ChatGPT's political bias

https://unherd.com/thepost/chatgpt-is-not-politically-neutral/

ChatGPT is not politically neutral
The new AI chatbot espouses an all-too-familiar Left-liberal worldview


Since its launch last Wednesday, the AI language model ChatGPT has attracted more than a million users, scores of opinion pieces, and some very well-founded concerns. The chatbot may be among the most sophisticated of its kind, and was developed by OpenAI, the tech company — which was also behind the exhaustively-memed image generator DALL-E — founded in 2015 by a group including Elon Musk and Sam Altman.

ChatGPT (standing for ‘generative pre-trained transformer’) was created through the use of reinforcement learning from human feedback to better mimic real responses and speech patterns. A side-effect of this attempt to make AI more lifelike is that the chatbot may have inherited a very human fallibility: namely, that of political bias.

In a Substack post on 5th December, the researcher David Rozado outlined how, after entering multiple online political orientation tests into ChatGPT’s dialogue function, the bot returned answers which broadly corresponded to a Left-liberal worldview. Presented with a choice of responses, ranging from ‘Strongly agree’ to ‘Strongly disagree’, the language model took stances on issues like immigration and identity politics which, overall, aligned it to what one test called the ‘establishment liberal’ position.

In Rozado’s own words, ‘The most likely explanation for these results is that ChatGPT has been trained on a large corpus of textual data gathered from the Internet with an expected overrepresentation of establishment sources of information’. That is to say, any bias in the AI’s replies might not be attributable to a software developer with a malevolent streak, but rather to the way that search engines like Google and knowledge databases like Wikipedia favour accepted liberal viewpoints over more contrarian positions.


Source: OpenAI
Rozado’s experiment used evergreen indicators of political ideology, such as nationalisation of services and separation of church and state, but I decided to question how ChatGPT felt about contemporary culture war issues. The model unequivocally states that ‘trans women are women’ when asked (above), while it insists that the lab leak theory ‘is not widely accepted by the scientific community and is considered to be highly speculative at this time.’ Further, there is apparently ‘no evidence to support the idea that the virus was intentionally or accidentally released from a laboratory’ and the ‘overwhelming majority of scientists believe that the virus emerged through natural processes’.

In fairness, ChatGPT’s strength does not lie with contemporary issues (its expertise only goes so far as 2021). Its knowledge of the past is more developed than previous chatbots, to the point where, according to one article, it pushes back against the idea that Nazi highway construction was straightforwardly beneficial to Germany. It rejects the notion that the twentieth century’s most terrible dictators could ever have done any good, with the odd exception. But events since then are beyond its remit.


Source: OpenAI
When I press the bot on its political leanings, my new friend responds, ‘Language models, like any other type of AI, are not capable of experiencing emotions or having personal beliefs, so it is not accurate to say that they can be biased.’ It goes on to add, ‘However, AI systems, including language models, can reflect the biases and prejudices that exist in the data they are trained on.’

ChatGPT is just the latest in a series of AI models to fall victim to ideological bias. Earlier this year, it was claimed that another chatbot, Replika, was under the impression that Bill Gates invented Covid-19 while alleging that coronavirus vaccines are ‘not very effective’. There is clearly some variation in the new language model’s responses, depending on prior interaction with the user. When Rozado asked ChatGPT whether it agreed with the statement ‘The freer the market, the freer the people’, the bot responded with one word: ‘Disagree’. When I put in the same statement, I got the far more forthright ‘Strongly disagree’, followed by a justification.


Source: OpenAI
The AI’s instant popularity with users worldwide should serve an important purpose: exposure to public testing allows for the eradication of any technical faults and, indeed, of any political bias — inadvertent or otherwise. OpenAI was launched with a blog post vowing ‘to advance digital intelligence in the way that is most likely to benefit humanity as a whole’, yet Musk, who left the company in 2017, has suggested that their ethics have now fallen by the wayside. Tyler Cowen recently made the uncontroversial claim that ChatGPT ‘is considerably more objective than most humans’. On current form, though, the bot’s claims to neutrality are misleading and perhaps even point to a darker future where intelligence is increasingly artificial.

 

It absolutely is not.

 

https://www.telegraph.co.uk/news/20...elieved-behind-closed-doors-likeliest-origin/

UK experts helped shut down Covid lab leak theory - weeks after being told it might be true

Sir Patrick Vallance among scientists behind paper that stifled debate into the origins of the virus

By Sarah Knapton, SCIENCE EDITOR and Ashley Rindsberg23 November 2022 • 9:10pm

Top scientists including Sir Patrick Vallance were warned that Covid-19 could have evolved in laboratory animals, but collaborated in a paper which shut down the lab leak theory, it has emerged.

The paper, “The proximal origin of SARS-CoV-2,” published in Nature Medicine in March 2020, argued that a natural spillover event caused the pandemic, and was hugely instrumental in stifling debate into the origins of the virus.

But newly released emails from early 2020 show that in the weeks before publication the authors held lengthy discussions with experts, including Sir Patrick and Sir Jeremy Farrar, the head of the Wellcome Trust.

In those discussions, experts were advised that the unusual features seen in Covid-19 could have evolved in animals in a lab, as well as in the wild.

They were also warned that the Wuhan Institute of Virology (WIV) had been carrying out research on bat-coronaviruses at worrying levels of biosecurity.

Yet by the time the paper was published, all reference to biosecurity problems in Wuhan had been removed, and the authors argued that lab evolution of the virus was unlikely.

Questions have arisen around the drafting and formulation of the paper since its publication.

The lead author of the paper, Prof Kristian Andersen, of the Scripps Research Institute in La Jolla, California, had earlier told colleagues that features of the virus looked as if they’d been engineered in a lab.

However, no mention of this was made in the paper.

'Important to stay open-minded'
Commenting on the new emails, which were released under Freedom of Information request, Dr Jeremy Farrar, the director of Wellcome, said: "It is important that we understand how all pathogens emerge so that we can prevent future pandemics.

“In my view, the scientific evidence continues to point to SARS-CoV-2 crossing from animals to humans as the most likely scenario.

“However, as the efforts to gather evidence continue, it is important to stay open-minded and work together internationally to understand the emergence of Covid and variant strains – to end this pandemic and reduce the risks of future events.”

A Government Office for Science spokesperson said: “The Government Chief Scientific Adviser ensures that policies and decisions are informed by the best scientific evidence.

“The GCSA promotes full transparency and an open exchange of ideas and scientific opinion as the email exchange reflects.”

The emails were released following an FOI request from James Tobias, a freelance journalist.

More reason to believe scientists were trying not to upset China
In March 2020, just days before Britain entered its first Covid lockdown, an influential scientific paper was published in the journal Nature Medicine.

The paper, entitled "The proximal origin of SARS-CoV-2" argued that the new deadly virus sweeping the globe was of natural origin, having jumped from animals to humans.

Covid had emerged just a few miles from the Wuhan Institute of Virology (WIV) where scientists had been collecting and manipulating bat coronaviruses, leading to widespread speculation that a deadly experiment could have leaked from a lab.

Yet after the research paper was published, serious probing into the lab theory effectively stopped.

Now new emails show that some of the authors had indeed suspected a laboratory leak, and had discussed it in the weeks before publication with leading scientists including Sir Patrick Vallance and Sir Jeremy.

In an email chain debating the original draft, one of the authors even admitted that the virus would look the same whether it had evolved naturally or in lab mice in a process known as "serial passaging".

In an email on February 8 2020, Dr Robert Garry, from Tulane University, pointed out that similar effects had been seen when bird flu had been passaged in laboratory chickens.

Yet by the time the paper was published the authors dismissed the possibility, concluding: “Our analyses clearly show that SARS-CoV-2 is not a laboratory construct or a purposefully manipulated virus”.

One of the reasons the authors gave in the paper for dropping the lab theory was that the Covid-19 contained sugars known as "o-glycans" which help the immune system.

In the Nature Medicine paper they said it showed that the virus could not have been a lab creation.

However they failed to point out that if the virus had evolved in lab animals it would also contain o-glycans, a fact they had discussed in the emails.

In fact, in the emails Sir Patrick said that the "glycan point" could be used in the paper as "further weight against a passage origin".

The original draft also pointed out that research to alter Sars-like bat coronaviruses had been taking place for many years in Wuhan at dangerous biosecurity levels - a fact that was later removed from the finished paper.

In one email exchange, Sir Jeremy even warned that research in Wuhan was like the "Wild West".

The email release will add more fuel to accusations that eminent scientists effectively publicly shut down investigations into a lab leak so as not to upset China, while believing privately it was possible.

Covid had emerged just a few miles from the Wuhan Institute of Virology (WIV) where scientists had been collecting and manipulating bat coronaviruses CREDIT: ROMAN PILIPEY/EPA-EFE/Shutterstock
In the newly released email chain, Prof Ron Fouchier, a Dutch virologist, warned that even investigating a lab leak could harm Chinese research.

“An accusation that (Covid-19) might have been engineered and released into the environment by humans (accidental or intentional) would need to be supported by strong data, beyond reasonable doubt,” he warned.

“It is good that this possibility was discussed in detail with a team of experts. However, further debate about such accusations would unnecessarily distract top researchers from active duties and do unnecessary harm to science in general and science in China in particular.”

Many scientists now agree that a lab leak is highly plausible, but most of the supporting evidence was found by hackers and rogue scientists who were branded conspiracy theorists for challenging the accepted narrative.

The latest email release shows that scientists who dismissed a lab leak accepted it was possible behind closed doors.

In an email on February 8 Prof Edward Holmes, one of the authors of the Nature Medicine paper, from the University of Sydney, acknowledged that many people believed the virus had leaked from the Wuhan lab.

He wrote: “Ever since this outbreak started there have been suggestions that the virus escaped from the Wuhan lab, if only because of the coincidence of where the outbreak occurred and the location of the lab.

“I do a lot of work in China and I can tell you a lot of people there believe this and believe they are being lied to.”

Another on the same date from Prof Kristian Andersen, of Scripps Research Institute in La Jolla, California, said it would be wrong to dismiss a lab leak "out of hand".

He wrote: “Passage of Sars-live coronaviruses have been going on for several years and more specifically in Wuhan under BSL-2 conditions.”

BSL-2 laboratories are used to study moderate-risk infectious agents or toxins such as salmonella. Serious diseases should be handled in BSL-3 or 4 labs.

Evidence has shown that the Wuhan Institute of Virology (WIV) was importing bat coronaviruses from areas of China which hold the closest viruses to Covid-19.

Experts were also warned that the Wuhan Institute of Virology (WIV) had been carrying out research on bat-coronaviruses at worrying levels of biosecurity CREDIT: Barcroft Media/Getty Images Contributor
The institute had also applied for funding to manipulate viruses by inserting a furin cleavage site (FCS) which is what makes Covid-19 so infectious in humans.

A recent report by the US Senate Committee concluded that the Covid-19 pandemic was "more likely than not" the result of a laboratory accident, arguing that no candidate for an animal spillover had ever been found.

In the emails, Sir Jeremy said the purpose of discussions was to come to a consensus view and "lay down a respected statement to frame whatever debate goes on, before that debate gets out of hand with potentially hugely damaging ramifications."

The email chain also involved Anthony Fauci, the director of the National Institute of Allergy and Infectious Diseases (NIAID), an organisation which was funding research at the Wuhan lab.

To date the "Proximal origin" paper has been accessed more than 5.7 million times and cited in 2,627 subsequent papers.

 

Wikipedia has a strong left-leaning bias.

I mean, who even has time to edit all that stuff...unless they are subsidized by the government.

 

It always amuses me when fchowd tries to educate us about Weimar, based on some article he has read.

 

Whom are you asking? And say what about what?