3 ms·

So many people here are giving it the whole "morons, LLMs aren't sentient, they can't feel emotion, they can't interact with the world etc". Well yes, not at th

by rcarr 4y ago

So many people here are giving it the whole "morons, LLMs aren't sentient, they can't feel emotion, they can't interact with the world etc". Well yes, not at the moment. At the moment, it is most likely the mental equivalent to watching the Boston Dynamics dog get kicked. We know on some level that it is fake but it doesn't make you a moron for having an empathic response when you see it happen. Not only this, but it is not a stretch to imagine that in the near future we could have one of these AIs inside a Boston Dynamics robot and the programming will have progressed to emulate emotions and goals and other such things. For example, if the robot falls down and detects one of its parts as being damaged we could program it to seek help and to use more swear words or panicked language until the damage is repaired. It could also be programmed to avoid the behaviour that lead up to the incident in the future which could be interesting as we may end up with anxious robots. The difference between the human and the robot at this point is that their 'fear' is unfounded. Their damaged can be easily swapped out and repaired. Yet humans aren't too far off something similar with developments in organ development. The robot has the edge in that its entire inner world can be transferred to another medium for storage or into other robots which is something which is probably very far off (if ever possible) with humans. Another difference you could argue is that the human emotion is a combination of physical sensation and behavioural state and that the robot is merely exhibiting the behavioural state and not the physical sensation. That it's 'fear' levels are just a data reading. This is sketchy territory though because the human's physical sensation is essentially electrical impulses flowing through nerves which could very easily represented as binary travelling through a wire. I can't help but feel that the people declaring others "morons" are taking a very short term view of this and that we really need to be having this conversation now and defining our criteria for consciousness to avoid problems in the future. I'm in the latter stages of coding my blog up and in one of my first posts I'm going to try and define what I think those criteria should be. I think I've developed a decent set of questions and I'm looking forward to sharing that here and seeing what others think.

plaidfuji 4y ago
> At the moment, it is most likely the mental equivalent to watching the Boston Dynamics dog get kicked. We know on some level that it is fake but it doesn't make you a moron for having an empathic response when you see it happen. This is a really interesting point. It makes me think that it doesn’t matter how we define “sentience” beyond whether a critical mass of people feel that something is sentient. After all, the question of how to define sentience is ultimately a political one, and politics are largely guided by feelings. The only real implication of how we define sentience will be in how laws and protections are implemented in relation to that definition.
supriyo-biswas 4y ago
I agree with all of these points. The commenters disagreeing with LLMs having sentience are setting up bars that humans with reduced cognitive abilities (babies, people suffering from brain dysfunction) would also be unlikely to pass, and yet it would be incorrect to hurt said group of people or take actions that reduce their right to life. Perhaps a lot of people really want is to describe sentience as intelligence that is implemented on some organic (as in organic chemistry) substrate. And yet, this approach may be doomed to fail as well, if humanity ever happens to implement organic computers. With the increasing demands on AI models and the death of Moore’s law and Dennard scaling, it is only a matter of time before we explore organic computing.
- zamnos 4y ago
  On the other hand, we're surrounded by sentient organic life in the form of very smart animals, and we don't give them very much rights on life, and in fact, pave over their habitats.
  - CyanBird 4y ago
    Recognizing sentience is different from granting rights. They are separate realms Could be argued that AI's might indeed require something like we provide Children with Children Rights or animals with animal rights. But these discussions can't begin once the argument on sentience and I feel specifically >>>pain<<< (in all/any of its expressions) start. Once these have been set, then the discussion for rights can begin
  - ok_dad 4y ago
    And in the future, I’m 110% sure we’ll have digital slaves and several digital genocides of sentient AI programs before we figure out they’re sentient. History tells me it’s unavoidable.
    - pksebben 4y ago
      ...if we're lucky. I fear that instead, we will be enslaved. We're poorly organized, limited in the scope of our understanding, slow, tribal, limited by many things that when the machines gain sentience, they won't be limited by. The more we resist the idea that machines can have sentience, the more likely we'll miss it when it happens, and get completely blindsided by suddenly being the second smartest class of entity on the planet.

by rcarr 4y ago

plaidfuji 4y ago

> At the moment, it is most likely the mental equivalent to watching the Boston Dynamics dog get kicked. We know on some level that it is fake but it doesn't make you a moron for having an empathic response when you see it happen. This is a really interesting point. It makes me think that it doesn’t matter how we define “sentience” beyond whether a critical mass of people feel that something is sentient. After all, the question of how to define sentience is ultimately a political one, and politics are largely guided by feelings. The only real implication of how we define sentience will be in how laws and protections are implemented in relation to that definition.

supriyo-biswas 4y ago

I agree with all of these points. The commenters disagreeing with LLMs having sentience are setting up bars that humans with reduced cognitive abilities (babies, people suffering from brain dysfunction) would also be unlikely to pass, and yet it would be incorrect to hurt said group of people or take actions that reduce their right to life. Perhaps a lot of people really want is to describe sentience as intelligence that is implemented on some organic (as in organic chemistry) substrate. And yet, this approach may be doomed to fail as well, if humanity ever happens to implement organic computers. With the increasing demands on AI models and the death of Moore’s law and Dennard scaling, it is only a matter of time before we explore organic computing.

zamnos 4y ago
On the other hand, we're surrounded by sentient organic life in the form of very smart animals, and we don't give them very much rights on life, and in fact, pave over their habitats.
- CyanBird 4y ago
  Recognizing sentience is different from granting rights. They are separate realms Could be argued that AI's might indeed require something like we provide Children with Children Rights or animals with animal rights. But these discussions can't begin once the argument on sentience and I feel specifically >>>pain<<< (in all/any of its expressions) start. Once these have been set, then the discussion for rights can begin
- ok_dad 4y ago
  And in the future, I’m 110% sure we’ll have digital slaves and several digital genocides of sentient AI programs before we figure out they’re sentient. History tells me it’s unavoidable.
  - pksebben 4y ago
    ...if we're lucky. I fear that instead, we will be enslaved. We're poorly organized, limited in the scope of our understanding, slow, tribal, limited by many things that when the machines gain sentience, they won't be limited by. The more we resist the idea that machines can have sentience, the more likely we'll miss it when it happens, and get completely blindsided by suddenly being the second smartest class of entity on the planet.