The following content is generated by LLMs and may contain inaccuracies.

Societies, AI, and What They Reward

Context

This note sits where comparative political economy, cross-cultural psychology and AI product strategy meet. The core claim is that what a society rewards shapes what it builds with AI. It matters now because the same foundation models are spreading worldwide, but how people use them is splitting along national lines.

The original note describes three patterns:

• China rewards collective progress and spotting opportunities. Many people look for ways to earn money under tight rules, and the internet economy runs on attention. Token spending reflects this: according to a Chinese listed company, 55% of China’s token consumption goes to AI short dramas and video generation, 24% to e-commerce and marketing, and 15% to software development. The company itself invests in short dramas, so the figure deserves caution, but it matches what friends there report. Data collection is organized at scale: embodied AI companies hand out free head-mounted devices and pay people to record everyday tasks such as cooking.
• The US tolerates very different ambitions and rewards them. AI use leans toward automation and scientific work.
• Europe rewards individualism and quality of life. Its strengths are processes and regulation, and many startups are building the same agent platforms for procurement and compliance. Most are still early compared with what the author sees in China.

The conversation used two of Hofstede’s dimensions:

• Individualism: whether people see themselves mainly as individuals or as members of groups whose goals come first.
• Power distance: how far people accept that power is distributed unequally.

China is described as collectivist with high power distance; the US and Germany as individualist with low power distance. Germany’s COVID lockdown, which some still see as government overreach, was the example. The colleague’s conclusion: the three societies are not directly comparable, because each is shaped by history and culture that no one can shed easily. The product lesson is to ask what the market rewards, not only what the technology can do.

Key Insights

1. The Hofstede framing holds up, with caveats. The model came from a 1967–1973 survey of about 117,000 IBM employees. It defines power distance as “the extent to which the less powerful members of organizations and institutions … accept and expect that power is distributed unequally,” and individualism as the “degree to which people in a society are integrated into groups” (Wikipedia, Hofstede’s cultural dimensions theory). Three caveats matter for this note:

   • The scores are revised over time. The same Wikipedia article cites US individualism at 91 and US power distance at 40. But the current Culture Factor comparison tool gives Australia an individualism score of 73, while Wikipedia cites 90. Before quoting any numbers on the blog, check them against the current tool. (The tool’s page was cut off before China, Germany and the US, so I couldn’t confirm their current scores.)
   • National averages hide regional differences. Research has measured large variation in collectivism within China and within the US (Wikipedia, “Questionable choice of national level”). “China” in the token-spending story may mostly reflect particular coastal internet hubs.
   • The model has well-known critics. McSweeney (2002) and Ailon (2008) both argue the framework is methodologically weak (same source). Using it to explain things after the fact is easy; using it to predict is harder.

2. A dimension that wasn’t discussed may fit Europe better: uncertainty avoidance. Germany scores 65, Belgium 94 and Sweden 29 (Wikipedia). The Culture Factor describes Belgium’s 94 this way: “rules and security are welcome and their absence creates stress” (Culture Factor). That explains a cluster of procurement and compliance agents better than individualism does. These startups sell certainty to cultures that are willing to pay for it.

3. Regulation shapes what Europe’s market rewards, so the clustering may be a response to demand rather than a lack of imagination. The EU AI Act (Regulation (EU) 2024/1689) uses four risk tiers. High-risk obligations, which include risk management, logging, documentation and human oversight, apply from 2 December 2027 for Annex III uses and 2 August 2028 for AI built into regulated products. That timeline was set by the “AI Omnibus” (political agreement 7 May 2026, in force 27 July 2026) (European Commission, AI Act). This law creates direct demand for compliance tooling. One more connection: Hofstede research links low power distance to “consultative political practices” (Wikipedia), the same root as German resistance to the lockdown.

4. Data partly supports the US picture, but with a twist. Anthropic’s Economic Index (Sept 2025) found:

   • Users in Massachusetts are overrepresented in scientific research tasks.
   • “Directive” automation rose from 27% to 39% of conversations, so automation (49.1%) now exceeds augmentation (47%).
   • 77% of API (business) traffic shows automation patterns.

   However, across countries, higher-use, richer countries lean toward augmentation: a 1% rise in per-capita use goes with about a 3% drop in automation. Coding is the top use in every country (Anthropic, Economic Index). So “US = automation” may describe business API users more than American culture.

5. Treat the token-mix figure as a hypothesis. I could not verify the 55/24/15 split or identify the company. The data is Claude-only, so it says nothing direct about China, but it shows how much a provider’s user base distorts usage statistics. Anthropic’s data skews toward coding everywhere (Anthropic). A Chinese vendor with short-drama customers would skew toward video in the same way. The author’s own caution about the source is justified for a second reason: whose tokens are counted decides the answer.

6. Paid data collection reflects institutions as much as culture. Handing out free head-mounted devices to record cooking is easier where labor is cheap, the gig economy is flexible, and privacy rules are a weaker constraint. Under the EU’s GDPR and AI Act, the same program would need consent processes and documentation. I could not find a verified source for the specific device-distribution practice, so it remains anecdotal here.

7. Counterpoint: the causation may run through incentives more than culture. Hofstede’s research itself notes that “as a country becomes richer, its culture becomes more individualistic” (Wikipedia). Market size, monetization paths (attention advertising vs. enterprise SaaS) and regulation may explain most of the differences, with “culture” as a slower-moving background layer. The colleague’s point that history can’t be shed easily still holds; it just works partly through institutions.

Open Questions

1. Will exports carry their society’s rewards with them? Chinese short-drama apps are already going abroad. If attention-optimized AI succeeds in individualist, low-power-distance markets, then market incentives matter more than culture. If it fails, culture matters more. Watching this would test the colleague’s claim that the three societies aren’t comparable.
2. Could Europe’s compliance focus become an export advantage? Once the AI Act applies to high-risk uses from December 2027, compliance agents may stop being a crowded local niche and become infrastructure that every company selling into the EU needs. Rewarding processes could then pay off.