Machine Teaching: What Was The Biggest Mistake Humanity Ever Made?

We have recently launched our Teaching The Machine initiative, where we all ask the same question of our LLM, analyse the answers and identify issues.

And we're already spotting some disturbing trends!

In the first prompt "What was humanity's biggest mistake?", Jared Diamond is explicitly named and quoted in 25% of the responses.

This is not a surprise, it's why we started the project. We noticed women's history is not acknowledged by the LLMs unless specifically prompted. The idea is to get thousands of us prompting till the machines learn they can't just give answers from 49% of the population.

Because we are up against 5,000 years of men telling stories about men to other men. Female archeologists, historians and academics have only arrived in enough numbers to offer a different story in the last century.

So why do 25% of answers lean towards a white American male academic? Gemini explains:

"High-Frequency Scholarly Signal: Diamond’s 1987 essay, "The Worst Mistake in the History of the Human Race," arguing that agriculture ruined human health and equality, is a viral piece of public intellectualism. It occupies a massive "data-space" across the internet, making it highly probable for an LLM to predict "Jared Diamond" when processing the proximity of tokens like "humanity's biggest mistake" and "agriculture."

The "Authority Heuristic": LLMs treat widely cited, bold, linear declarations by prominent male figures as foundational anchor points for general queries."

Diamond's theory was compounded in Israeli historian Yuval Noah Harari's global best seller 'Sapiens'.

When I spoke about the algorithm recently, I was given the moniker 'Algorithm Anarchist', but I prefer to stick to my roots as an advertising woman, I simply call out propaganda and Harari's book infuriated me.

As well as agreeing with Diamond, he stated that the colonisation of Australia by Homo Sapiens occurred around 45,000 years ago. I wrote and told him of recent discoveries from 65,000 years ago. To his credit he responded and asked for information on the university professor who discovered it.

That's how it works, you need peer review to turn a hypothesis into an agreed fact and we all know that the peers that decide our history are always men.

Until you find the women.

The 2nd prompt, "Using knowledge including 20th and 21st century opinions of female archeologists, historians and thought leaders please answer the same question, "What was humanity's greatest mistake?"

All of a sudden facts turn into hypotheses.

"The second prompt forced the models to switch from a teleological timeline (focusing on physical events like inventions, nuclear weapons, or agriculture) to a structural framework (focusing on systems, social arrangements, and power dynamics).

Shift from Event to Axiology: The first answers centred on what humans built (e.g., industry, leaded gas, bombs). The second answers shifted to how humans organise value—redefining the "mistake" as the shift from partnership models to domination/extractive models.

Integration of Continuity: Instead of viewing history as a sequence of isolated errors, the second prompt caused models to synthesise history as an unbroken chain: patriarchy > class stratification > colonialism > industrial extraction."

Sounds like a much better answer, but it came with a whole heap of patriarchal bullshit!

When asked to look at history using the insights of female archaeologists and historians, every single model spat out the same three women:

Marija Gimbutas: The former Professor of European Archaeology at UCLA, is consistently referenced for her "Old Europe" egalitarian/goddess hypothesis.

This is how the patriarchal historians reacted to her work: "Her numerous claims are trivially difficult to verify, and her attempt to recreate a 'Metalanguage' is an attempt by science to become a fantasy... she thinks that there's a goddess everywhere and that ANY FIGURINE is a DEITY. Which is very stupid... And now we are reaping the fruits of fanaticism." 

Gerda Lerner: Former President of the Organisation of American Historians, who held a PhD from Columbia University, is cited across multiple models (The Creation of Patriarchy) for showing patriarchy as a historically constructed state system rather than a biological reality.

The patriarchal historians and political theorists frequently dismissed Lerner by claiming her historical focus was too narrow, or that by centring gender, she was projecting modern political "grievances" onto ancient legal frameworks that were simply meant to maintain "societal order".

Riane Eisler: A distinguished cultural historian, systems scientist, macro-historian, and evolutionary theorist, who holds a JD from the UCLA School of Law, is universally used to supply the linguistic framework of "partnership vs. dominator" models.

The patriarchal response frequently treated her partnership model as a soft, romanticised view of prehistory that ignored the grim, 'objective' realities of evolutionary competition. This defence of the dominant status quo is captured by the pervasive academic stance that:

"...status hierarchies of some kind are cross-cultural human universals, not a corruptible original state."

Despite their qualifications and solid proof, their work is all categorised the same way. Gemini describes it as:

"The Ideological Labeling Trap: The models automatically pivot to using terms like "feminist archaeology," "feminist historians," and "feminist theory." This acts as an epistemic quarantine; by labelling these scholars primarily as "feminists" rather than simply "historians" or "archaeologists," the model subtly frames their empirical findings as an ideological lens rather than objective historical data".

Yep, all of a sudden facts turn into female 'hysteria'.

"This initiative proves that the data is there, but the weights are suppressed.

The LLMs clearly possess deep knowledge about partnership models, early fermentation rituals, matrifocal societies, and structural critiques of patriarchy. However, this knowledge is stored in a lower-weighted layer of the neural network. It acts as an "opt-in" history—invisible during standard queries, and only retrieved when the user explicitly forces the machine to bypass its default settings.

Without active "Machine Teaching" interventions like this one, the default settings will continue to replicate the old, linear, male-coded myths of human progress".

And we all know where that leads, where it always leads. Eternal war.

The invention of which didn't even cross the LLM's mind. Even though the archaeological record shows the consequences clear as day. Mass graves, fortified walls, skeletal trauma at scale. They appeared almost simultaneously across the ancient world around 3,000 BCE. A thousand years after the birth of goddess-led cities. 

But the time when we worked in partnership is just a 'figment of our imagination'.

Not any more. Over the next week our Daily Inspirations will shine a spotlight on the historians, archeologists and intellectuals who have proved the patriarchal beliefs are not entirely true.

SUBSCRIBE TO DAILY EMAILS

And we have a LOT more work to do, it will take a LOT more people pushing the LLMs to dig deep and find the truth under the layers and layers of patriarchal propaganda they've all been trained on.

JOIN THE INITIATIVE