---
title: "Reframing Superintelligence: Comprehensive AI Services as General Intelligence"
authors: "K. Eric Drexler"
date: 2019-01 (sections circulated as drafts from 2016-12)
kind: work (Future of Humanity Institute Technical Report 2019-1)
status: published; text unchanged
rights: CC0 1.0 (dedicated by the author)
related: ["../../commentary/2026-reframing-superintelligence-review/"]
---

# *Reframing Superintelligence* (2019)

K. Eric Drexler, *Reframing Superintelligence: Comprehensive AI Services as General Intelligence*, Future of Humanity Institute Technical Report 2019-1, University of Oxford, January 2019. Sections were written and circulated as drafts from December 2016. The report argues that general, superintelligent-level capability can take the form of a growing set of AI services, including the service of developing new services, and that this reframes both the prospects for superintelligence and the problems of AI safety.

## Files

| File | What it is | Use it for |
|---|---|---|
| [`reframing-superintelligence-fhi-tr-2019-1.pdf`](reframing-superintelligence-fhi-tr-2019-1.pdf) | The report as published (210 pages), the copy archived at Oxford's research archive: https://ora.ox.ac.uk/objects/uuid:9c05427a-6390-4b42-9c55-ee45f73a26ad | Authoritative text, figures and page numbers |
| [`extended-toc.md`](extended-toc.md) (and [`.pdf`](extended-toc.pdf)) | The extended table of contents: every section and subsection heading, with each section's thesis. The headings were written as propositions, so it reads as a summary of the whole argument. Prepared with a model; all 419 entries were checked verbatim against the edition | A short overview of the whole report |
| [`edition/`](edition/) | A verbatim Markdown edition converted from the author's LaTeX source: one file per section, plus [`index.md`](edition/index.md) (section theses), [`references.md`](edition/references.md), [`VERIFICATION.md`](edition/VERIFICATION.md) (conversion checks) and [`full.md`](edition/full.md). Figures are not reproduced; their model-written descriptions are marked editorial | Reading, quoting and retrieval |
| [`precursors/`](precursors/) | The author's slides for an FHI/DeepMind AI safety seminar (Oxford, 2 March 2017) and for a talk at OpenAI (21 June 2018), which date the R&D-automation model of recursive improvement, the "no self" argument and the conditions that disrupt collusion | Dating and development of the ideas |

**Duplication.** `edition/full.md` is the section files concatenated in reading order, for single-file use. Use either it or the section files, not both; a retrieval index built over both will return every passage twice. The PDF and the edition have the same text; the edition omits the figures.

## Commentary in this library

- [*Reframing Superintelligence* in 2026: a review](../../commentary/2026-reframing-superintelligence-review/), written by Claude models (Anthropic) in October 2026: what remains forward-looking, what failed, and what is still open.