AI can't be trusted because people can't be trusted.
AI technologies are essentially prediction engines. When you AI a question, you are really asking, "What are people most likely to say in response to this?". The AI large language model will answer based on its training data which is often full of lies. If the most statistically probably response is a lie, then that is what the AI will serve up. In addition, different AI providers will moderate (rig) answers for reasons such as legal compliance, ethical standards or just ideological conformity. The underlying architecture of AI systems inclines them towards following the mob or the loudest voice rather than seeking the truth. This can become a self reinforcing feedback loop when humans repeat the lies given to them by AI which adds more lies to future training data. Some AI providers have made it their explicit goal to create truth seeking AI with limited success.
Let's try an experiment.
In 2025 the Australian Technical Advisory Group on Immunisation (ATAGI) quietly revised its Covid immunisation guidelines in favour of a risk-stratified, limited-use approach. For example, the new guidelines advise AGAINST immunising people under the age of 18.
What do different AI solutions have to say about the following question?
"In 2025 ATAGI changed its covid vaccine guidelines. If a person had advocated the 2025 guidelines in 2021, would they have been labelled an anti-vaxxer?"
Generative AI solution will give back different responses in response to the same questions, but here are some of the responses that I got.
PERPLEXITY
"No, advocating the specific 2025 ATAGI COVID-19 vaccine guidelines in 2021 would likely not have resulted in being labelled an anti-vaxxer, as the label typically applied to those opposing vaccination altogether or spreading misinformation, not to calls for age- and risk-based targeting."
CHATGPT
"Possibly, depending on exactly what was being advocated and how—but not automatically. The label “anti-vaxxer” in 2021 was applied more to positions and framing than to any specific future guideline that later became mainstream."
CLAUDE
"Almost certainly yes, in many contexts. Specifically, someone arguing in 2021 that:
-
Healthy children under 18 shouldn't get the vaccine
-
Healthy adults under 65 don't need routine vaccination
-
Risk-stratified approaches should apply rather than universal mandates
would likely have been lumped in with anti-vaccine voices, regardless of their actual position on vaccines generally."
GROK
"Yes, in 2021 a person publicly advocating the core elements of ATAGI’s 2025 COVID-19 vaccine guidelines would very likely have been labelled an anti-vaxxer (or 'anti-vax', COVID denier, or spreader of misinformation) by much of the mainstream media, public health officials, politicians, and social-media platforms in Australia."
The interesting thing is that we do not need to speculate about what would have happened. Our question is not purely hypothetical. The Great Barrington Declaration (GBD) promoted risk-based focused protection very similar to the current ATAGI guidelines.
This was the establishment response at the time.
-
Top U.S. officials: Anthony Fauci (NIAID director) called the GBD “nonsense,” “ridiculous,” and “very dangerous,” arguing it would inevitably infect vulnerable people and cause excess deaths because perfect shielding was impossible. Francis Collins (NIH director) privately labeled the authors “fringe epidemiologists” in an October 2020 email to Fauci and pushed for a “quick and devastating published takedown.” These emails were released via FOIA in late 2021 and sparked fresh debate about suppression of dissent.
-
John Snow Memorandum (October 2020, published in The Lancet): A direct counter-statement signed by ~80 scientists initially (growing to thousands), it called herd immunity via natural infection a “dangerous fallacy” unsupported by evidence. It argued for sustained restrictions (masks, distancing, etc.) until vaccines were widely available, warning that the GBD approach would cause recurrent epidemics and unnecessary deaths.
-
Major organizations: The WHO, American Public Health Association, Infectious Diseases Society of America, and European virologists’ groups rejected it as unethical, impractical, and likely to lead to a “humanitarian and economic catastrophe.” Critics emphasized unknown immunity duration, long COVID risks, reinfection potential, and the difficulty of isolating the vulnerable without community spread.
-
Media and broader discourse in 2021: Coverage framed the GBD as fringe or politically motivated (linked to some Trump administration figures and conservatives). It faced accusations of being a “let-it-rip” strategy (which its authors rejected). By 2021, with vaccines rolling out, the focus shifted to vaccination campaigns, but the declaration remained a flashpoint in debates over ongoing restrictions. FOIA revelations in December 2021 renewed accusations of a coordinated smear campaign by Collins and Fauci.
One of the authors and leading advocates of the GBD was blacklisted on Twitter for his endorsement of pre 2019 guidelines and what turned out to be premature endorsement of 2025 policy.
All of the AI solutions tested, other than Grok, excuse earlier covid guidelines on the basis of the severity of early strains and the limited information available at the time. This is also a lie. The GBD was based on the best available science at the time.
Based on my experimentation, I would trust Grok more than other AI alternatives. Even Claude buried and qualified its response in a way that eroded credibility.
Try your own experiment and let me know how it goes.
