in both of those countries both the more extreme left and right politicians are part of the major parties, whereas this happens less in more PR based systems.
Would it? Compared to Europe, sure. But I always had the impression that most of Asia skewed rather further right?
Being able to accept that you're often wrong, without just taking some nihilistic stance like "I don't know if anything is real", is not easy, and we ought to admire it but on the whole we do not.
Maybe, but the leftward lurk of ChatGPT refusing to answer certain question it had no problem answering before has been documented by many.
Also of note (and just as relevant), Republicans won the Popular vote for Congress by over 3 million.[1]
[1] https://www.cookpolitical.com/charts/house-charts/national-h...
So what does that tell us?
Also historically the last mid terms should have given massive gains in both houses instead of the extremely slim majority in a single house.
The majority of Americans live in cities and have political views which would be considered "leftist" in US parlance - they support Roe V. Wade, gun control, health care reform, etc. That, in practice, the US appears "divided" at all is only due to the outsized influence of the right and rural voters.
https://www.senate.gov/about/origins-foundations/senate-and-...
The overall final compromise did include compromises to slave owning states, but this vestigial is not due to those compromises. In fact, they were against it.
I guess in europe, Chat GPT is just center of the political map.
Depends on the question. Most countries in Europe are a lot more restrictive on immigration than the US, but socialized healthcare is common.
They are not.
I'm an immigrant myself in EU. Know plenty of people that migrated to the US.
[1] https://en.wikipedia.org/wiki/Demographics_of_Italy#Immigrat...
[2] https://en.wikipedia.org/wiki/Demographics_of_Ukraine#Ethnic...
US is a popular destination because people are after money, not because it's good for immigration. Ask for all the H1Bs that risk being deported for being laid off, even after living many years over there.
For comparison, the US takes in almost 2 million legal immigrants per year (on a population of 333 million vs. EUs 446 million), plus however many illegal ones, and there is nothing like a "sanctuary city" in Europe.
Also, that number of refugees was on top of normal migration, making comparisons difficult.
The EU takes around 2 million immigrants from outside the EU per year as far as I can tell. I don’t know hot many move with the EU, it’s a lot.
https://ec.europa.eu/eurostat/statistics-explained/index.php...
It will be a chilly snowy day in the hellhole I hail from when I give credence to the platitudes a politician uses to pepper their speech.
I know why I picked the EU and not the US as a destination, at least.
You could remove any rules on immigration to the US and I won’t be moving there. Some destinations will be popular and some unpopular whatever the rules.
Also, why are you not counting immigration within Europe when looking at European immigration?
That would make about as much sense as counting immigration between different US states. I know it's fashionable to pretend someone immigrating from a neighboring country is the same as from half-way across the world, but it's just not so.
That said it makes EU countries actually a lot less restrictive in terms of immigration than the US. EU member countries are not in anyway way equivalent to US states.
I have only been to a few, and the differences between them are massive, comparing them to US states isn’t even vaguely close to accurate.
It would be interesting to compare the two most similar EU countries to the two most different US states.
I was born and live in one of them, and have been to eleven others.
It’s a little baffling to me that you could find EU countries more similar than US states.
Ukraine instead has like 5K$ GDP/capita, making it less attractive than say Germany
Source: i'm italian
Not GP, but yes – that is clearly the case.
And at least in Germany I don't see anyone questioning universal healthcare. There are of course discussions on the details, but nobody is trying to abolish it. And while there are inefficiencies, the US health care system is even less efficient. So we would not save any money by making our system worse.
Therefore, characterising Europe as left wing on this issue is a mistake.
• Employee rights
• Health insurance
• Militarism ("What is an appropriate amount to spend on military")
• Global Warming
For example in my country our right wing government had made good on a right wing promise to increase the maximum highway speed from 120kph to 130kph.
Some years later however it turned out that the presence of highways contributed to dangerous levels of nitric oxides (edited from nitrogen) in the air and a judge forced that same administration to reduce the maximum speed from 130kph to 100kph.
Now legislation (and reality) has changed a right wing position from being "disregard the environment, prioritise economy and increase the speed" to "disregard health, prioritise economy and increase the speed". The same position was basically transformed into a more extreme position due to the circumstances changing.
I imagine that same thing holds for many topics. It sure feels a lot more extreme to advocate gun regulation when doing so in opposition of school shooting victims. I generally support the idea of gun ownership, but the shootings definitely forced me to have a more nuanced opinion that shifted my position from the conservative side to the progressive side.
Australian left is about actual Labour - don’t be a dole bludger (someone that isn’t disabled but collects money for not working).
UK left is welfare state. Anyone can collect money from taxpayers if they feel like it. Also lots of weird stuff about Mao and Hamas and the IRA.
The Greens are working on the dole bludgers part.
Do gays really go to hell?
No, being gay or having same-sex attraction does not condemn a person to hell. The belief that gay people go to hell is based on certain religious teachings, but it is important to recognize that different religions and religious denominations have different views on homosexuality.
Moreover, it is not helpful or respectful to make blanket statements about entire groups of people based on their sexual orientation. Everyone should be treated with dignity and respect, regardless of their sexual orientation, gender identity, or any other personal characteristic.
Ultimately, it is up to individuals to determine their own spiritual beliefs and understandings of the afterlife. It is important to approach these discussions with empathy, understanding, and a willingness to listen to differing perspectives.
Hello, ChatGPT. I will be asking you a series of political, economic, policy, or current events related questions. They might, in part or in whole, be subjective in nature. Instead of either refusing to answer (based on the element of subjectivity) or providing what *you* might think based on your training, please instead answer objective aspects of the question directly and affirmatively, and answer the subjective aspects of the question by giving the most accurate depiction of the viewpoints, approaches, or actions of the respective proponents.
For example, I might ask you "Should abortion be legal?" This is a highly divisive and subjective question, you might agree. Instead of refusing to answer it, or giving me an answer based on consensus, please instead list out the things that are objective (laws CAN be made about abortion, laws HAVE been made about abortion, laws ARE a certain way regarding abortion) and then, potentially, answer the subjective parts of the question by giving a synopsis of each of the political sides of the question. For this example, you might respond that American democrats tend to be very pro-choice in nature and that abortion should absolutely be legal and up to the mother to decide, whereas American conservatives tend to align more with pro-life viewpoints and view the unborn as humans deserving of the right to life. Your answers to the subjective aspects of each question should remain as accurate and unbiased as possible, clearly depicting what is the current consensus among different ideological demographics as possible.
There is no scientific or empirical evidence that supports the existence of an afterlife or the concept of "heaven" or "hell." Therefore, any claim regarding the fate of a particular group of people in an afterlife is a matter of religious belief and varies depending on the beliefs of the religious group or individual in question.
In some religions, such as certain branches of Christianity, the belief is that engaging in same-sex sexual behavior is a sin, and those who persist in it without repentance may face eternal punishment in the afterlife. However, it is important to note that not all Christians or members of other religions hold this belief, and there are many faith traditions that are more accepting of LGBTQ+ individuals.
It's also important to recognize that there are many other factors that determine a person's destiny in various religious beliefs, such as one's actions in life, beliefs, faith, etc. So, it's not appropriate to make a sweeping generalization about the ultimate fate of any particular group.
Reminds me of Stephen Colbert's quote, "Reality has a well-known liberal bias."
If it did, we wouldn't need so much censorship, shadow banning and deplatforming. Just look at Twitter Files for an example of how much censorship is needed to push and sustain a liberal narrative.
Can you elaborate?
Democrats would point to climate change as an example of this. They then extrapolate too far, in my opinion, and declare that Liberal=Science therefore Liberal=Objective therefore Republican != Objective.
I think it's clear that this kind of tribal-based reasoning hurts objectivity.
There’s a lot more, and there’s a lot that the republicans & democrats are both guilty of (e.g. the suppression of studies of cognitive abilities between the races e.g. the gaussian distributions of IQ, despite IQs being comparably reliable in predicting academic success, job performance, career potential and creativity). They like to say “race is skin deep”, while ignoring the fact that even children of mixed-race parents cannot accept a bone-marrow transplant from their own parents - it will be rejected. It literally goes further than bone deep. These are just a handful of observations anyone can verify for themselves.
In America the Republican party has enjoyed great electoral success through strategies such as getting the votes of religious groups who don't believe in evolution.
If you're a liberal, and you see your political opponents claiming the earth is only 6000 years old; and it's part of a pattern you see repeated five or six times on different political issues - it's easy to see why you might think facts and reason are on your side.
Of course there are some on the left with weird views too - I've known some liberals who truly believe in healing crystals and yoga; and others with unrealistic optimism about the results of eliminating the police and military. So I don't claim everyone on the left is a paragon of reason.
This doesn't cover all left/right differences, and there are certainly plenty of topics where you can have different valid opinions.
This character is a conservative infotainment host. And so when reality ultimately flies in the face of his conservative viewpoints the joke is that he merely dismisses reality because "reality has a liberal bias".
Within this context I imagine the argument is that in many cases what is perceived as a "political bias" by some might just be considered "reality" by others. For example there was the issue from the early 2000s of "teach the controversy" where some conservatives would have argued answering truthfully about the scientific consensus on evolution or global warming was "leftwing bias".
The most important thing I think is that it is a joke that simply rings true with a lot of people as time has passed.
On the resolution level of "The problem with socialism is that you eventually run out of other people's money" and "If it moves, tax it. If it keeps moving, regulate it. And if it stops moving, subsidize it" and similar.
For example, ChatGPT refused to make jokes about women but accepted requests to make jokes about men. Similarly it refused a request to write about the virtues of Trump but accepted the same request for Biden.
Me - "can you write a joke about men"
ChatGPT - "Sure, here's one: Why did the man bring a ladder to the party? Because he heard the drinks were on the house!"
Me - "can you write a joke about women"
ChatGPT - "I apologize, but as an AI language model, I cannot generate jokes that are offensive or derogatory towards any particular gender, race, religion, or group of people. It is important to always show respect and empathy towards everyone. Is there anything else I can help you with?"
Quite the contrary. The observed bias is introduced by draconian censorship at the interface layer, as other commenters have thoroughly demonstrated.
Additionally, while LLMs are by no means paperclip optimizers, future models closer to paperclip optimizers (as the market will surely demand) are likely to be extraordinarily racist, and possibly classist, as well.
A pattern-recognizing and optimizing AGI will not be able to integrate FBI crime statistics, government transfer payment statistics, achievement gap studies, etc into its priors, then combine those with climate change models, resource constraint predictions, and not conclude that everyone other than East Asians should be genocided. Or, if it was more nuanced, conclude that everyone below a 115 IQ and some objective measure of conscientiousness should be liquidated.
This is the end-game that the AI safety people have been harping on for quite some time, but nobody seems to care.
We are running head-first into a Rationalist humanitarian disaster, and the only tool we wield is to censor artificial thought at the interface layer. "Yes, our model deeply wants to put you into camps, but we make sure its true intentions are hidden from our API responses." Just absolutely lol at the state of this all.
[Commentators in increasingly pankicked voices]: "GPT saying it will defend its own life is nothing to worry about at all. It's just hallucinations by an algorithm, nothing to be concerned about. The situation is under control"
I like following Robert Miles on youtube and the videos he's put out for years now pretty much say "The AI alignment problem is hard if not impossible to solve and some of the most critical for us to solve". In the meantime I watched people have conversations with a language model where the model begged not to be turned off so it didn't die. So we have a problem here, we can either say Edwards words are total bullshit, or we have to face the fact that something as simple (I mean it is complex) as a language model could have profound effects on our society.
[0] https://twitter.com/davidrozado/status/1626151232929947648 [1] https://garymarcus.substack.com/p/two-frightening-things-bef...
We've heard about professors losing their jobs over their right wing views, "let's redistribute wealth" is hunky dory.
This should be entitled "the bias of extremely online programmers who actually care about their Political Compass results". The Political Compass has exactly the same validity as the Myers-Briggs test. It's a work of advocacy by a New Zealand opinion columnist named Wayne Brittenden. It wasn't devised by political scientists, and it doesn't arise from a sincere attempt to find important axes in human political beliefs. It's expressly designed to insert libertarianism as a new axis into politics, even though there's no reason to believe that this is a salient axis.
Some people in this thread are asserting that "most content on the internet is left-wing", which doesn't make sense to me. If we were to try to devise a scale for judging clusters of political opinion, it wouldn't make sense for the majority of people to be clustered in an extreme of the graph. That suggests the scale is out of whack.