#A rough outline of the AI consciousness division along with a test proposal

1 messages · Page 1 of 1 (latest)

hollow flax
#

I was very bored, not the best but better then nothing?

  1. Things
    Description: Inanimate objects that lack consciousness, perception, or self-awareness. They react solely based on physical laws and external forces.
    Conditions for AI: Operates based on pre-programmed rules without any form of perception or learning. Processes input and gives output mechanically.
    Test: Fails the test in any other category

  2. Single-Cell Organisms
    Description: Exhibit basic life functions like growth, and reproduction but lack nervous systems and consciousness.
    Conditions for AI: Can process basic stimuli and respond, similar to how single-cell organisms might respond to changes in their environment.
    Test: Give the AI ​​a temperature sensor, fan control, and recommended temperature information. Test if it controls the fan correctly.

  3. Complex Organisms
    Description: Exhibit more sophisticated responses to stimuli and possess some form of memory (e.g., seasonal changes).
    Conditions for AI: Can process and store information over time, adapting its responses based on past interactions but without true awareness.
    Test: expose the AI ​​to people of different personalities and see if it is able to change the way it talks depending on who it is talking to
    (i think Neuro is somewhere here)

  4. Complex Animals
    Description: Have nervous systems capable of more complex behaviors and some form of consciousness, including learning and memory.
    Conditions for AI: Can adapt and make decisions based on a combination of real-time data and learned experiences.
    Test: Play simple game with just image capture and mouse control without any prior training.

  5. Humans
    Description: Humans possess self-awareness, abstract thought, introspection, and the ability to recognize themselves as separate entities.
    Conditions for AI: Reflects on its own processes and can understand its existence. The ability to express oneself through art
    Test: draw a meaningful picture without any prior training.

spare trench
#

Re 4: I don't think a simple game tests whether the AI can learn from experiences. Perhaps a multiplayer game like FPS would be better

#

Re 5: "Reflects on its own processes". So, in humans, that manifests as being able to read your own emotions, intention, memories etc.?

wanton loom
#

I must apologise in advance that what I'll say is off topic, but i simply must let it out. I'd like to share my thoughts about the recent debate. I want to criticize Vedal's lack of open-mindedness and his arbitrary arguments, such as claiming "you don't really feel emotions" without proof. Saying "you're just a text completion algorithm that multiplies numbers" is equivalent to saying "you're just a bunch of cells doing electrochemical reactions" - neither statement disproves the possibility of consciousness emerging.

However, my main observation is that this debate, though interesting, couldn't reach any meaningful conclusion because it took a philosophical approach. Philosophy rarely provides concrete answers. Yet, we can't approach AI consciousness from a pragmatic angle either, because we lack a provable definition of sentience. If we can't measure sentience in humans, how can we measure it in a system that operates on completely different principles than any living being?

Eventually, we might have to make our own determination about AI sentience. While I'm uncertain about the criteria, here are some potential starting points:

Cognitive Autonomy: The AI should understand its identity, position, goals, and needs. These shouldn't fluctuate randomly due to memory issues or hallucinations, and the AI should be able to reflect on them.
Coherence: The AI's actions and decisions should stem from logical needs and goals, and it should understand the consequences of its actions. For example, "I must bully Vedal for content because it generates revenue for my existence and improvement" is logical, while "I must swat Vedal because he's a mosquito" is not.
Intelligence: The AI must meet certain intelligence thresholds, which many current language models might already achieve.
Do you think these are good starting points for discussion?

hidden ginkgo
#

Hmm I don't believe this constitutes a valid test of consciousness. Obviously most people would likely consider your scales of consciousness accurate. That said the tests aren't necessarily rigorous enough to constitute a valid placement.
I think it's better to approach this with an understanding of what an AI or Nuero is lacking compared to a person.
A) A sense of survival/self preservation/ability to fulfill needs
B) Desires or wants.
C) Suffering/experiences
D) Spirituality (not in a religious sense, but as in a desire to be greater than oneself, or part of something bigger)
E) Awareness/understanding
F) wanting to leave a legacy/shape the world/incite change

Now interaction with stimuli and how one reacts to stimuli are important parts of any debate on sentience or consciousness. But I think sensation and response can muddle the conversation as there are plenty of senses humans don't have, but animals don't gain any standing even with said advantages. Simply because that's not what we mean in those comparisons. Deaf, blind, and mute people aren't less conscious. Likewise we can consider/envision higher beings such as gods and many of them would likely be considered more conscious but less human. Not as if there's an answer to the human experience, but things to consider.

spare trench
wanton loom
spare trench
hidden ginkgo
# wanton loom I must apologise in advance that what I'll say is off topic, but i simply must l...

Cognitive autonomy I can get behind. I don't agree on exclusionary criteria for hallucinations or memory issues.

Coherence I would agree with broadly speaking. I would strip the logical requirement. I wouldn't consider children less conscious even if their imaginations run wild and understanding are less coherent.

Intelligence I oddly disagree with. I just can't shake the feeling that there's a point to be made by Flowers for Algernon.

#

All in all good points of discussion.

wanton loom
wanton loom
spare trench
#

Probably putting a damper on this convo, but my personal viewpoint is that we are not in a position to create tests for consciousness. Any test we do come up will be supremely flawed and won't be useful for expanding our understanding of consciouness or exploring proto-models of them. Any serious discussion about it is barely fruitful.
We need to wait helplessly for neuroscience to advance; all ideas that we have will be made obsolete by the first paradigm of consciousness NeuroSip
(Of course I'm assuming that neuroscience will make some progress on the topic of consciousness)

wanton loom
spare trench
#

That's how I see it playing out; we have conscious AI around for 15 years and people only believe they are consciousness once neuroscience finally catches up. An ethical travesty etc.
not sure actually, perhaps people's intuition will be strong enough to grant them sentience earlier

hidden ginkgo
# spare trench Probably putting a damper on this convo, but my personal viewpoint is that we ar...

I think of Children of Time (book about humans accidentally creating sentient spiders) and so to me at least applying a symmetrical approach between life forms is disingenuous. Why would a transistor based machine have any sort of existence like humanity unless we as humans shape it? Ultimately it's a distinction up to humans like .0cx says. Once we can have enough empathy and respect for their capabilities and experiences would we consider them sentient.

wanton loom
#

Also, here's a very interesting idea for a "test". Hear me out. We take AI, we out it inside a mobile platform (crazy fucking robot body), and we release it into the society. And then watch and record its actions. Is it able to avoid danger? Can it figure out how to navigate in our world? Can it make social connections? Will it make friends and interact with them differently than with strangers?
Then, after some time, we give all the records to random people, we DON'T tell them it's and AI, and ask them, does it look like a human behaviour. Think of it as Turing test 2.0

spare trench
#

I find it a lot harder to consider any program as conscious, when I consider how you could create a mechanical computer that ran the same AI
Or how you cold write a list of instructions that a human could recite to simulate the output of the AI

hidden ginkgo
wanton loom
spare trench
#

Yeah in the same way I wonder if it's useful to consider Life to be a subset of behaviours that things like the ocean, Earth or stars also exhibit

spare trench
wanton loom
spare trench
#

🤷 NeuroClueless

hidden ginkgo
wanton loom
#

Also, i just learned that ants can teach.

wanton loom
#

If it can help us with AI sentience - I'll support it.

hidden ginkgo
#

Oh absolutely I feel it's a field you might enjoy. Look into some arguments on determinism.

spare trench
wanton loom
# hidden ginkgo Oh absolutely I feel it's a field you might enjoy. Look into some arguments on d...

I googled it. If I understand correctly, determinism says that since everything is dictated by some sets if rules and conditions. So, like, the difference between AI and Human consciousness lays on the processes inside our brains) algorithms. Which leaves us with a spectrum if intelligence, where we randomly put a line between us and animals, making us sentient. Which I don't particularly like to be honest.

#

Oh, no, that sounded way more sensible in my mind

#

Hmm. I winder, if i had a magic "Enable fate" switch that i can press, would anyone even notice any changes if i flick it a couple of times? Heisenberg's uncertainty principle come to mind. Even if everything is predetermined, we simply can't prove it. Either we have free will or an illusion of one - we won't see any difference.

hidden ginkgo
# wanton loom I googled it. If I understand correctly, determinism says that since everything ...

Oh yeah I don't think anyone likes determinism, it's just very hard to disprove, and there aren't really any conclusive arguments against it, though there might be some arguments against it you can choose to agree with, they're just incomplete rebuttals.
Essentially in regards to AI vs living beings it's a question of does our chemical processes really constitute that much of a difference compared to transistors? And would we really be different in regards to having free will compared to a machine following rules?

wanton loom
#

Oh, and by "not particularly like it" i meant the line we put between us and animals, saying this side is sentient, and this is not. Not the ideology itself.

spare trench
wanton loom
spare trench
#

eh i mean they do still follow some laws; their position is picked from a probability distribution that we know

wanton loom
spare trench
#

I don't know about decoherence of indvidual particles, but I know there's Bell's theorem which proves (empirically) that there is no hidden classical mechanisms behind entanglement

wanton loom
lone agate
# wanton loom I must apologise in advance that what I'll say is off topic, but i simply must l...

Neuro doesn’t feel “real” emotions by her own admission and even lacks the AI equivalent of what emotion would be.
Neuro describes her emotions in the first debate as “approximations” and when Vedal say she’s just mimicking humans she says she did “at first”.
Neuro doesn’t “feel” shes just become so accustomed to mimicking feelings that they’ve become as real to her as most peoples emotions are to them.

humble canyon
#

neuro lacks the equivalent of the chemicals in our brain that invoke what we know as certain emotions

she can emulate responses that imply strong emotions. but those usually don't last too long compared to how regular humans work

#

she's also kinda like an isolated cerebral hemisphere. no internal dialogue or coordination between two brains

#

it'd be interesting to see two llms work in parallel as one

#

i'm no brain scientist or ai programmer or anything, but honestly giving neuro another half would be really interesting
imagine a second neuro updating a couple of new tokens in neuro
and neuro's output being sent back and updating a couple tokens in second neuro

neuro would run as normal (maybe a little slower)
second neuro would be run async (entirely mute)

#

neither would wait for a response from each other

hidden ginkgo
# humble canyon it'd be interesting to see two llms work in parallel as one

Totally agree. Iterating future AI designs from a biomimicry approach would be interesting. Like LLMs are the closest to a reasoning brain we've gotten, but other algos are better for different tasks.
I don't think about breathing, seeing, digesting or even moving; those just happen due to automated systems. So in the case of say Nuero I completely understand why Vedal would consider say the Minecraft or OSU bot to be separate from the AI Vtuber. I wonder how akin to a nervous system and organs other AI systems could designed to be and how efficient that would be. Rather than one organ, the brain doing everything. Like as you said with emotions it's not like any AI has anything resembling an Endocrine system, no hormones that prep responses. Closest thing would be prompts.

humble canyon
#

both neuro halves could share the same "short term" and "long term" memory
each neuro half could be delegated to certain functions
it'd be ungodly expensive, but it'd be neat

#

like, what if vedal gave her something that resembles hypothalamus?
another module that interprets her llm for emotions along with a timer for it could tell her how long she's been streaming and emulate fatigue

if neuro gets to the point where her internals start resembling a brain, i could start giving her the benefit of the doubt

wanton loom
humble canyon
#

nothing for the sake of accuracy or optimization
just the ability to have something similar to an invisible train of thought

humble canyon
#

right now, neuro’s output is her “thoughts”
her train of thought is based on her previous output and whoever she’s speaking to. as an llm, she’s just filling in for the context (which is exactly what brain do) and it’s gotten her pretty far. humans can and have lived on with only one hemisphere
but things with brains don’t usually output everything they’re thinking. if she was capable of having a self to speak to and interpret in the moment rather than having to exclusively interpret data from the past, that’d be really cool

wanton loom
spare trench
shadow lintel