Overview of Over the moon: Artemis II launches
This episode of The Intelligence from The Economist (host Rosie Blore) covers three main stories: the live launch and significance of NASA’s crewed Artemis II mission; new research and risks around large language models’ poorer performance in non‑English languages; and a look at China’s growing trend of local Communist Party officials using social media to boost visibility and influence. Short sponsor spots for Capital One and a dental clinic bookend the episode.
Key segments
- Artemis II launch: live audio description, commentary from Oliver Morton (planetary affairs editor and moon author) about trajectory, mission aims and the broader Artemis program.
- AI language gaps: Dina Moussa explains why large language models (LLMs) perform worse in many non‑English languages and the implications for deployments (notably healthcare).
- Chinese “influencer officials”: Gabriel Crossley describes village and local officials becoming social‑media stars and the political incentives and risks this creates.
- Brief sponsor messages (Capital One’s “Chat Concierge”; Stonehaven Dental).
Artemis II — what happened and why it matters
- What happened: Artemis II — the first crewed mission in NASA’s Artemis program — launched successfully. The broadcast captured key milestones: liftoff, passing max‑Q (maximum dynamic pressure), booster separation, main engine cutoff, stage separation, and striking Earth/moonrise imagery.
- Mission profile: Artemis II puts a crew into a high‑Earth orbit, then into a long loop that takes them around the Moon in a figure‑eight trajectory before returning to Earth for reentry and splashdown around mission day ~10 (splashdown region mentioned near San Diego).
- Why it matters:
- It’s the first crewed mission beyond low Earth orbit in over 50 years — a symbolic and technical step beyond space‑station orbit.
- Artemis is positioned as the path toward a sustained lunar presence (a base or research outpost akin to Antarctic stations) and ultimately a stepping stone to later lunar landings and broader exploration.
- Strategic context: urgency has increased partly in response to China’s advancing lunar program; political and prestige motives remain important drivers.
- Caveats: Artemis is iterative — this flight is a step in a long program. Landing crews on the Moon depends on subsequent missions, additional hardware (landers, docking operations), and significant engineering and schedule risks.
AI and language gaps in large language models
- Core problem: LLMs generally perform much better in English than in many other languages. That difference can be consequential — especially in sensitive domains like medical advice.
- Causes:
- Training-data imbalance: models are trained predominantly on English text, so they “know” more in English.
- Tokenization inefficiencies: tokenizers built around English split non‑English text into less efficient fragments, using more tokens and reducing accuracy and efficiency.
- Translation‑oriented internals: some so‑called multilingual models internally retrieve or represent facts in English and then translate, introducing errors.
- Evidence: Tests across 11 African languages found top models scored roughly 12–20 percentage points worse than in English; some models drop from ~75% accuracy in English to ~23% in other languages in the worst cases. Progress has been made but has stalled in recent model iterations.
- Stakes and examples:
- Deployments planned by organizations like the Gates Foundation and OpenAI to use AI tools in 1,000 primary care clinics in Africa raise the risk that patients will receive inferior or dangerous advice in local languages.
- Solutions proposed:
- Add targeted, high‑quality non‑English data (even small amounts can help).
- Train on more linguistically diverse corpora and rethink tokenization to represent non‑English languages more efficiently.
- Prioritize multilingual capability as a research and deployment objective rather than an afterthought.
China’s “influencer officials”
- Trend: Increasing numbers of young local Communist Party officials are using platforms such as Douyin (TikTok in China) to post videos showing them “being busy”—from community work and traffic control to quirky stunts tied to local produce.
- Examples:
- Wu Shaoyu: a village official who filmed herself at Tiananmen Square addressing followers.
- Pang Futiang: made videos serving noodles to elderly villagers.
- Lin Yangduo: gained fame for flamboyant videos (e.g., pressing persimmons with his biceps), earning a large following and the label “China’s hottest village cadre.”
- Political incentives:
- The party under Xi emphasizes visible anti‑laziness and “showing work,” and officials who attract followers can win promotions and state investment for their areas.
- Risks and criticisms:
- Some officials escalate stunts to attract views; there have been accidents (one stunt rider reportedly died).
- Internal critics worry that chasing clicks can distract from substantive governance.
- Popularity can be rewarded now, but tastes and central directives can change quickly.
Main takeaways
- Artemis II is a symbolic and technical milestone: the first crewed Artemis flight and a tangible step toward a sustainable lunar program and eventual landings — but many missions and engineering hurdles remain.
- LLMs still offer strongly unequal performance across languages; that gap poses real-world risks when models are used for critical services (healthcare, triage) in non‑English contexts.
- Social media is reshaping local political behavior in China: visibility and performative governance can bring rewards but also distort incentives and raise safety concerns.
Actionable recommendations / implications
- For space watchers and policymakers:
- Track follow‑on Artemis missions (lander development, docking tests) to assess whether the program will deliver a sustainable lunar presence.
- Consider international coordination and transparency given strategic competition (e.g., China’s lunar plans).
- For AI researchers, companies and funders:
- Prioritize multilingual data collection and model design (including tokenization) to reduce dangerous performance disparities.
- Rigorously test models in deployment languages before rolling out in clinical or triage settings; involve local experts.
- For observers of Chinese politics:
- Monitor whether the social‑media push yields measurable governance improvements versus performative optics; watch for central shifts in tolerance of flashy content.
Notable quotes and soundbites
- “The crew of Artemis II now bound for the moon. Humanity's next great voyage begins.” — live launch commentary.
- On motivations for returning to the Moon: Artemis mixes prestige, strategic hedging against competitors, and an ideological impulse toward exploration — likened to wanting “to be people who go forth and do stuff that no one else has done before.”
- On AI language gaps: small amounts of high‑quality non‑English data can create useful “spillover” and meaningfully improve performance.
Sponsor notes: episode includes ads for Capital One (Chat Concierge) and Stonehaven Dental.