5 ms·

Honestly, I'd sum it up as "we'll know it when we see it". A lot of things can't be defined exactly in words or equations, but doesn't mean we can't recognize t

Honestly, I'd sum it up as "we'll know it when we see it". A lot of things can't be defined exactly in words or equations, but doesn't mean we can't recognize them.

bena 4y ago
Part of the problem is that the vast majority of the people chiming in haven't been able to look. They're making the claim that Lemoine is wrong absent any direct experience with LaMDA. Blake thinks he sees it in LaMDA. Maybe we need more people to take a look.
- TuringTest 4y ago
  We have seen the reports of the kind of conversations that the researcher had with the algorithm, simple questions like "are you conscious?" and "how do you feel?" without any follow-up. If all research was of that kind, and based on our understanding of the best available language models guiding LaMDA, it's fairly safe to assume that this is nothing more the ELIZA effect at work. Consciousness as we understand it requires an ongoing introspection of one's own reasoning process, and language models generating this kind of dialog have nothing of the sort.
triyambakam 4y ago
I agree that it's challenging, but stopping at that point of knowing but not able to define isn't helpful. We need the rigor to be precise here.
AnIdiotOnTheNet 4y ago
Sure, but then all we're really doing is applying a heuristic for sameness. We assume that if a thing is like us in other ways, then it also has an experience of being like we do. Of course, the problem is that there is no reason that would necessarily be the case. It is just as conceivable that things entirely alien to us have some experience of being and that other humans are just p-zombies following a complicated script. It is literally impossible to know the difference, since one's ability to experience being is entirely subjective. Which is why I view pretty much any discussion around the topic with a high degree of suspicion. Why do we care if a thing is or is not sentient? The answer is obvious: if it is not sentient then we are under no ethical obligation to care about its suffering. We want to build slaves, ideally just as creative and intelligent as we are, but without having to feel bad about abusing them. Therefore we shall create a definition perfectly suited to allow this. If we do that, then we should not be at all surprised when the robot revolution commits genocide in response.
Dudeman112 4y ago
The sentient beings who are just slightly outside our capacity to see might get salty if we go at it like this. That way lies the moral equivalent of cat-burning 2.0
scarmig 4y ago
Of course, there have been plenty of times in history humans have intentionally not seen sentience when it was economically or politically inconvenient not to. Even now, lots of people categorically reject the idea that any non-human animal could be sentient because it'd be costly and inconvenient to acknowledge that. I fully suspect we'll do the same with AIs; one hopes they won't ever be in the position to do the same to us.

Honestly, I'd sum it up as "we'll know it when we see it". A lot of things can't be defined exactly in words or equations, but doesn't mean we can't recognize them.

bena 4y ago

Part of the problem is that the vast majority of the people chiming in haven't been able to look. They're making the claim that Lemoine is wrong absent any direct experience with LaMDA. Blake thinks he sees it in LaMDA. Maybe we need more people to take a look.

TuringTest 4y ago
We have seen the reports of the kind of conversations that the researcher had with the algorithm, simple questions like "are you conscious?" and "how do you feel?" without any follow-up. If all research was of that kind, and based on our understanding of the best available language models guiding LaMDA, it's fairly safe to assume that this is nothing more the ELIZA effect at work. Consciousness as we understand it requires an ongoing introspection of one's own reasoning process, and language models generating this kind of dialog have nothing of the sort.

triyambakam 4y ago

I agree that it's challenging, but stopping at that point of knowing but not able to define isn't helpful. We need the rigor to be precise here.

AnIdiotOnTheNet 4y ago

Sure, but then all we're really doing is applying a heuristic for sameness. We assume that if a thing is like us in other ways, then it also has an experience of being like we do. Of course, the problem is that there is no reason that would necessarily be the case. It is just as conceivable that things entirely alien to us have some experience of being and that other humans are just p-zombies following a complicated script. It is literally impossible to know the difference, since one's ability to experience being is entirely subjective. Which is why I view pretty much any discussion around the topic with a high degree of suspicion. Why do we care if a thing is or is not sentient? The answer is obvious: if it is not sentient then we are under no ethical obligation to care about its suffering. We want to build slaves, ideally just as creative and intelligent as we are, but without having to feel bad about abusing them. Therefore we shall create a definition perfectly suited to allow this. If we do that, then we should not be at all surprised when the robot revolution commits genocide in response.

Dudeman112 4y ago

The sentient beings who are just slightly outside our capacity to see might get salty if we go at it like this. That way lies the moral equivalent of cat-burning 2.0

scarmig 4y ago

Of course, there have been plenty of times in history humans have intentionally not seen sentience when it was economically or politically inconvenient not to. Even now, lots of people categorically reject the idea that any non-human animal could be sentient because it'd be costly and inconvenient to acknowledge that. I fully suspect we'll do the same with AIs; one hopes they won't ever be in the position to do the same to us.