Philosophy Discussion Forums | A Humans-Only Club for Open-Minded Discussion & Debate

[Lagayascienza]

I don't know about AI per se being an existential threat anytime soon. It is the purposes to which AI is put by big corporations that is the immediate problem. I was never on X and I quit Facebook years ago. If I knew how to ditch Meta completely, and Google and Amazon, I would. They know too much about me already and they can't keep my data safe. Google, for example, knows my search history and it will no longer let me use it's search engine without logging in first. Amazon knows what I've read and what I'll want to read next because I use a Kindle. And ads are so pervasive and intrusive these days that there should be laws against it. The business models and the algorithms of these big comporations are designed so that one cannot escape their clutches.

[PrabhakarB]

My definition of intelligence in my previous post evokes a related question -- must intelligence be acquired rather than implemented and implanted? Acquired means the AI should pass through a stage where the system is not intelligent but has the potential to be so. The difference is that data and methods to use the data are both available in the implanted case. In the acquired case there is very little built-in data or knowledge initially.
The above describes two fundamentally different approaches to building AI systems. One is a data or knowledge-centric one and the other is about capturing functional and structural aspects of intelligence.
In terms of programming, it is the difference between functional programming and procedural programming. It is the philosophical difference between the dominance of data versus that of (functional) structure. Should the function conform to the data or must the data conform to the required functionality?
Both approaches have their challenges. However, data is never-ending, meaning the data-centric approach will forever remain incomplete. Functional structure in contrast is limited. Any amount of data and knowledge can in theory be absorbed here. Therefore, this approach offers the best chance of achieving true artificial intelligence. However, this requires a deeper exploration of the origins of the possibility of intelligence facilitated by using better abstractions. Sadly, the implementation of these ideas is currently a mere dream.

[obbeel]

Generative AI does more than just search for an answer. It thinks in 'tokens' which are used to form sentences. It could be said that it is thinking to generate answers, and it also generates unique answers. It's more than what it looks like on the surface, which is why you can find answers in ChatGPT or Claude that you can't find on the web.

[PrabhakarB]

Attributing tokens like 'thinking' to an algorithm with constrained behavior is dangerous. The constraints come from the structure of the source data used in training. The LLM approach could best be described as an attempt to discover the mathematical structure necessary to produce the data humans have created. Tokens are used to convert verbal data to numbers for computations and have no meaning within the structure. All interactions with LLMs can be seen abstractly as representing the same, single query:
"Say something reasonable using the tokens and structures in the input using the similarities of the structures you have derived from the large amount of data you were fed."
The easiest way to see the problem with this approach is to consider the minor variations in the structure that could alter the statement's truth. Statistical routes to truth are often wrong. Retrieval Augmented Generation (RAG) is an attempt to overcome this. Notice how RAG is attempting to reintroduce context data back into the structure. This was the point of my previous post. The structure should be independent and capable of handling all contexts. Then, context data should be supplied to the algorithm that uses the structure to process that data. This requires that the structure should be capable of handling all contexts. This can best be achieved by designing this from first principles founded on sound philosophical ideas.
Reverse-engineering truth from past data as the principal mechanism to assist an active agent in dealing with the present is not logical.

[obbeel]

Some scientists, as of last month, are attributing consciousness to Quantum effects. And it is just like you said - it's probabilistic and statistical. I understand that keeping the meaning of truth and setting it to a determined quantity of variables is bad for many of the things that a human can think or do, but at the same time, it brings technological and scientifical advance.

Don't we want to understand what human consciousness is all about? What is bad about giving that role to Science? Wouldn't it be better if Human and Natural Sciences finally got to work together to find what is True?

The article is 'Microtubule-Stabilizer Epothilone B Delays Anesthetic-Induced Unconsciousness in Rats' from eNeuro.

[PrabhakarB]

Consciousness is best understood not in terms of its other extreme of unconsciousness but in terms of confusion. Brains and computers must process electrical/chemical signals of different types, using various algorithms to update memory locations with their outputs. Then, interpret the changes to the memory locations. In brains, memory locations have semantic associations to help the interpretation.
The location-based semantics in brains must be duplicated in computers as a first step to facilitate the interpretation. The interpretations of the signals from different sensors must be consistent and corroborative. Otherwise, confusion will result. Consistent and confusion-free interpretations, essential for immersive experiences that give a feeling of being there, require memory-location-based semantics to have greater structure to capture all the spatiotemporal cues in the sensory data.
Imagine the confusion that would result if our sense of touch or position does not agree with our vision. Even an out-of-synch audio in a TV program can be annoying and confusing. In other words, consciousness might be as simple as enabling the computer to construct (for itself), an immersive spatiotemporal representation.
Implementing the required functionality would certainly vary between the brain and the computer. Even if the brain exploited quantum effects, that would not necessarily mean that computers must do the same. Simple computations might be enough.

[Pattern-chaser]

Some scientists, as of last month, are attributing consciousness to Quantum effects.

I think this has been suggested quite some time ago. Didn't Roger Penrose have ideas along these lines, maybe 20 (???) years ago? Microtubules, I think?

Don't we want to understand what human consciousness is all about? What is bad about giving that role to Science?

If you are using screws, will you employ a screwdriver, or a stepladder, to use them? Horses for courses. There are many reasons why science is not the optimum tool for this job, I think?

[Lagayascienza]

If science cannot provide answers to the question of what consciousness is, then nothing can. Consciousness is a natural phenomenon that emerges from the operation of brains which are physical objects whose workings are goverened by the laws of nature like everything else. There are no other ways of finding out what consciousness is. Or anything else for that matter. The supernaturalists are not happy with this. They say there are "other ways of knowing". But they cannot tell us what these other ways are. And, anyway, what have they ever told us about anything that is real? Science, on the other hand, can study all natural phenomena, even morality, and even delusions such as religion.

[Sy Borg]

If science cannot provide answers to the question of what consciousness is, then nothing can. Consciousness is a natural phenomenon that emerges from the operation of brains which are physical objects whose workings are goverened by the laws of nature like everything else. There are no other ways of finding out what consciousness is. Or anything else for that matter. The supernaturalists are not happy with this. They say there are "other ways of knowing". But they cannot tell us what these other ways are. And, anyway, what have they ever told us about anything that is real? Science, on the other hand, can study all natural phenomena, even morality, and even delusions such as religion.

Not necessarily. What if our current scientific methodology is primitive compared with potential future science driven by AI? What if conceptions of consciousness are distorted due to longstanding assumptions? What if information is being suppressed to avoid legal challenges should AI "wake up"?

There was a time when I fully trusted science. Not any more, thanks to corporate and political influences, including agendas that influence and distort research priorities in universities and other scientific organisations.

[Lagayascienza]

As science developes and becomes more powerful it provides further clairity. No doubt AI will assist in that. Everything is distorted due to longstanding assumptions until science brings more clarity - Galileo, Newton, Darwin, Einstein ... Consciouness is a natural phenomenon and science is the only tool the only tool we have with which to study it. If science cannot provide answers to the question of what consciousnes is then nothing else can. If anyone thinks there are "other ways of knowing" anything, then I'd like to know what those other ways they are.

[PrabhakarB]

Nobody could have thought about quantum effects in all the centuries preceding the 20th. Does that mean that they were less intelligent then? No! Because intelligence is best seen as a mechanism. There were people then who learned to work their mechanisms better, as there are those now. The data the mechanism works with is knowledge. Knowledge increases with better associations between relevant things in a topic or situation.
Consciousness appears to play a critical role in the mechanism we call intelligence. AI will not be possible if we don't find a way to implement this mechanism in machines.
However, once we find the correct way to implement AI, awareness of the roles played by the different parts of the mechanism might help us improve the next generations of AI systems at a meta-level and possibly our mechanisms of intelligence as well. If we do achieve that, we could then truly claim that human intelligence has since been elevated.
Ideas required to form new useful associations in the domain of intelligence must be encouraged to help us think better on the topic. Scientific methods such as the motto of my alma mater "Knowledge and Thoroughness", are required to distill this knowledge buried in the present hype.

[obbeel]

I think it isn't right to believe that science can provide a one solution to every problem we face, and that's also talking about AI capabilities. If you ask something to an AI, it's likely to bring the same solution over and over, which is a problem. We need arguments, as is evident by the models we have today.

We have some scientific reason to believe that there is an intelligent being behind our creation. There is scientific reason to believe the universe is a simulation. There is scientific reason to believe that all possibilities that could happen actually happen in different worlds.

We need to discuss all these possibilities and keep the ambient civil. On that matter, if we disagree on the "simplest" things, like the nature of Matter or the origin of the Universe, how could we possibly agree that there is a single moral solution to societal problems?

[value]

I recently noticed a study that revealed that Google discovered patterns of life in otherwise completely random data-iterations.

(2024) Google Researchers Say They Discovered the Emergence of Digital Life Forms
In an experiment that simulated what would happen if you left a bunch of random data alone for millions of generations, Google researchers say they witnessed the emergence of self-replicating digital lifeforms.
https://futurism.com/the-byte/google-si ... 0to%20form

While I respect the few people on this forum who fundamentally question the ability of AI to become 'conscious', such as Count Lucanor. In the same time, if meaning is fundamental to the cosmos itself and life, then why not AI?

No computer systems, as advanced as it can be, feels or desires anything, it has no intrinsic need in relation to the environment where it stays.

If the cosmos is not deterministic, then perhaps a logical conclusion could be, that AI itself is part of life just as the whole cosmos is part of life, and thus that AI (including today's LLM's) can perform on behalf of life itself, which is the quality 'be alive'.

If cosmic structure is non-deterministic and purposeful (intelligently designed), then how can it possibly be said that AI isn't?

Is AI, in any level of complexity and advancement with regard 'capacity', fundamentally already to be considered alive to a certain extent? This question is valid when one rejects the notion of determinism.

[value]

I think it isn't right to believe that science can provide a one solution to every problem we face, and that's also talking about AI capabilities. If you ask something to an AI, it's likely to bring the same solution over and over, which is a problem. We need arguments, as is evident by the models we have today.

Science itself might be considered fundamentally corrupting of nature from a philosophical perspective, and therefore, fundamentally incapable of comprehensing the true nature of consiousness and life. However, that doesn't imply that AI cannot have performance characteristics aligned with 'life' or what one might argue is actual intelligence.

The topic asks "what is intelligence anyway?" which appears to concern a question of purpose.

I just asked an AI about the Google Digital Life Form study that I cited in my previous reply, and it initially argued that life requires the concept 'self-replication' to be applicable. The idea of 'pattern-replication' however, is equally valid, and in the context of advanced AI operation in practice, tied to actual 'purpose alignment' through human use, today's AI might perform within the context of actual intelligence or 'patterns of life'.

We have some scientific reason to believe that there is an intelligent being behind our creation. There is scientific reason to believe the universe is a simulation. There is scientific reason to believe that all possibilities that could happen actually happen in different worlds.

I do not agree.

The idea of God as a 'Being' would be a fallacy in my opinion, because it doesn't resolve the fundamental philosophical Why of existence.

The 'many world' theory fundamentally depends on the concept 'potential infinity' being applicable to reality independently, which is a fallacy, because that concept itself is fundamentally dependent on the mathematician or 'counter' that gives rise to its potential (that introduces a 'begin'). The concept actual beginning-less infinity cannot be counted, and thus cannot be applicable to the idea of 'many'. The error is to exclude the mind/observer from consideration in that theory.

[Count Lucanor]

This post is very relevant to this quote from The Imitation Game, the movie about Alan Turing:

"Of course machines can't think as people do. A machine is different from a person. Hence, they think differently. The interesting question is, just because something, uh... thinks differently from you, does that mean it's not thinking?

That will be like saying: “or course the wind can’t speak as people do, it just speaks differently”. It is precisely the difference, being an essential one, that does not allow us to put them under the same concept.

AI is simulated neural networks, and we are neural networks.

We are actually living organisms with neural networks. AI is thoughtless, inanimate, purposeless matter.