3 ms·

Because statistics don't squeal when you stick a pin in them. We have bodies, we're influenced by our bodies, our entire perceptual universe is filtered throug

Because statistics don't squeal when you stick a pin in them. We have bodies, we're influenced by our bodies, our entire perceptual universe is filtered through the limitations of our bodies, and - most of all - our consciousness is literally embodied. By far the worst thing about AI research - and in fact CS in general - is that it overvalues abstraction, especially language, and undervalues embodiment. In fact there's an almost culty refusal to take embodiment seriously as a bedrock factor in consciousness. Not conscious-like simulated behaviour, but the experience of being conscious. Without embodiment you can make a ChatBot, but it's just Eliza++. It has no idea what blue looks like, or how fear or joy feel. It can supersearch for human-generated linguistic associations around those experiences, but the experiences themselves do not exist for it. It doesn't even have the capacity to have experiences in an analogous way. You can argue that it's having experiences in a different way, but that's exactly the point of the article. We can't know, but we will ascribe consciousness to certain behaviours - and especially to language - whether or not it's there. At the same time we ignore evidence of sentience in animals - or dismiss it as unsophisicated - because we don't see animals speaking or writing. Ignoring or minimising evidence of animal sentience - even animal language - while being impressed by a self-designed automaton explicitly designed to mimic our language is an ironically un-self aware take.

MrScruff 4y ago
If that's the point then it feels like a straw man. It literally quotes a couple of the articles saying 'I know it's not sentient'. If the suggestion is that there are a load of 'smart people' claiming Bing chat is sentient then I've yet to see it. People are saying, wow, it's pretty crazy how well LLMs are scaling - what's coming next? I agree that just because LLMs have been trained to model human language, doesn't at all mean they'd inherit any of the conscious experience of humans. But then since we don't know what consciousness is, it's rather hard to say whether or not an analogous phenomenon could arise in an AI agent. I would assume it would be very alien to us if it did, given it would have arisen through a completely different process. But if it did, and the AI had been trained to avoid a specific state, I could imagine such a 'conscious' AI might experience something analogous to pain when pushed towards that state.
riknos314 4y ago
This is a great insight. Considering the idea of embodiment made me think of the context in which the neurons in our brain operate. They don't fire in isolation, but rather in the presence of neurotransmitters which change the context surrounding the neurons and influence the way in which they activate. If you have too much coffee your brain behaves differently than if you haven't. The current AI models I'm familiar with only have analogs for the neurons themselves, with no ability to model any of the surrounding neurochemical context (which comes almost entirely from embodied interaction with the environment). ChatGPT (et al) has no ability to adjust how it interacts with a user based on their handshake, body odor, what the weather is like, if it slept well the previous night, if they have eaten too little potassium recently, etc.. All of these things are critical to the embodied experience of humans. Being able to mimic speech in response to a textual input in no way implies that the model actually understands and mimics the process that humans go through to produce that speech.
jdietrich 4y ago
>In fact there's an almost culty refusal to take embodiment seriously as a bedrock factor in consciousness. We're researching artificial intelligence, not artificial consciousness. People who are building AI systems largely don't care about the question of consciousness, because there's no evidence to suggest that it's in any way necessary to do the kind of things we want to do with AI. The people who are very concerned about issues like consciousness and embodiment are in no way being prevented from running their own experiments, but they aren't running many experiments and those experiments they do run largely aren't particularly successful or interesting. The experiments of people who are just training models to do stuff are generally very successful and very interesting.
- moremetadata 4y ago
  >We're researching artificial intelligence, not artificial consciousness. They go hand in hand, otherwise how can you have consciousness? One element of consciousness is being aware of existential threats in order to make intelligent decisions. Some of those threats are easier to learn from than others, namely because it causes pain or does something that harms the feedback loop ego but is not an existential threat.
  - Jensson 4y ago
    > One element of consciousness is being aware of existential threats in order to make intelligent decisions No, your subconsciousness does that. When you are asleep and your subconsciousness hears a scream it might decide to wake you up, or it might deem it wasn't a signal of danger and let you sleep. So a process can be very smart without being conscious, they are two separate things. ChatGPT doesn't behave like a conscious being, it behaves like a subconsciousness at best.
    - moremetadata 4y ago
      >When you are asleep and your subconsciousness hears a scream it might decide to wake you up You arent taking into account the sleep phase an individual is in at the time the noise is heard, or the state of their immune system. For example the state of an individuals immune system could mean they are a light sleeper or a deep sleeper. >ChatGPT doesn't behave like a conscious being, it behaves like a subconsciousness at best. Its using a reinforcement learning model but these are basic because they dont take into account how chemistry works in biological systems. Chemistry in biological systems is the reinforcement learning model for lifeforms, including higher species and so when a reinforcement learning model for AI can quantify the chemicals effects, then you can get closer to imitating biological lifeforms and when that happens lifespans will drastically increases, which is why big pharma also has a dog in this race!
hackinthebochs 4y ago
Embodiment is an intellectual dead end in explaining consciousness/sentience. The fact that some pattern of signals on my perceptual apparatus is caused by an apple in the real world does not mean that I have knowledge or understanding of an apple in virtue of this causal relation. That my sensory signals are caused by apples is an accident of this world, one we are completely blind to. If all apples in the world were swapped with fapples (fake apples), where all sensory experiences that have up to now been caused by apples are now caused by fapples, we would be none the wiser. The wide content of our perceptual experiences is irrelevant to literally everything we know and how we interact with the world. Our knowledge of the world is limited to our sensory experiences and our deductions, inferences, etc derived from our experiences. Our situatedness in the world is only relevant insofar as it entails the space of possible sensory experiences. Our sensory experience is the medium by which we learn about the external world. We learn of apples not because of the redness of the sensory experience, but because the pattern of red/not-red experience entails the shape of apples. Conscious experience provides the medium, modulations of which provide the information about features of the external world. It is analogous to how modulations of electromagnetic waves provides information about some distant information source. Understanding consciousness is an orthogonal matter to one's situatedness in the world, just like understanding electromagnetic waves is orthogonal to understanding the information source being modulated into them.

MrScruff 4y ago

If that's the point then it feels like a straw man. It literally quotes a couple of the articles saying 'I know it's not sentient'. If the suggestion is that there are a load of 'smart people' claiming Bing chat is sentient then I've yet to see it. People are saying, wow, it's pretty crazy how well LLMs are scaling - what's coming next? I agree that just because LLMs have been trained to model human language, doesn't at all mean they'd inherit any of the conscious experience of humans. But then since we don't know what consciousness is, it's rather hard to say whether or not an analogous phenomenon could arise in an AI agent. I would assume it would be very alien to us if it did, given it would have arisen through a completely different process. But if it did, and the AI had been trained to avoid a specific state, I could imagine such a 'conscious' AI might experience something analogous to pain when pushed towards that state.

riknos314 4y ago

This is a great insight. Considering the idea of embodiment made me think of the context in which the neurons in our brain operate. They don't fire in isolation, but rather in the presence of neurotransmitters which change the context surrounding the neurons and influence the way in which they activate. If you have too much coffee your brain behaves differently than if you haven't. The current AI models I'm familiar with only have analogs for the neurons themselves, with no ability to model any of the surrounding neurochemical context (which comes almost entirely from embodied interaction with the environment). ChatGPT (et al) has no ability to adjust how it interacts with a user based on their handshake, body odor, what the weather is like, if it slept well the previous night, if they have eaten too little potassium recently, etc.. All of these things are critical to the embodied experience of humans. Being able to mimic speech in response to a textual input in no way implies that the model actually understands and mimics the process that humans go through to produce that speech.

jdietrich 4y ago

>In fact there's an almost culty refusal to take embodiment seriously as a bedrock factor in consciousness. We're researching artificial intelligence, not artificial consciousness. People who are building AI systems largely don't care about the question of consciousness, because there's no evidence to suggest that it's in any way necessary to do the kind of things we want to do with AI. The people who are very concerned about issues like consciousness and embodiment are in no way being prevented from running their own experiments, but they aren't running many experiments and those experiments they do run largely aren't particularly successful or interesting. The experiments of people who are just training models to do stuff are generally very successful and very interesting.

moremetadata 4y ago
>We're researching artificial intelligence, not artificial consciousness. They go hand in hand, otherwise how can you have consciousness? One element of consciousness is being aware of existential threats in order to make intelligent decisions. Some of those threats are easier to learn from than others, namely because it causes pain or does something that harms the feedback loop ego but is not an existential threat.
- Jensson 4y ago
  > One element of consciousness is being aware of existential threats in order to make intelligent decisions No, your subconsciousness does that. When you are asleep and your subconsciousness hears a scream it might decide to wake you up, or it might deem it wasn't a signal of danger and let you sleep. So a process can be very smart without being conscious, they are two separate things. ChatGPT doesn't behave like a conscious being, it behaves like a subconsciousness at best.
  - moremetadata 4y ago
    >When you are asleep and your subconsciousness hears a scream it might decide to wake you up You arent taking into account the sleep phase an individual is in at the time the noise is heard, or the state of their immune system. For example the state of an individuals immune system could mean they are a light sleeper or a deep sleeper. >ChatGPT doesn't behave like a conscious being, it behaves like a subconsciousness at best. Its using a reinforcement learning model but these are basic because they dont take into account how chemistry works in biological systems. Chemistry in biological systems is the reinforcement learning model for lifeforms, including higher species and so when a reinforcement learning model for AI can quantify the chemicals effects, then you can get closer to imitating biological lifeforms and when that happens lifespans will drastically increases, which is why big pharma also has a dog in this race!

hackinthebochs 4y ago

Embodiment is an intellectual dead end in explaining consciousness/sentience. The fact that some pattern of signals on my perceptual apparatus is caused by an apple in the real world does not mean that I have knowledge or understanding of an apple in virtue of this causal relation. That my sensory signals are caused by apples is an accident of this world, one we are completely blind to. If all apples in the world were swapped with fapples (fake apples), where all sensory experiences that have up to now been caused by apples are now caused by fapples, we would be none the wiser. The wide content of our perceptual experiences is irrelevant to literally everything we know and how we interact with the world. Our knowledge of the world is limited to our sensory experiences and our deductions, inferences, etc derived from our experiences. Our situatedness in the world is only relevant insofar as it entails the space of possible sensory experiences. Our sensory experience is the medium by which we learn about the external world. We learn of apples not because of the redness of the sensory experience, but because the pattern of red/not-red experience entails the shape of apples. Conscious experience provides the medium, modulations of which provide the information about features of the external world. It is analogous to how modulations of electromagnetic waves provides information about some distant information source. Understanding consciousness is an orthogonal matter to one's situatedness in the world, just like understanding electromagnetic waves is orthogonal to understanding the information source being modulated into them.