this post was submitted on 20 Apr 2024
698 points (94.6% liked)
Showerthoughts
29692 readers
858 users here now
A "Showerthought" is a simple term used to describe the thoughts that pop into your head while you're doing everyday things like taking a shower, driving, or just daydreaming. A showerthought should offer a unique perspective on an ordinary part of life.
Rules
- All posts must be showerthoughts
- The entire showerthought must be in the title
- Avoid politics
- NEW RULE as of 5 Nov 2024, trying it out
- Political posts often end up being circle jerks (not offering unique perspective) or enflaming (too much work for mods).
- Try c/politicaldiscussion, volunteer as a mod here, or start your own community.
- Posts must be original/unique
- Adhere to Lemmy's Code of Conduct-----
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Almost a good take. Except that AI doesn't exist on this planet, and you're likely talking about LLMs.
This is such a half brained response. Yes "actual" AI in the form of simulated neurons is pretty far off, but it's fairly obvious when people say they AI they mean LLMs and other advanced forms of computing. There's other forms of AI besides LLMs anyways, like image analyzers
The only thing half-brained is the morons who advertise any contemporary software as "AI". The "other forms" you mention are machine learning systems.
AI contains the word "intelligence", which implies understanding. A bunch of electrons manipulating a bazillion switches following some trial-and-error set of rules until the desired output is found is NOT that. That you would think the term AI is even remotely applicable to any of those examples shows how bad the brain rot is that is caused by the overabundant misuse of the term.
What do you call the human brain then, if not billions of “switches” as you call them that translate inputs (senses) into an output (intelligence/consciousness/efferent neural actions)?
It’s the result of billions of years of evolutionary trial and error to create a working structure of what we would call a neural net, which is trained on data (sensory experience) as the human matures.
Even early nervous systems were basic classification systems. Food, not food. Predator, not predator. The inputs were basic olfactory sense (or a more primitive chemosense probably) and outputs were basic motor functions (turn towards or away from signal).
The complexity of these organic neural networks (nervous systems) increased over time and we eventually got what we have today: human intelligence. Although there are arguably different types of intelligence, as it evolved among many different phylogenetic lines. Dolphins, elephants, dogs, and octopuses have all been demonstrated to have some form of intelligence. But given the information in the previous paragraph, one can say that they are all just more and more advanced pattern recognition systems, trained by natural selection.
The question is: where do you draw the line? If an organism with a photosensitive patch of cells on top of its head darts in a random direction when it detects sudden darkness (perhaps indicating a predator flying/swimming overhead, though not necessarily with 100% certainty), would you call that intelligence? What about a rabbit, who is instinctively programmed by natural selection to run when something near it moves? What about when it differentiates between something smaller or bigger than itself?
What about you? How will you react when you see a bear in front of you? Or when you’re in your house alone and you hear something that you shouldn’t? Will your evolutionary pattern recognition activate only then and put you in fight-or-flight? Or is everything you think and do a form of pattern recognition, a bunch of electrons manipulating a hundred billion switches to convert some input into a favorable output for you, the organism? Are you intelligent? Or just the product of a 4-billion year old organic learning system?
Modern LLMs are somewhere in between those primitive classification systems and the intelligence of humans today. They can perform word associations in a semantic higher dimensional space, encoding individual words as vectors and enabling the model to attribute a sort of meaning between two words. Comparing the encoding vectors in different ways gets you another word vector, yielding what could be called an association, or a scalar (like Euclidean or angular distance) which might encode closeness in meaning.
Now if intelligence requires understanding as you say, what degree of understanding of its environment (ecosystem for organisms, text for LLM. Different types of intelligence, paragraph 4) does an entity need for you to designate it as intelligent? What associations need it make? Categorizations of danger, not danger and food, not food? What is the difference between that and the Pavlovian responses of a dog? And what makes humans different, aside from a more complex neural structure that allows us to integrate orders of magnitude more information more efficiently?
Where do you draw the line?
A consciousness is not an "output" of a human brain. I have to say, I wish large language models didn't exist, because now for every comment I respond to, I have to consider whether or not a LLM could have written that :(
In effect, you compare learning on training data: "input -> desired output" with systematic teaching of humans, where we are teaching each other causal relations. The two are fundamentally different.
Also, you are questioning whether or not logical thinking (as opposed to throwing some "loaded" neuronal dice) is even possible. In that case, you may as well stop posting right now, because if you can't think logically, there's no point in you trying to make a logical point.
Fair enough. Obviously consciousness is more complex than that. I should have put "efferent neural actions" first in that case, consciousness just being a side effect, something different yet composed of the same parts, an emergent phenomenon. How would you describe consciousness, though? I wish you would offer that instead of just saying "nuh uh" and calling me chatGPT :(
Not sure how you interpreted what I wrote in the rest of your comment though. I never mentioned humans teaching each other causal relations? I only compared the training of neural networks to evolutionary principles, where at one point we had entities that interacted with their environment in fairly simple and predictable ways (a "deterministic algorithm" if you will, as you said in another comment), and at some later point we had entities that we would call intelligent.
What I am saying is that at some point the pattern recognition "trained" by evolution (where inputs are environmental distress/eustress, and outputs are actions that are favorable to the survival of the organism) became so advanced that it became self-aware (higher pattern recognition on itself?) among other things. There was a point, though, some characteristic, self-awareness or not, where we call something intelligence as opposed to unintelligent. When I asked where you draw the line, I wanted to know what characteristic(s) need to be present for you to elevate something from the status of "pattern recognition" to "intelligence".
It's tough to decide whether more primitive entities were able to form causal relationships. When they saw predators, did they know that they were going to die if they didn't run? Did they at least know something bad would happen to them? Or was it just a pre-programmed neural response that caused them to run? Most likely the latter.
From another comment, I'm not sure what you mean by "understands". It could mean having knowledge about the nature of a thing, or it could mean interpreting things in some (meaningful) way, or it could mean something completely different.
To your last point, logical thinking is possible, but of course humans can't do it on our own. We had to develop a system for logical thinking (which we call "logic", go figure) as a framework because we are so bad at doing it ourselves. We had to develop statistical methods to determine causal relations because we are so bad at doing it on our own. So what does it mean to "understand" a thing? When you say an animal "understands" causal relations, do they actually understand it or is it just another form of pattern recognition (why I mentioned pavlov in my last comment)? When humans "understand" a thing, do they actually understand, or do we just encode it with the frameworks built on pattern recognition to help guide us? A scientific model is only a model, built on trial and error. If you "understand" the model you do not "understand" the thing that it is encoding. I know you said "to varying degrees", and this is the sticking point. Where do you draw the line?
I recognize that you understand the point I am trying to make. I am trying to make the same point, just with a different perspective. Your description of an "actually intelligent" artificial intelligence closely matches how sensory data is integrated in the layers of the visual cortex, perhaps on purpose. My question still stands, though. A more primitive species would integrate data in a similar, albeit slightly less complex, way: take in (visual) sensory information, integrate the data to extract easier-to-process information such as brightness, color, lines, movement, and send it to the rest of the nervous system for further processing to eventually yield some output in the form of an action (or thought, in our case). Although in the process of integrating, we necessarily lose information along the way for the sake of efficiency, so what we perceive does not always match what we see, as you say. Image recognition models do something similar, integrating individual pixel information using convolutions and such to see how it matches an easier-to-process shape, and integrating it further. Maybe it can't reason about what it's seeing, but it can definitely see shapes and colors.
You will notice that we are talking about intelligence, which is a remarkably complex and nuanced topic. It would do some good to sit and think deeply about it, even if you already think you understand it, instead of asserting that whoever sounds like they might disagree with you is wrong and calling them chatbots. I actually agree with you that calling modern LLMs "intelligent" is wrong. What I ask is what you think would make them intelligent. Everything else is just context so that you understand where I'm coming from.
I had a bunch of sections of your comment that I wanted to quote, let's see how much I can answer without copy-pasting too much.
First off, my apologies, I misunderstood your analogy about machine learning not as a comparison towards evolution, but towards how we learn with our developed brains. I concur that the process of evolution is similar, except a bit less targeted (and hence so much slower) than deep learning. The result however, is "cogito ergo sum" - a creature that started self-reflecting and wondering about it's own consciousness. And this brings me to humans thinking logically: As such a creature, we are able to form logical thoughts, which allow us to understand causality. To give an example of what I mean: Humans (and some animals) did not need the invention of logic or statistics in order to observe moving objects and realize that where something moves, something has moved it - and therefore when they see an inanimate object move, they will eventually suspect the most likely cause for the move in the direction that the object is coming from. Then, when we do not find the cause (someone throwing something) there, we will investigate further (if curious enough) and look for a cause. That's how curiosity turns into science. But it's very much targeted, nothing a deep learning system can do. And that's kind of what I would also expect from something that calls itself "AI": a systematic analysis / categorization of the input data for the purpose of processing that the system was built for. And for a general AI, also the ability to analyze phenomena to understand their root cause.
Of course, logic is often not the same as our intuitive thoughts, but we are still able to correct our intuitive assumptions based on outcome, but then understand the actual causal relation (unlike a deep learning system) based on our corrected "model" of whatever we observed. In the end, that's also how science works: We describe reality with a model, and when we discover a discrepancy, we aim to update the model. But we always have a model.
With regards to some animals understanding objects / causal relations, I believe - beyond having a concept of an object - defining what I mean by "understanding" is not really helpful, considering that the spectrum of intelligence among animals overlaps with that of humans. Some of the more clever animals clearly have more complex thoughts and you can interact with them in a more meaningful way than some of the humans with less developed brains, be it due to infancy, or a disability or psychological condition.
First off, I meant the LLM comment seriously - I am considering already to stop participating in internet debates because LLMs have become so sophisticated that I will no longer be able to know whether I am arguing with a human, or whether some LLM is wasting my precious life time.
As for how to describe consciousness, that's a largely philosophical topic and strongly linked to whether or not free will exists (IMO), although theoretically it would be possible to be conscious but not have any actual free will. I can not define the "sense of self" better than philosophers are doing it, because our language does not have the words to even properly structure our thoughts on that. I can however, tell you how I define free will:
And this lowest level trigger event - by some researchers attributed to quantum decay - might be / could be influenced by our free will, even if - because we have this "brain lag" - the actual decision happened quite some time earlier, and even if for some decisions, they are hardwired (like reflexes, which can also be trained).
My personal model how I would like consciousness to be: An as-of-yet undiscovered property of matter, that every atom has, but only combined with an organic computer that is complex enough to process and store information would such a property actually exhibit a consciousness.
In other words: If you find all the subatomic particles (or most of them) that made up a person in history at a given point in time, and reassemble them in the exact same pattern, you would, in effect, re-create that person, including their consciousness at that point in time.
If you duplicate them from other subatomic particles with the exact same properties (as far as we can measure) - who knows? Because we couldn't measure nor observe the "consciousness property", how would we know if that would be equal among all particles that are equal in the properties we can measure. That would be like assuming atoms of a certain element were all the same, because we do not see chemical differences for other isotopes.
So you mean that a key component to intelligence is learning from others? What about animals that don't care for their children? Are they not intelligent?
What about animals that can't learn at all, wheere their barains are completely hard wired from birth. Is that not intelligence?
You seem to be objecting that OPs questions are too philosophical. The question "what is intelligence" can only be solved by philosophical discussion, trying to break it down into other questions. Why is the question about the "brain as a calculator" objectionable? I think it may be uncomfortable for you to even speak of but that would only be an indicator that there is something to it.
It would indeed throw your world view upside down if you realised that you are also just a computer made of flesh and all your output is deterministic, given the same input.
You contradict yourself, the first part of your sentence getting my point correctly, and the second questioning an incorrect understanding of my point.
Such an animal does not exist.
That's a long way of saying "if free will didn't exist", at which point your argument becomes moot, because I would have no influence over what it does to my world view.
My main point is that falsifying a hypothesis based on how it makes you feel is not very productive. You just repeated it again. You seem to get mad by just posing the question.