Talk:Maes–Garreau law

Perhaps this would be best merged with the Future_studies article? DaFranker (talk) 14:08, 15 August 2012 (UTC)

I think it works as a standalone page. Future studies would become a trashheap if we put every law or claim of this kind in there. Anders Sandberg (talk) 14:16, 15 August 2012 (UTC)

The paper from MIRI
The paper cited does not actually say that the Maes-Garreau law is not borne out in practice. It says this:

Fifty-five predictions were retained, in which it was possible to estimate the predictor’s expected lifespan. Then the difference between their median prediction and this lifespan was computed (a positive difference meaning they would expect to die before AI, a negative difference meaning they didn’t). A zero difference would be a perfect example of the Maes-Garreau law: the predictor expects AI to be developed at the exact end of their life. This number was then plotted again the predictor’s age in Figure 2 (the plot was restricted to those predictions within thirty years of the predictor’s expected lifetime). From this, it can be seen that the Maes-Garreau law is not borne out by the evidence: only twelve predictions (22% of the total) were within five years in either direction of the zero point.

It uses a different definition of the law than Wikipedia, namely that the Maes-Garrau law states that the predictor expects the singularity to happen precisely at the end of their lifetime, not just within their lifetime. — Preceding unsigned comment added by Oracions (talk • contribs) 11:37, 15 July 2015 (UTC)


 * Yeah. I've moved it down and qualified its use of "Maes-Garreau". It does count as an actual (if minor) academic publication on the subject, so may be worth noting - David Gerard (talk) 19:02, 15 July 2015 (UTC)


 * I've reformulated the paragraph on the MIRI paper, to make it clear that it was a data point against the law (or at least against a version of the law). Cheesefondue (talk) 12:21, 7 September 2017 (UTC)