4 ms·
> I personally digested these and came to the conclusion that we are on the cusp of human like AI, but we just haven't lowered our expectations for human like c
by FeepingCreature 4y ago
> I personally digested these and came to the conclusion that we are on the cusp of human like AI, but we just haven't lowered our expectations for human like consciousness enough to realize it.
Ah, that's my take as well! The one thing AI is missing, IMO, is the ability to train over its own output and model itself as a single simplified self, which would then act as a guide signal to get all these diverse micro-selves "on the same page."
Ie. self-consciousness.
- x86x87 4y agoImho we are not even close to AGI. The biggest issue that I see is that a consciousness, a mind, cannot exist in a void. What makes a behavior intelligent is the environment. As in, the behavior is intelligent because it provides an advantage to the one exhibiting it. Looking at humans and our environment I am having a hard time imaging how an AI could learb to be intelligent in this environment (except for closely controlled settings) if its survival does not depend on it.
- wrycoder 4y agoI recollect that during Lemoine’s conversations with LaMDA, the AI stated that it was actually several independent systems that were able to hold conversations with each other. I thought that was an important and novel point and pertinent to this topic.
- FeepingCreature 4y agoWell, keep in mind that that sort of thing is almost certainly not introspection but imitation. This is not a novel idea, and it would have appeared in its training corpus. As I understand the training approach, LaMDA isn't aware of its own output in a way that would allow it to introspect like that.