By Katja Grace, 10 January 2015
If you want to know when human-level AI will be developed, a natural approach is to ask someone who works on developing AI. You might however be put off by such predictions being regularly criticized as inaccurate and biased. While they do seem overwhelmingly likely to be inaccurate and biased, I claim they would have to be very inaccurate and biased before they were worth ignoring, especially in the absence of many other sources of quality information. The bar for ridicule is well before the bar for being uninformative.
So on that note, we made a big summary of all of the surveys we know of on timelines to human-level AI. And also a bunch of summary pages on specific human-level AI surveys. We hope they are a useful reference, and also help avert selection bias selection bias from people only knowing about surveys that support their particular views on selection bias.
It’s interesting to note the consistency between the surveys that asked participants to place confidence intervals. They all predict there is a ten percent chance of human-level AI sometime in the 2020s, and almost all place a fifty percent chance of human-level AI between 2040 and 2050. They are even pretty consistent on the 90% date, with more than half in 2070-2080. This is probably mostly evidence that people talk to each other and hear about similar famous predictions. However it is some evidence of accuracy, since if each survey produced radically different estimates we must conclude that surveys are fairly inaccurate.
If you know of more surveys on human-level AI timelines, do send them our way.
Here’s a summary of our summary:
| Year | Survey | # | 10% | 50% | 90% | Other key ‘Predictions’ | Participants | Response rate | Link to original document |
| 1972 | Michie | 67 | Median 50y (2022) (vs 20 or >50) | AI, CS | – | link | |||
| 2005 | Bainbridge | 26 | Median 2085 | Tech | – | link | |||
| 2006 | AI@50 | median >50y (2056) | AI conf | – | link | ||||
| 2007 | Klein | 888 | median 2030-2050 | Futurism? | – | link and link | |||
| 2009 | AGI-09 | 2020 | 2040 | 2075 | AGI conf; AI | – | link | ||
| 2011 | FHI Winter Intelligence | 35 | 2028 | 2050 | 2150 | AGI impacts conf; 44% related technical | 41% | link | |
| 2011-2012 | Kruel interviews | 37 | 2025 | 2035 | 2070 | AGI, AI | – | link | |
| 2012 | FHI: AGI | 72 | 2022 | 2040 | 2065 | AGI & AGI impacts conf; AGI, technical work | 65% | link | |
| 2012 | FHI:PT-AI | 43 | 2023 | 2048 | 2080 | Philosophy & theory of AI conf; not technical AI | 49% | link | |
| 2012-present | Hanson | ~10 | ≤ 10% progress to human level in past 20y | AI | – | link | |||
| 2013 | FHI: TOP100 | 29 | 2022 | 2040 | 2075 | Top AI | 29% | link | |
| 2013 | FHI:EETN | 26 | 2020 | 2050 | 2093 | Greek assoc. for AI; AI | 10% | link |
(Image: AGI-09 participants, by jeriaska)
2 Trackbacks / Pingbacks