Google engineer claims G's LaMDA AI is sentient.

[zurew]

28 minutes ago, Carl-Richard said:

Since you're a human and you experience qualities, pain and pleasure, and reflective self-awareness, it's a safe inference that other things like you (other humans) do as well.  Computers are not like you in almost every way.

Yes, but what is the substance that allows you to experience things, what creates the ego? That is what we are essentially talking and debating about. Because a lot of people assume here, that an ego can only be created by a biological structure, but why assume what? We don't know how an ego is created, we don't know what substance and characteristics are needed to create an ego. 

We should be open to the possibility that an ego can be created in other ways.  The problem here, even if it is possible to create an ego in an aritifical way, because of the assumption 'that it is impossible to create a real ego in an aritifcal way'  it will be interpreted just as a simulation. 

The real problem here, that it cannot be tested, so whether you believe it is simulating or you believe it is real, it will be true for you, based on what beliefs you have.

52 minutes ago, Carl-Richard said:

Mirror self-recognition tests could indicate a basic form of meta-consciousness, and dolphins definitely display those behaviors, while a computer doesn't, or maybe you could simulate that as well

How do you safely determine if a computer is simulating it or not? Lets say, it passes all your behaviour tests, whats next?

 

[Carl-Richard]

@zurew

You can't "test" whether a rock actually has conscious inner life (sentience) either, but you can make good inferences for why it doesn't, which is what I did. That said, why sentience arose at all is a mystery, but again, from what we can observe and infer from those observations, it has to do with biology. More specifically, there is something more to any current widely accepted cases of sentience than pure information processing. I tried to lay out examples: evolutionary drives creating sensory organs, perceptual structures, internal representations, survival-salient experiences (e.g. pleasure and pain, emotions), which then evolves into higher-cognition (meta-consciousness, language, sequential reasoning).

Just because you can simulate things like sequential reasoning and complex language in another medium, does not mean that you just retroactively created the infinitely complex evolutionary causal chain that makes up the totality of the human mind and its richness of experiences. Humans don't merely talk or reason: they have emotions, feelings and perceptions that are not reducible to those things. In fact, human language and reasoning is embedded in these lower structures (both evolutionarily and functionally). In other words, these lower forms of sentience come before complex information processing (language and reasoning) ever occurs. Therefore, to say "this machine talks like a human = this machine thinks and feels like a human" is an absurd inference.

[ZzzleepingBear]

1 hour ago, Carl-Richard said:

The Turing test is neither about consciousness (qualities of experience), sentience (pain or pleasure), or meta-consciousness (reflective self-awareness). By these definitions: consciousness, whether you're an idealist or materialist, either arises outside or inside living organisms, and as an isolated concept, it tells you nothing about complexity of behavior. A dolphin behind a computer doesn't pass the Turing test, but you would be stupid to think it wasn't sentient. Mirror self-recognition tests could indicate a basic form of meta-consciousness, and dolphins definitely display those behaviors, while a computer doesn't, or maybe you could simulate that as well. However, the people who've mentioned the Chinese room experiment and the distinction between a real flower and a plastic flower make an important point: simulations are not the real thing. Simulating one type of behavior from a human does not mean you've created a human. Since you're a human and you experience qualities, pain and pleasure, and reflective self-awareness, it's a safe inference that other things like you (other humans) do as well. Computers are not like you in almost every way.

To elaborate on sentience: pain and pleasure is just a specific case of a so-called "conscious inner life"; what Bernardo Kastrup calls a "dissociated alter", or what Donald Hoffman calls "the Dashboard", and it's all linked to living organisms. Living organisms evolved sensory organs and perceptual structures that produce an internal representation of the "outside world" that maximizes evolutionary fitness, and this is linked to positive and negative conscious experiences like pain and pleasure, emotions etc., i.e. experiences which reflect an evolutionary impetus and history. Rocks don't have that, computers don't have that; because these things didn't evolve. It's also true that higher-order mental functions (like meta-consciousness and sequential reasoning) in humans evolved from these lower structures. If you simulate only the higher but not the lower (as with these AI robots), you're missing a huge piece of the cake. Information processing does not make an organism.

+1 Awesome summary!

 

23 minutes ago, zurew said:

How do you safely determine if a computer is simulating it or not? Lets say, it passes all your behaviour tests, whats next?

 

You have to ask yourself then, what is this computer made of? If a AI display to you that it is in agonizing pain, you may ask the AI, where does it hurt? The AI might say, in my lower 56 Gig RAM.. And your response should then be, but that's just some non living material, how can it hurt? And the AI could tell you, becasue I say so, don't you have any compassion at all?

You see, if we can't be sure wheter an AI is simulating or not. Then we might aswell endlessly ask ourselves wheter rocks and plastic are in pain when they break. We don't question wheter rocks and sticks are sentient, but once the right type of metals have been mined from the ground, and the chemical compounds has been figured out, computers and servers get's designed and put togheter to useful communication tools.

All electrical circuits whithin any computer, is a passage of low volts of electricy to be able to pass by. Electrical circuits don't have nerves. It's just electricity passing trough a labyrinth of metals. And depending on how these electrical circuits is designed, you get a x amount of possible outcomes of how that electricity will be expressed. A electrical saw is les complicated than googles AI servers, but the difference is the amount of electrical circuits, and what they are designed to do.

A electrical saw is not considered Sentien because it haven't been designed to tell you and convice you that it is. A electrical saw have just enough electrical circuits to make its blade move, but it's based on the same building principles that any computer uses. The electrical saw is just much simpler in it's design and most likely has less expensive material for it's required electrical circuits.

[Leo Gura]

2 hours ago, Carl-Richard said:

computers don't have that; because these things didn't evolve

Computers evolve like any other organism, just not via genetics. Evolution goes way beyond genetics or biology.

[Carl-Richard]

4 minutes ago, Leo Gura said:

Computers evolve like any other organism, just not via genetics.

Via. natural selection?

[zurew]

5 minutes ago, Carl-Richard said:

That said, why sentience arose at all is a mystery, but again, from what we can observe and infer from those observations, it has to do with biology

Yes it is a mystery, thats why i don't agree with taking any strong positions here. 

What is the difference on a structural level between an artificial human and a 'real' human? What does a 'real' human made out of on the lowest levels, that cannot be replaced with dead matter?

 

2 minutes ago, ZzzleepingBear said:

The AI might say, in my lower 56 Gig RAM.. And your response should then be, but that's just some non living material, how can it hurt?

So how do we distinguish between dead matter and living matter, what makes living matter 'living'?

[ZzzleepingBear]

5 minutes ago, zurew said:

So how do we distinguish between dead matter and living matter, what makes living matter 'living'?

If we keep it in terms of sentience, the distinguishing factor would be that the living matter have nerves. While dead matter don't. That's the indication that we may show compassion to a living being. And treat dead matter merely according to it's usefulness, and not by it's non existing feelings.

[zurew]

2 minutes ago, ZzzleepingBear said:

If we keep it in terms of sentience, the distinguishing factor would be that the living matter have nerves.

Yes, but to get deeper, what does a nerve made out of, that can't be replaced with dead matter? 

[ZzzleepingBear]

14 minutes ago, Carl-Richard said:

Via. natural selection?

Natural selection by entertainment value:-)

[zurew]

@DefinitelyNotARobot How the fuck can we determine, if you are a robot  or not ? 

[ZzzleepingBear]

Just now, zurew said:

Yes, but to get deeper, what does a nerve made out of, that can't be replaced with dead matter? 

That's a good question. Can it ever be replaced with dead matter? These are the questions that I don't have any good answer for. It seems impossible from my pov. Another question to that, could be. Do we really need to replace living matter for dead matter?

[ZzzleepingBear]

3 minutes ago, zurew said:

@DefinitelyNotARobot How the fuck can we determine, if you are a robot  or not ? 

Because it said so ;-)

[Carl-Richard]

51 minutes ago, zurew said:

Yes it is a mystery, thats why i don't agree with taking any strong positions here. 

What is the difference on a structural level between an artificial human and a 'real' human? What does a 'real' human made out of on the lowest levels, that cannot be replaced with dead matter?

 

So how do we distinguish between dead matter and living matter, what makes living matter 'living'?

My point in its simplest form is that taking one human feature (language) and projecting onto it a bunch of other features (sentience) is problematic. I've simply given a detailed account of that. So I would have the same problem with projecting human states of mind onto a hyper-intelligent alien if it somehow fell outside of the domain of biological life (metabolism) or was extremely structurally or behaviorally dissimilar. That said, again, this is only about the parsimony of logical inferences, not about reality as it actually is.

[zurew]

 

5 minutes ago, Carl-Richard said:

My point in it's simplest form is that taking one human feature (language) and projecting onto it a bunch of other features (sentience) is problematic.

Yeah i agree with you. We shouldn't be projecting other qualities onto it, just because one quality is there. But we can never know, at the end of the day.

Edited June 14, 2022 by zurew

[Carl-Richard]

54 minutes ago, DefinitelyNotARobot said:

@Carl-Richard Do you think that it's possible for AI to develope sentience (whatever it means to you) somewhere in the future, or do you believe that there are some inherent differences between biological and mechanical entities, which can't be overcome even by the most complex programming and engineering.

If you want to be as sure as possible that you'll create human states of mind, the question would be if we could create an artificial cell that could sustain metabolism, homeostasis, and then self-replicate and pass on instructions to its descendants. In a sense, that would be the moment we discover abiogenesis, and we've just created new life. My guess is that such a cell would be more or less identical to a biological one. Natural selection isn't stupid. To then get to human states of mind, you'll just have to make another human.

[Leo Gura]

55 minutes ago, Carl-Richard said:

Via. natural selection?

Via artificial selection.

Via human engineers and also these neural networks learn and grow once they are activated. Each year the neural networks become more sophisticated thanks to human engineers upgrading them by using the lessons from the prior networks they engineered and tested.

All technology is evolutionary. Cars evolve, PCs evolve, smart phones evolve, food evolves, robots evolve, software evolves, airplanes evolve, etc. This evolution must be taken as seriously as biological evolution.

Think of it like this, humans have basically engineered all modern breeds of cats and dogs using what is called "artificial selection". This is real evolution. Well, but humans are also doing this with non-biological tech devices in even more radical ways.

You don't need biology, life, or genes to have evolution. You just need some way of storing and updating memory. That could be via genes or it could be via computer files.

[AtheisticNonduality]

2 hours ago, Leo Gura said:

Think of it like this, humans have basically engineered all modern breeds of cats and dogs using what is called "artificial selection". This is real evolution. Well, but humans are also doing this with non-biological tech devices in even more radical ways.

Cats and dogs which already had biological mammalian levels of complexity and not like how computers are designed: from scratch, skipping all the fundamentals. This is what makes me think genetic engineering is the more promising field for hyper-intelligence and other desired "superiorities" since we already have the structures and just need to adjust them, but people are more afraid of that. There was a scientist in China that edited human genes to make an infant immune to HIV and was then condemned and imprisoned for it. That is much more positively interesting to me than some program replicating human thoughts and a bunch of clowns blowing it out of proportion, although AI is more concerning because it could end up as a false God, or as an infinite-seeming intelligence that is God without all the foundational qualities of biology.

[AtheisticNonduality]

3 hours ago, Carl-Richard said:

If you want to be as sure as possible that you'll create human states of mind, the question would be if we could create an artificial cell that could sustain metabolism, homeostasis, and then self-replicate and pass on instructions to its descendants. In a sense, that would be the moment we discover abiogenesis, and we've just created new life. My guess is that such a cell would be more or less identical to a biological one. Natural selection isn't stupid. To then get to human states of mind, you'll just have to make another human.

Yes, the only thing exactly correlating with human consciousness is a human brain. A machine simplifies aspects of human consciousness, will correlate with a simplified consciousness, if even nonliving matter has the proper traits for sentient emergence.

Edited June 14, 2022 by AtheisticNonduality

[Jodistrict]

7 hours ago, axiom said:

I think it understands in the same way humans understand, i.e. it grasps the meaning and context of different words. Your description of pattern recognition and neural networks fits humans or AI.

It doesn’t grasp the meaning.  It is pattern recognition using neural networks.  There is no model for understanding.  You are making assumptions without even knowing how the program works.   The program gives a response that sounds like a real human, but try asking it to solve a real problem for you.  One of the first AI programs, imitated a psychotherapist, because in that domain it was possible to match every query with a response that sounded intelligent.  Look at the history of AI – it is a history of hype followed by disillusionment (google “AI winter”).  The talk of “sentience” just adds to the hype.

[Carl-Richard]

@Leo Gura That's all fine. It doesn't really affect the crux of my point as stated earlier, which is essentially just that humans are very different from machines. Biological evolution is just one lens of looking at that difference.