3 ms·
I can't remember who, but an AI expert tried challenging AI sceptics to come up with a future year and a feat that they were sure an AI wouldn't be able to achi
by dane-pgp 6y ago
I can't remember who, but an AI expert tried challenging AI sceptics to come up with a future year and a feat that they were sure an AI wouldn't be able to achieve by that year.
The sceptics were reluctant to give any such prediction, but he managed to persuade one to try.
Their prediction was that no AI would achieve a score above 90% on the Winograd Schema Challenge before 2025 (or something similar). The AI expert writing about this pointed out that a couple of months after the sceptic had made their prediction, in 2019, this threshold was passed by the BERT language model.
- rytill 6y agoEliezer Yudkowsky: https://intelligence.org/2017/10/13/fire-alarm/ https://intelligence.org/2017/10/13/fire-alarm/
- caturopath 6y agoLord, I hate the Rationalists' tendency to write long articles with so little structure. I don't understand how people make it through a post like this, which has a ton of juicy parts but doesn't stick to a thesis and doesn't commit to clarity.
- yters 6y agoWhat if all the quantifiable tests we have are AI achievable, but the tests only cover an insignificant portion of human intellectual ability?
- caturopath 6y agoWhat's an AI sceptic? It's apparently typified by some unnamed person who thought that incremental progress in natural language work would not be made? Was this person's guess worth while in any way? Were they an eminent researcher in an adjacent field or something?