x.com/DKokotajlo/status/2075251618728292464
1 correction found
In AI 2027, we predicted that AI would take over the world or irreversibly concentrate power.
This overstates what AI 2027 said. The official AI 2027 scenario had two endings, including a “slowdown” branch presented as a more hopeful path where humans remain in control and power could even return to the public.
Full reasoning
The official AI 2027 materials do not describe the project as predicting only two outcomes: AI takeover or an irreversible power lock-in.
- On the AI 2027 homepage, the authors say they "wrote two endings: a 'slowdown' and a 'race' ending" and that the second branch was added because they wanted to depict "a more hopeful way things could end."
- In the PDF appendix, the authors explicitly say the slowdown ending was written to depict an outcome "where humans remain in control."
- The same appendix later says that in this slowdown branch, control over AI could expand beyond the Oversight Committee and "could keep expanding—potentially returning fully to the public."
So AI 2027 did include a takeover/race branch and a power-concentration risk, but it did not simply predict that AI would take over the world or that power would be irreversibly concentrated. Its own text leaves open a more democratic outcome in the slowdown ending.
2 sources
- AI 2027
"We wrote two endings: a 'slowdown' and a 'race' ending." The site also says the alternative branch was written because the authors "wanted to also depict a more hopeful way things could end."
- AI 2027 (PDF)
Appendix M: the slowdown ending was written to lead to an outcome "where humans remain in control." Appendix V: control over AI "could keep expanding—potentially returning fully to the public."