Chalmers: "As for consciousness, I am open to the idea that a worm with 302 neurons is conscious, so I am open to the idea that GPT-3 with 175 billion parameters is conscious too."
I think, if some future system could get rid of the boundary between training and runtime, so that training happened continuously – then it would be closer to consciousness in my view. (It would also mean that different instances of the system would begin to diverge and become unique individuals, because even if the initial training was identical, the ongoing operation would be different, and the networks would diverge over time.)
Consciousness is what it feels like to have a thought, an emotion, an idea, in that exact moment while you are feeling it. It can happen in a brain at a moment, during a single second of time, and doesn't have anything whatsoever to do with learning, reacting to environments, or any of that other stuff the normally goes along with life.
Consciousness itself is the actual experience itself. Nothing more. Nothing less. This is why we think anything with a brain has some kind of consciousness. If anything it's likely more related to quantum mechanics and waves than it is to "information processing" which is the common misconception even AI experts have.
It must be either a side effect of intelligence or something that human intelligence uses to an end. Either consciousness is something composed of information processing, or it is something inherent to the universe that has some evolutionarily efficient use towards information processing. I favor the former.
I believe this very strongly. That said, the subject matter is a personal obsession, and I would love to hear counterpoints.
I touched on this in my previous comment, it is my belief that consciousness is not the only way that intelligence can be made, but that it is somehow efficient for the purposes of evolution. Using consciousness may consume the least energy (the brain uses a lot of energy), take the least genetic material to describe, have the safest learning curve (so that children are more intelligent and more likely to survive), or any combination of these and other features.
I think of experience as a sophisticated mathematical object with useful functionality. We have a disconnect with physical reality, and a strong connection with informational reality. I can assert that I exist, and the abstract model of my phone I keep in my head exists, but I can't assert that the phone exists and in reality its existence is very different from how I perceive it. It certainly seems like I am an information construct that was formed within a physical reality.
Beyond that I'm mostly in the dark though. You can see that consciousness is involved in learning and adapting- you are highly conscious of new skills and change, but old skills sink into the subconscious and you gradually ignore repeated stimulus. You can see that consciousness integrates much of our intelligent functionality (perception, memory, executive function) and you can feel that your role is to run things. How is experience related to all of this? I do not know.
Sometimes I try to imagine the later case, and it really flips reality on its head. The limit and most extreme case is that reality is fundamentally experiential -- that is, what comes first is "being", "feeling", "embodiment", and through this lens is found structure, objects, form, etc. Obviously this is just the reverse of the idea that consciousness emerges from an underlying physical substrate performing complex processes.
Either way, there is a definite correlation between the two -- feelings have their correlate molecular, biochemical basis, and molecules working together through processes have their transcendent embodiment as feelings experienced.
The question of "what is real?" can boil down to this: are things external to consciousness fundamentally real and consciousness an ephemeral, emergent flourish floating "on top", or is consciousness real and everything observed by it a kind of flourishing of it?
This is a bit of a rabbit hole with many different paths to fall down, as I'm sure you know. Scientific knowledge is rooted in observation and the dusting away of uncertainty to reveal an objective reality we all share. From this standpoint, the objective substrate being revealed and it's complex processes is taken as fundamental, and we have all the great successes of scientific knowledge to show as justification for this to be true. The only hole seems to be, why the hell am I embodied, then? -- why am I conscious at all? Life would probably be easier if I didn't see that hole and want to search for more satisfying answers!
What is real?
Consciousness self-asserts: (1a) 'I think, therefore I am' (or else, 'thought is occurring, so thought must exist'). If you accept the reasoning there, you can also grapple in (1b) 'I see blue, therefore blue exists', etc.
In that sense, our consciousness is a rare example of something that definitively exists. A statement like 'there is a rock in space called Earth' would be false if we lived in a computer simulation. The correct statement becomes 'there are a bunch of numbers representing a rock in space called Earth, in this computer'. Consciousness doesn't answer to the abstraction in the same way. 'I see a rock in space that I think of as Earth' is true regardless of whether you're inside of the simulation.
We can also assert that reality exists, as far as (2) 'there is a thing that my experience interacts with which I do not consciously control and which exhibits complex behavior', and also, (3) 'I exist (per 1a), therefore I am somewhere. I can perform computations, therefore the place I am in must allow for computations to occur. I have experience (per 1b) therefore there I am somewhere in which experience can exist. Reality exists (per 2) therefore there must be something sophisticated enough to produce it.'
But that's strictly an informational definition, again equally true whether or not you're in your own dream- it only addresses the complexity of the mind producing the dream.
So to conclude: information is quintessentially real. Our consciousness and reality are real at least to the extents that they are information, which are 'very much so' and 'a lot, maybe more', respectively. Physical reality as we know it might be real, Occam's Razor says 'probably', Simulation Hypothesis says 'probably not'. Anyone's game. I think that a physical reality of some form must exist in order to perform computations and produce information, but I'm open to a rebuttal.
And then why the hell am I conscious? This seems to be the crux of the matter. It is my opinion that the answer is of the form 'consciousness solves problem X efficiently along dimensions Y and Z' where X is some fundamental component of intelligence, and Y and Z are environmental constraints. I think it's unlikely that the answer is related to the fundamental makeup of the universe. Evolution follows the path of least resistance, and entangling our minds with some innate property of quanta, from the scale of proteins seems more challenging than other conceivable non-conscious solutions to general intelligence.
I definitely follow you up to your last paragraph and it all rings true to me, however I don't quite understand, "It is my opinion that the answer is of the form 'consciousness solves problem X efficiently along dimensions Y and Z' where X is some fundamental component of intelligence, and Y and Z are environmental constraints." Maybe the rest of what I have to say is just because I don't understand the fundamental component or constrains very well.
To me mathematics is the limit of description. I can assign a word to some observable thing and distinguish it from all other observable things. I can draw a picture of it to distinguish it even more precisely. I can use various mathematical techniques to describe it even better, perhaps even to arbitrary degrees of precision. But I fail to see how any mathematical technique can capture --the feeling of-- happiness, pain, etc.. These embodiments can not be fully realized by description alone. They can be pointed to, hinted at, and I think great artists can stir echos of them in other people, but actually experiencing them is beyond the capacity of description. That's why I wonder if experience/consciousness is something fundamental. A subsequent worldview would have as its central concern 'beings' instead of 'objects'; it would not exclude any current or future science, it would just shift it's focus away from abstractions and toward experiential beings -- with conscious beings, which we are, perhaps a special case of a much larger set. The gains would not be material, but perhaps there would be some improvements in the ways we interact with ourselves, each other, and our surroundings.
There are two criteria I'm addressing here. Consciousness is either physical (produced in the universe) or informational (produced in the mind). Consciousness is either important to intelligence or incidental to intelligence. My position, which I'll justify below, is informational/important. If you accept that consciousness is manufactured in the mind and important to intelligence, that means we evolved it. Because it is a widespread evolved trait, it very probably is an effective solution to a problem against environmental constraints, towards the larger goal of reproduction.
Constraints might include the amount of genetic data needed to produce a useful output, how well it deals with failure cases, how well it responds to genetic mutations or how well it withstands viruses or cancer. The kind of stuff that is irrelevant from the perspective of an intelligent designer like us with access to basically limitless indestructable computational resources.
Physical/important I responded to previously, but briefly: the big issue is scale. Humans run on proteins and large organic molecules. If there was something nonmathematical at that size and in our bodies, we would very probably know about it by now.
Both informational/ and physical/ irrelevant are 'side effect' models. They have at least two flaws. Consciousness follows attention, not brain activity. If I do something subconsciously, I am engaging the same neurons but not producing the same side effects. Consciousness is not a disconnected afterimage of intelligence because I am aware of it and can perform reason on it. It affects and is affected by my brain. If it's a side effect, it's one that has been knitted into me, presumably to some benefit.
So what does that make consciousness? Taking it as an informational tool to some end, we can probe some interesting questions. Self-assertion, which I referred to earlier, is an interesting mathematical property. A set of rules that allow the system within them to prove its own existence? And it's a global property across all conscious experience, that's certainly of note. The benefit of consciousness seems to be related to awareness of self and environment (that's all experience seems to be) as well as executive function- we experience a sense of free will, presumably because evolution wants us to help run things from here. There's a remote possibility that free will is real, and consciousness is somehow an non-deterministic process. That and beyond are all speculation, though.
The belief system you describe is how I got out of nihilism and escaped what was an agonizing conflict between romanticism and realism (I like the song Imitosis by Andrew Bird for depicting that conflict). There's a cold, meaningless reality out there, but somehow there's meaning that is made of it. We matter even though (or because) if we didn't, nothing would.
I agree with most of what you guys are saying. Here's a wiki post I wrote, outlining a kind of theory for what consciousness actually is, that attempts to explain some of the 'mechanics' of it, or a description of what memory itself actually is.
So intelligence uses memories as it's building blocks for recombining and recognizing patterns, but as you'll see in a nearby reply in this comment thread by me, I have a theory about memory which is that it's not "stored" but your brain accesses essentially the 'actual' event thru spacetime, and that the 'accessing' of these past events and merging those 'wave-forms' and entangling them with the present brain state, is what we call consciousness, regardless of whether any intelligent processing is happening.
Some philosophers (of a pan-psychic bent) would say that consciousness doesn't require any intelligence, so even something completely unintelligent (like a pebble, or an individual proton) could be conscious. Others think that consciousness requires some minimal degree of intelligence, a standard which (non-human) animals may or may not meet, but a pebble certainly can't. We don't know who is right here. We don't have any agreed upon objective standards to determine what is conscious and what is not.
But, if it were true that some minimal degree of intelligence is required for consciousness, then it may well be that primitive animals have that degree of intelligence yet GPT-3 lacks it. While GPT-3 can perform at seemingly human level on some tasks, there are other tasks on which even quite primitive animal intelligences vastly outperform it. Maybe, if intelligence is necessary for consciousness, the kind of intelligence underlying the later tasks may be more essential to consciousness than the kind underlying the former.
> If anything it's likely more related to quantum mechanics and waves than it is to "information processing" which is the common misconception even AI experts have.
The idea that consciousness is some kind of special quantum phenomena is highly speculative. Sure, some philosophers and physicists think it may be true, but others think the whole idea is baloney. When you say "likely", that's just your own opinion of what is more likely, there is no hard evidence to support that probability judgement.
You can invent new thoughts and ideas, but they are always built from existing ones, as their building blocks. I think the answer to the question 'where/how' is memory 'stored' in the brain is: "It's not". I believe the brain is quantum mechanically connected to all prior states of itself (like all matter is), and so what we call 'memory' is actually a 'direct connection' thru spacetime to the actual event.
Needless to say it would take a book to describe this theory in detail, so maybe I'll write up all my thoughts at some point, but it explains lots of mysteries of intelligences once you accept this interconnection model. Everything from savantism, to instinct behaviors, to fungal intelligences falls into place.
Once you accept that all complex patterns in nature that 'evolve' are always still 'connected directly to' all prior copies of themselves, and able to exchange wave potentials, it makes many things that used to seem 'paranormal' or 'magical' suddenly have a more scientific explanation.
And that's a fundamental difference between conscious biological life and GPT-3. Conscious biological life experiences a two-way interaction with its environment, in which organism and environment act on each other simultaneously. GPT-3's experience of that is very limited. It has experienced the environment act on it (training), and it has experienced itself act on the environment (runtime), but those two experiences are largely siloed off from each other. (It effectively does have some runtime memory, so to a very limited degree it can dynamically react to the environment, but it can't actually learn anything at runtime.)
Now is that experience, which humans and animals have, but which GPT-3 lacks, essential to consciousness? Who really knows. The fact is, we don't really know what consciousness is, or what are the conditions for its existence. Maybe at least some history of that kind of two-way interaction is essential for consciousness, in which case GPT-3 can't have it (but some future successor system might). Maybe not. Nobody really knows.
Unless of course we find out what that is and engineer a machine to have it. But it might end up being something deep within us, like particle spin within our DNA, for example.
You should consider familiarising yourself with the ideas before dismissing then as silly.
Forgive me for not wasting time familiarising myself with his made up distinction between consciousness and cognition.
There are basic medical tests, of course, but maybe they're faking it. You know, as a person, what it feels like to be a person; you feel when you're conscious. Is this feeling universal, or are there people who just don't have it?
This question is known as the question of philosophical zombies [0], or p-zombies, and it is worth taking seriously, if for no other reason than that GPT-3 and friends are very much like p-zombies; they sound very cogent and coherent but are definitely not conscious in the same brain-based way that humans are conscious.
Isn't it enough that we can perceive consciousness or its absence in ourselves and others? Isn't it enough to define it (my words, not very carefully thought out) as something that at least entails the capability to think about a situation, real or hypothetical, including your own state and thoughts and communications, understanding some of the consequences of action in such a state, choosing an action, accepting responsibility, etc., up to philosophy, meta-cognition and beyond?
The fact that it's hard to define, doesn't mean it doesn't exist.