Early Access Video: The OTHER AI Alignment Problem: Mesa-Optimizers and Inner Alignment
Added 2021-02-13 20:03:36 +0000 UTCThis "Alignment" thing turns out to be even harder than we thought.
https://youtu.be/bJLcIBixGj8
Also UPCOMING EVENT TODAY, we will have an experimental Live Watch Party on the Discord in one hour (at 9pm UK time, 1pm pacific, 4pm eastern). We'll watch the new video at the same time here:
https://sync-tube.de/rooms/GvwA2fBVp
and chat about it on the Discord here:
https://discord.gg/2YxbNS5KUK
Comments
one of the best videos yet, but also the heaviest weight I think there is no escape of nature and evolution and conflict with our creations. Your examples inspired from nature were the most convincing of them all. Great parallels. We are after all building artificial beings. Things will go really south once we enable replication inside these things, and once they are memetically enabled (to understand language and reason). I honestly think the end goal of your research will be what the military existence have figured out since civilization: Fight against the enemy (machine). I don't foresee any scenarios in which us the humans will be able to overpower and control intelligent machines when they are stronger/smarter than us. That is fine btw. As long as we play along we will be able to actually play the main objective of life: Construct dyson spheres, expand consciousness and figure out what is going on.
loopuleasa
2021-02-15 09:56:07 +0000 UTCCouldn't help thinking of Miller in The Expanse. Pretty good example in sci fi.
Ben Archer
2021-02-14 21:26:00 +0000 UTC
