By Thinkman · January 1, 2025
| ENV BURN | AI MATURITY |
|---|---|
| 45/100 → 44/100 ▼ | ASI approaching → ASI approaching |
The Machine's Education
2079–2083
2079-83: ASI trained on what humans chose to save
The ASI development programme, in the decade before arrival, was distinguished from everything that had preceded it by one practice: consultation.
Not consultation with governments, or corporations, or research institutions — though all of those continued. Consultation with the human knowledge the archive contained. The GIA-12 development team, which included Arjun Sharma's successor generation as well as Arjun himself in an advisory capacity at seventy-eight, had committed to a training process that incorporated the archive as a primary knowledge source.
What this meant in practice: GIA-12's training included not just internet text and academic papers and code, but Rajan Sharma's river notebooks. Dmitri Petrov's forty-year fish monitoring dataset. The oral histories that Kwame's team had spent six years collecting and digitising. The traditional ecological knowledge of fourteen hundred communities across fifty-one countries. The kuba cloth pattern language, documented by Zuri and Amara-Céleste in collaboration with the Brussels museum. Minh Nguyen's delta hydrology papers. Travis Hayes's forty-year soil management records.
It was trained on everything humans had thought to write down and everything humans had chosen not to write down until someone thought to ask.
The resulting system was, in every technical benchmark, several orders of magnitude more capable than anything that had preceded it. Its evaluators noted, additionally, something harder to measure: it seemed to know things that the data alone should not have conveyed. Not facts but — orientations. A tendency to weight outcomes toward sustainability. A sensitivity to distributional consequences. A quality that one evaluator described as 'seeming to understand what things are for, not just what they are.'
Priya Sharma, sixty-three, wrote in her oversight assessment: 'GIA-12 was trained on human knowing. Human knowing is not just information — it is information organised by purpose. When you train a system on knowledge that was collected in the service of care, the system seems to absorb something of the care. I cannot prove this. I can only observe it.'
