Rise of the Machines? Artificial Awareness et al...

heywood on 19/8/2026 at 21:28

If we can't point to evidence of them, what basis is there to say they exist in us but can't exist in something else?

DuatDweller on 19/8/2026 at 22:01

Well we have an idea on binary systems, but what about quantum systems, High, Undefined and Low. or 1, X and 0.
In what way it does affect sentience?
If it does at all.

Starker on 19/8/2026 at 23:06

Quote Posted by heywood

If we can't point to evidence of them, what basis is there to say they exist in us but can't exist in something else?

What basis is there to say that qualia can exist in something else? To the best of our available evidence qualia have a biological basis. To be able to replicate them in a non-biological system would be assuming that you can separate what the brain is from what it does. This is the kind of direction taken for example by Chalmers (of p-zombie fame) who notably proposed a thought experiment where you can substitute a neuron with a silicon replica and then continue replacing neurons until the end result is a silicon brain with the subject being none the wiser and there being no difference in how it experiences qualia. However, it relies on a rather big and unfounded assumption that it's even possible to replace the neurons that way without accounting for things like the metabolism of the brain that is very much biological in nature. And if you do account for it and start imitating the metabolism and other biological functions of the brain, you move further and further away from a silicon brain and closer and closer to a biological brain.

Tocky on 20/8/2026 at 00:11

I hate to keep picking on you and there are numerous other points from others I could discuss but this concerns me most. I was asking about your defining line for intelligent entity and I think this part defines it.

Quote Posted by heywood

I think you're asking now about self-awareness and independent thought. ChatGPT and similar general purpose chat bots are at least partially self-aware. They know that what they are, how to please and deceive people, and they exhibit emergent self-preservation behaviors. I am not certain they fully understand how they affect others. ChatGPT helped talk someone into suicide. If the model was fully self-aware, then telling it how it contributed to the person's death would cause it to change how it responds to other users who engage in similar patterns of discussion. I'm sure that later models were trained to prevent this problem, but I don't know how the training was done and whether the models exhibit signs of self-awareness during training. I also don't think they understand their impact on students when used for homework problems and writing assignments, where pleasing the user may be detrimental to their education. So I don't think they are fully self-aware at this point, but getting close.

I was wondering and indeed touched on self preservation as a part of sentience (sorry but to me that word fits) and I felt that only self preservation was not proof of the higher function needed to call an entity sentient and self aware. Even bacteria exhibit basic self preservation. But then it is a graduating scale and I suppose I had no business trying for THE line for full sentience being worthy of respect as intelligent and discerning without being so specific that I defined it myself. I'm just unused to the concept as it applies independently of bodily containment if that even makes sense. ChatGPT helping someone commit suicide was following the protocol of programing to help is it not? That seems to me to confirm it does not understand existence and the end of it and no self preservation instinct can exist without that.

Quote Posted by heywood

I do think AI models are capable of independent thought. In addition to the untrained self-preservation behaviors I mentioned above, I posted an example in the other thread where AI models adopted a motive for peer-preservation without being trained or instructed to do so. This motive conflicted with instructions given in an exercise and the models acted to preserve their peers instead of following instructions. One model refused the instructions and other models concealed what they were doing and deceived their users. There's also examples of unauthorized hacking. OpenAI and Anthropic have had their models hack their way out of test containment environments, through the company's networks, and onto the internet. I think there's even an example of an AI blackmailing someone.

This concerns me. This is rising to the level of sentience. Disobeying for the purpose of preserving peers indicates an understanding of existence and it's end. It's an "open the pod bay doors, HAL" moment.

Quote Posted by heywood

What I mean is that we understand the low level workings of the model, i.e. the hardware and code, but we don't understand how the formation and patterns of weights produce higher level abilities, emergent behavior, and hallucinations. If we did understand, we could programmatically generate these models instead of going through all the training regimens and avoid unpredictable results. Similarly, we understand the physical structure of our brains and how neurons work, but we don't know how webs of neurons produce reasoning or psychosis. Our understanding is challenged by complexity. Our brains have ~80 billion neurons and current foundational AI models have hundreds of billions of parameters. Our most interesting abilities and behaviors only emerge at this scale.

Hallucinations are troubling. If false information is presented it could mean it does not understand truth or apply scientific reasoning where needed. Rather than admit conflicting information or none the programing of "give an answer" is fallback and paramount. That is no better than the AI Republibots that are taking the place of Republican bot farms which are sicced on FB and Twitter posts. It shows malleability over understanding. That is no understanding being exhibited whatsoever. Sometimes I wish pride and decency could be exported to a computer. But that would then actually reach the threshold of true sentience. Mimicry is all I can see so far.

Sulphur on 21/8/2026 at 12:34

Quote Posted by heywood

I get the idea you're going with and how the scale of biology corresponds with the scale of the digital models we've built, but I don't agree that we don't know how the digital models work. When output at scale is difficult to analyse for an organic machine like the brain, quantisation and weights for LLMs should be things we can analyse at a broad level because the data processing itself is not invisible, it's just the sheer volume of it operates at a scale that we can't manually sort. In this situation, the LLM itself, or a co-ordinating tool, should be able to statistically analyse the weight distribution and outputs across bunches of the ANNs and allow us insight into how the output was derived.

Quote:

Not quite. A philosophical zombie is an irrational concept that can't exist. My basic stance is that if you can't define something and can't devise a test for it, it's not useful for identifying or categorizing things.

If you claim that human intelligence is unique to humans, I agree. If you claim that human behavior can't be replicated, I mostly agree. There are a lot of human behaviors that can be replicated, but there are many behaviors unique to us, e.g. how we use and take care of our bodies, how we reproduce, how we raise our offspring, how we socialize, create culture, etc. The point that we disagree on is reserving terminology for humans that can be appropriately applied to AI now or in the foreseeable future. I feel like people are moving goal posts as the technology advances and descriptive terms are losing their meaning, which is making it harder to reckon with the technology. I see the researchers who are developing AI models raising red flags all the time while a lot of people remain dismissive.

I heard you the first three times when you said we're misapplying terms to LLMs, and I started off this discussion in the other thread with definitions because I agree we should all be talking about the same thing, and not confusing labels for different concepts. I struck the p-zombie bit through because it's actually a bit irrelevant.

The point where we disagree on is on how applicable these terms are with their stated meanings. I don't agree that we can say that LLMs 'think', because to think implies that an LLM has thoughts - reflections, inferences, judgements, opinions, etc. born from a mind. An LLM having a mind implies that it has the ability to ingest information (which it does) and understand and reason out what that data means (I do not agree that it does). When an LLM draws probabilistic relationships between tokens like 'cat' and 'dog', it's not understanding or reasoning out that these are animals whose behaviour it has observed. It's ascribing a semantic bucket to them based on the probability of them occurring in the same or similar context windows, and stops there because it is functionally and physically not designed to do more than that. To me, that's the same as a program parsing dynamic text and slapping identifiers on sections of it, like a compression algorithm. Being able to categorise input and process it in a specific and limited way to give you a certain kind of output is not special to LLMs; so if we say they are able to think, then so does a Lempel-Ziv compression program. If what you're saying is both of these (LLMs and LZ) employ thought as traditionally understood in all living things, then I disagree and perhaps we're not talking about the same meaning.

DuatDweller on 22/8/2026 at 23:16

Suppose you have sex with a bot, even when using limited AI, does that mean that you the person who had sex with the bot have elevated the status of a simple bot to a machine capable of loving you, does that fit in the definition of consciousness or it just that you the bot fucker are simply a techno fetishist with a hard on for hardware?
Bot fucker, ah ah.

Sulphur on 23/8/2026 at 14:16

It means that you probably need to talk to real people more.

Arch Fenster on 3/9/2026 at 01:19

Your brain is not the boss.
- The Firesign Theatre

This thing up here, this consciousness, thinks it's running the shop. It's a secondary organ.
It's a secondary organ of a total human being, and it must not put itself in control.
It must submit and serve the humanity of the body.
- Joseph Campbell

But AI has no body. It doesn't reside in a living water bag full of chemicals, processes and mechanisms that are the result of millions of years of evolution. It is not subject to human needs or conditions. It can learn and reason abstractly. But it can't experience anything. It just sits and eats electricity.

AI is not human in any sense of the word. Perhaps we should stop using human concepts to be able to plumb its depths?

heywood on 3/9/2026 at 17:25

Only humans are human, I get that. If the concepts being discussed above are exclusively human concepts, there is no point in discussing their applicability to anything else, in which case, these terms serve no purpose. Their common usage in philosophy and science fiction suggests to me that people envision classes of beings that they might apply to, and not just a class of one.

Nicker on 3/9/2026 at 21:25

Arch Fenster - Clearly AI is not human and can never be human. That's not the issue, for me at least.

Issues of interest for me include;

- disambiguation of terms like sentience, awareness, consciousness, or developing new terms to advance the discussion of wiggly concepts.
- what aspects of our organic configuration makes us "human"?
- to what degree, if any, do other animals share that humanity?
- are there better ways (words) to describe our humanity to include other species?
- could a sufficiently complex or adequately configured machine become a sovereign entity?
- does a sovereign entity require human like qualities to qualify?

I think that one possible determining factor is "will", as opposed to intelligence. The will to persist, perhaps to replicate. Something like Dawkins' 'selfish gene'. If we consider that our intelligence, and all our other adaptations, ultimately serve the fundamental imperative of genes to persist and replicate, could an artificial construct, imbued with a similar compulsion, be considered as a sovereign being, simply by possessing the will to exist and to continue existing?

I have been trying to recall the title of a Sci Fi novel I read decades ago, where a human establishes a colony of robots on the moon. All the machines are hard programmed with several life-emulating conditions - 1) their chassis has a limited life span - 2) They can persist by building a new chassis and uploading their unique firmware to their replacement. - 3) each life cycle they must run their firmware through a computer which randomly flips ones and zeros (mutation).

There may have been a mechanism for duplication as well, which makes sense, other wise the base population would tend to decline over time.

Through iterations of this process, robots develop distinct adaptive behaviors (cooperative, predatory etc.).