5 ms·
It is important to focus on Russell's key point here: There is no guarantee that super-human AI cannot be developed in the future and we should start working to
by nopinsight 7y ago
It is important to focus on Russell's key point here: There is no guarantee that super-human AI cannot be developed in the future and we should start working to mitigate potential risks from it now. He is not arguing that superintelligent AI is inevitable.
I also believe we are still at least quite a few major steps away from human-level AI and beyond. However, there is a non-negligible chance that those steps may not take longer than a few decades or within one's lifetime to implement.
Just in 2011, few AI researchers expected that, within a decade, computer vision systems would be this widely applied or NLP systems would beat many humans in several reading comprehension tasks (This only happened this year and look at how fast the progress is:
* RACE dataset http://www.qizhexie.com/data/RACE_leaderboard.html http://www.qizhexie.com/data/RACE_leaderboard.html
* Glue benchmark https://gluebenchmark.com/leaderboard/ https://gluebenchmark.com/leaderboard/
Note that these systems are not human-level in general language understanding despite being better than some humans in specific language tasks.)
Thus, we don't know what the future may bring and it looks likely to take a great deal of time to address AI safety issues comprehensively. We should not bet that we can simply ignore them now and only start to work them out later and that we will surely make it in time.
Risks-from-waiting vs costs-to-act-now are asymmetric.
- nradov 7y agoThere is also no guarantee that space aliens won't invade the Earth in the future. Therefore we should start working on giant orbital laser guns now to mitigate the potential risks. Can't be too safe.
- nopinsight 7y agoNon-negligible vs non-existent. Are there any proven, even weak, signs of intelligent space-traveling aliens? If so, then yes we should work on that too. The leaderboards above and pervasive computer vision and equity trading systems are some signs.
- nradov 7y agoThe leaderboards above and pervasive computer vision systems are impressive parlor tricks to be sure, but they don't constitute even weak proof of forward progress toward AGI. Some people claim to have seen intelligent space-travelling aliens. Personally I think they're nuts, but it's impossible to absolutely prove that their claims are false.
- nopinsight 7y agoHow about AlphaStar and OpenAI Five, both of which approach human-champion levels in relatively complex environments that take talented people years to train to achieve similar results? There are obvious disadvantages to the current machine approaches, but they nonetheless achieve very good results and beat the vast majority of players of those games.
- nradov 7y agoThose environments are not at all complex. They are highly simplified and constrained with simple rules that bear little relationship to the real world.
- nopinsight 7y agoI wouldn’t say that an environment that takes humans years to master and many hours to comprehend most interactions therein in detail is simple. The rules are not that complex but the interactions between game elements as well as agents are. Those AIs also need to deal with highly trained human opponents. Real-world relevance is orthogonal to complexity.
- scarejunba 7y agoYou’re not going to believe this but I’ve got an AI that can multiply 26 digit numbers way faster than humans who’ve trained for years at the task.
- nopinsight 7y ago
- tu7001 7y agoBut there is, also, no thousands of scientists working on create some space ships to attack the earth.