Comment by jdblair
3 years ago
Reading the abstract, the first thing I thought was this question:
Is it moral to bring an artificial general intelligence into existence and then hobble the intelligence’s capability? This sounds a lot like creating a new sentience and then dooming it be slave to humanity.
I expect there are less extreme interpretations, and lurking just under the surface are deep questions like “what is free will? do we have free will?”
We're scaling up what computers can do. They still have as much consciousness as a rock (none). Don't lose sight of what they are, just because their capabilities seem human-like.
Do you have proof that humans are conscious?
It's accepted that we are, by consensus, because of what we experience individually. We know that scientific proof is lacking, although perhaps scientists in the field may disagree, I don't know for sure.
6 replies →
Does your consciousness disappear without proof?
1 reply →
Oh sure, I don't think LLM are sentient. But this paper is about hypothetical "general intelligence," no?
Are we hobbled by the fact evolution made us into social creatures who value each other? If not, then I don't think it's wrong for us to make AI like that too.
Good point! On the other hand, we're not provably safe for social interaction. We have all manner of punishments for variance from expected behavior exactly because our altruistic tendencies are so unreliable.
It looks like Tegmark wants the same unbreakable rules for everyone, not just for AGIs: protect routers from being hacked, make it impossible for planes to fly into buildings, etc.
If it's moral to selectively breed animals to produce superior health outcomes and compatibility with our own species, it seems that the same should hold true for AI.
Just don't get PETA involved in the discussion
Most people don't think livestock are "general intelligences."
Those that do probably overlap with those you want to leave out of the discussion.
Are we creating machines with just enough reasoning to be useful, or are we building sentient general intelligences?
I'd need to see what the argument is for "Family dog, which can be trained, can communicate its needs, enjoys play, is capable of empathy, and feels emotion isn't generally intelligent/sentient"
IMO - We're building specialized devices on our way to building family pets, then humans, then perhaps something greater than humans.