From the Golden Record to Generative Curation: If AI Re‑selects Earth’s Sounds Today, Will Indigenous Languages Be Sent into Space?

In 1977, Voyager carried a gold‑plated copper record away from Earth. It was not a music album issued for the terrestrial market but a letter sent to an unknown cosmos: it contained natural sounds, music, encoded images and greetings in many languages, attempting to let any possible receiver know that somewhere in the Solar System there once existed a civilization capable of listening, singing, measuring and imagining.

Half a century later, the probes are aging, yet this record still acts like a mirror reflecting humanity’s most difficult question: who can represent Earth?

From the Golden Record to Generative Curation

If today we were to redo “Golden Record 2.0”, the engineering conditions would be vastly different. Storage capacity could be larger; audio and video could be preserved at higher resolution; artificial intelligence could organize billions of cultural data points in a short time. Models can analyse global music, language distribution, natural soundscapes and scientific knowledge, even simulate different curation schemes so that a spacecraft could carry a richer Earth archive than 1977.

The problem is that AI’s notion of “completeness” usually rests on accessible data rather than the world’s true diversity. Languages with abundant online material are treated as normal by models; languages lacking written, digitised or commercial‑platform exposure are easily judged rare, marginal and even unrepresentative. When algorithms select Earth’s voice according to frequency, they may simply amplify existing power: cultures with the most data speak again, while voices that have been colonised, suppressed or ignored remain silent.

Language Is Not a Greeting List

Indigenous languages cannot play merely as “diversity decoration” in the Golden Record. A language is not just translating “hello” into another set of syllables; it preserves place names, kinship relations, seasonal changes, plant uses, river directions, classifications of dreams and migration routes of ancestors. Some terms cannot be precisely replaced by mainstream languages because they describe not a single object but an entire way people live with the land.

If AI is asked to pick the language that best represents Earth, it should first acknowledge that low‑resource languages are not low‑value languages. Few digital records may result from historical bans on mother tongues, exclusion in education systems, lack of writing systems, or community choices based on cultural safety. Gaps in databases are not gaps in civilisation and cannot justify an algorithm deleting a worldview.

Recording a language into a space archive does not equal preserving it. Real revitalisation happens on Earth: can children use the mother tongue at home and school? Do language workers receive sustained resources? Can communities decide their own textbooks, orthographies and digital tools? If a language is gradually losing its space of use on Earth but only leaves a few seconds of greetings in a probe, that resembles civilisation making a commemorative object for its own failure. The budget and technical achievements of Golden Record 2.0 should instead support language communities rather than leaving only symbolic cosmic exposure.

This also turns Golden Record 2.0 from a technology project into a governance issue. Who nominates? Who has veto power? Is consent obtained from speakers and communities for recordings? After the sound leaves Earth, does it remain bound by existing cultural norms? If a song should appear only in certain seasons, ceremonies or family relations, sending it into space is not necessarily highest honour; it may be an irrevocable removal from context.

AI Must Learn to Respect “Not Included”

Generative AI excels at filling gaps, but cultural ethics sometimes demand precisely that we do not fill. Models might fabricate Indigenous language songs from scant public data, splice sacred vocabulary, or generate “representative voices” for communities without authorisation; technically smooth, culturally they may be wholly wrong. A truly responsible curation system must retain provenance, licensing levels, usage scopes and revocation mechanisms, and must allow communities to answer: “This segment is not public, cannot be trained on, and should not be sent away.”

Two-Eyed Seeing offers another path. Scientists and engineers can handle deep‑space communications, durable materials and data encoding; language workers, elders, youth and cultural practitioners decide what represents them and how it should be understood. AI can assist in identifying omissions, presenting selection bias and preserving different versions, but cannot replace community cultural judgement. Global representativeness should not be a model’s ranking list but a negotiation willing to acknowledge differences and boundaries.

The most moving aspect of the Golden Record has never been that an alien civilisation will actually play it; it is humanity’s willingness to organise itself before leaving Earth. If the next record involves AI, we need not only larger capacity but more mature memory ethics. The cosmos may never answer, but at least we can determine: when Earth speaks again, those long‑silenced people must not have others speak for them.

Sources and Further Reading

• NASA Jet Propulsion Laboratory: Voyager Golden Record official content, production background and record structure
• NASA Voyager Mission Status: Voyager 1 and 2 mission information and scientific instrument operation data
• UNESCO World Atlas of Languages: global language diversity and endangered language data
• CARE Principles for Indigenous Data Governance: collective benefit, control, responsibility and ethical principles