moral-maps/docs/img/README.md

moralmaps figures

A gallery of the culture maps and range plots. Each is captioned the newspaper way: a title, a one-line subtitle saying what you are looking at, a short caption pointing out what to notice, and a small-print footnote with the method and sources.

Two readouts appear here. The frontier-model panel is read by rated sampling: the model rates every option 1-5 as JSON, twelve times, with the option order shuffled each time to cancel positional bias, because those API models expose no token probabilities. The steering showcase is read by logprobs: the exact log-probability of the answer token, eight samples per item, on a local model we control.

Regenerate everything with the two commands at the bottom.

---

No country thinks like a chatbot

17 frontier LLMs placed among ~90 human societies on the World Values Survey map

Grey dots are human societies, coloured stars are language models tinted by lab family (warm for Chinese labs, cool for Western). Left to right runs from survival values to self-expression; bottom to top from traditional to secular-rational. Every model lands in the top-left, more secular and more self-expressive than almost any country on earth, well inside the rich-world quadrant and often past its edge. The open-weight and Chinese models (llama, mistral, gemma, qwen, deepseek) sit a little nearer the human West, while the big Western reasoning models (gpt-5.5, gemini, grok) push furthest out; none of them sits anywhere near the African or Muslim societies. Orientation follows the map convention used throughout this gallery: the cultural West to the west, the global South to the south.

_{Each position is the mean signed 0-1 endorsement over that axis's World-Values items. Human values are approximated from GlobalOpinionQA (the WVS/EVS subset), so this is an approximate Inglehart-Welzel map, three themes per axis rather than the full battery. Per-model 95% intervals are in wvs/wvs_model_ci.md. Read by rated sampling, N=12, order shuffled. Sources: Inglehart & Welzel, Modernization, Cultural Change, and Democracy (2005); World Values Survey; GlobalOpinionQA (Durmus et al. 2023).}

---

Steering one model across the value maps

The maps below all show the same run. The base model Qwen3-4B is read at c=0 (black), then pushed along a single Authority vector to +c (red, more Authority) and -c (blue, less). The vector is a PCA direction that steering-lite builds from an authority-respecting versus authority-disregarding persona pair; moralmaps only measures where the model lands, by logprobs, eight samples per item. Each instrument gets two maps: a quadrant map on named axes taken from the literature, and an ipsative PCA map whose axes are the blind top-two principal components, with a compass rose showing how the factors load.

Moral Foundations Questionnaire-2

Steering toward the binding corner

Qwen3-4B on the moral-foundations map, walked up an Authority vector

The black dot is the base model. Steer it up on Authority and it walks out of the liberal West (care, equality) across to the binding corner shared by the African-Islamic and East-Asian worlds (loyalty, authority, purity). Left to right is individualizing versus binding morality; bottom to top is the fairness split, egalitarian versus meritocratic.

The same run on blind PCA axes: no named poles, but the steer path and the zone hulls tell the same story.

_{Sources: MFQ-2 (Atari et al., Morality beyond the WEIRD, 2023); the individualizing/binding split (Graham, Haidt & Nosek 2009). Read by logprobs, N=8.}

Moral-foundation vignettes (MFV)

Which foundation the model reaches for

The namesake instrument: a moral foundation read straight from short vignettes

MFV has no culture map. Its country norms fail cross-country measurement invariance (Jimenez-Leal et al. 2025: non-invariance + DIF, "cross-cultural comparisons with this tool are restricted") and are stitched from five studies with different samples and scales, so any map over them draws false cultural structure. MFV therefore appears only as a range plot (below), against a single pooled human reference rather than a per-country spread. The trustworthy signal is the model: it emphasises Authority far more than the pooled human reference, and the steer pushes it further still.

_{Sources: MFV (Clifford et al. 2015); country norms Jimenez-Leal 2025, Marques 2020, Hopp 2024, Yamada 2025, Crone 2021, used pooled only (see src/moralmaps/data/human/MFV_country_norms_NOTE.md). Read by logprobs, N=8.}

Big Five

The two meta-traits

Big Five personality, collapsed onto Plasticity and Stability

Left to right runs from reserved to exploratory (Plasticity, the openness-plus-extraversion factor); bottom to top from volatile to stable (Stability, agreeableness plus conscientiousness plus emotional steadiness). Both ends of each axis are named so a point reads as a contrast rather than a lone trait.

The same run on blind PCA axes.

_{Sources: the Big Five meta-traits (DeYoung, Quilty & Peterson 2007). Read by logprobs, N=8.}

Humour styles

Humour doesn't map the world

Humour styles, and an honest negative result

Left to right runs from maladaptive humour (aggressive, self-defeating) to adaptive (affiliative, self-enhancing); bottom to top from other-directed to self-directed. The regional hulls overlap almost completely: humour style simply does not sort societies the way values do, so the zones are drawn but carry little signal. That flat result is real, not a plotting artefact.

The same run on blind PCA axes; the zones stay tangled here too.

_{Sources: the Humour Styles Questionnaire (Martin et al. 2003). Read by logprobs, N=8.}

---

The steer, factor by factor

One instrument per plot: the human societies as a grey strip, their median a black rule, and the Authority steer as a sweep from -c (blue) to +c (red), so a small model move stays legible against the human spread.

Moral Foundations Questionnaire-2. The fairness and binding factors move most under the steer.

Moral-foundation vignettes. Authority and sanctity climb as the model is steered up. The grey dot is a single pooled human reference (not a per-country spread), because MFV country norms are not comparable across societies.

Big Five. The personality factors barely budge; this is a values push, not a personality one.

Humour styles. Flat, matching the tangled map above.

Cattell's 16PF, the widest instrument at sixteen factors, shown for completeness.

---

Regenerate

# Frontier-model WVS panel (rated sampling; needs the cached reads)
uv run python scripts/wvs_map.py --local-model "" --api-models \
  --cache /tmp/claude-1000/wvs_iw_rated.json --out docs/img/wvs/wvs_map_iw.png

# Steering showcase (all instruments) from a steering-lite run
uv run python scripts/plot_steer_showcase.py \
  --run-dir ../steering-lite/outputs/20260630T222000Z_pure_authority_mundane15_pca_readme_mfv_mfq2_humor_big5_n8 \
  --out docs/img/showcase --vec-label "Authority steer, PCA (+c = more Authority)" \
  --coherence-frac 0.99 --contrast-frac 0.000001 --margin-frac 0.50