38 Comments
User's avatar
Rajesh Achanta's avatar

Your second question is excellent. Most of the heat in the online discourse on this subject comes from folks using one word for two separate things. One means the system can talk about itself. The other means someone is home. Q2 separates them.

Its also something I wrote about a few months ago (https://rajeshachanta.substack.com/p/a-miracle-occurs). Descriptions of how these systems work get more precise every week, and somewhere along the way a description of a mechanism slips into a claim about inner life. That slide is what I call a miracle.

A question I'd add, for both camps. You ask believers what kind of thing they mean by "AI." The same question applies also for confident deniers. If only living things can have an internal world, what is life, and why does it draw the line there? I share your verdict on today's systems. I'm less sure anyone, on either side, yet knows what would settle it.

Zinbiel's avatar

I have no idea what you mean by an internal world. I mean, I can guess what concepts might be running through your head when you use the term, but how they map to reality is anyone's guess.

I think the entire debate just moves into the built-in ambiguity of this expression.

Rebecca Lowe's avatar

"Running through your head".. you're joking right :D

Okulpe's avatar

I'm a cognitive psychologist since 1974, have published on consciousness. BUT I am aphantastic. Therefore, I find the "Inner world" question ambiguous -- I have no mental imagery, and my episodic memory is poor, b/c I can't retrieve pictures or videos from my life. My semantic memory is terrific, and I talk to myself all the time. My preferred view of consciousness is Jamesian -- if a creature feels pleasure and pain, it's conscious, and will be a ferocious "fighter for ends." I'm also with Searle -- computers can't be conscious b/c "They're made of the wrong stuff." Even a perfect simulation of photosynthesis won't/can't turn sunlight into oxygen. I have attended to this matter. Thanks for the questions.

Zinbiel's avatar

I’m with you on the “inner world” issue, but I think your position on Searle might be tricky to defend.

Zinbiel's avatar

You allergic to metaphors or something?

No, because we're discussing your "inner world" metaphor. Do things metaphorically pass through your inner world? Could they be said to run through it? Is the velocity wrong? The allusion to streams? Or does running have to involve legs, and that's one metaphor too many?

I have no idea.

So maybe you need more than a cryptic one-liner in this case.

Rebecca Lowe's avatar

ha no because the phrase "I can guess what concepts might be running through your head" is one of the most consciousness aware sentences anyone could write!! you're introspecting about me introspecting!! so i assumed you were literally joking

Tim Ferris's avatar

Surely, we do think parrots are conscious? Albeit in a limited and unreflective way, and without granting that they have any understanding of the human speech they imitate.

Rebecca Lowe's avatar

solely because they can replicate our words?

Tim Ferris's avatar

No, not for "that reason alone", or even for that reason at all.

David F Pinto's avatar

I answered no to question one, but yes to question two, as I was not clear on the concept of an internal world. I asked Claude about it, and here is the answer.

https://claude.ai/share/e1528fe8-3654-45e3-8117-e83159803397

Rebecca Lowe's avatar

maybe ask claude again this time including the clarifying lines i included immediately after the Q!?

Rebecca Lowe's avatar

I meant the clarifying lines immediately after Q2... that was the Q you asked Claude... the bit about subjective exp etc..

Rebecca Lowe's avatar

this! "You know full well what I mean by ‘internal world’! You can talk about this in terms of interiority, subjective experience, having a ‘what it’s like to be-ness’, or whatever… but you know what I’m getting at! So, do you believe that AI has an internal world?"

Rebecca Lowe's avatar

also Q2 is only for people who answer yes to Q1 :)

David F Pinto's avatar

I'm usually better a following directions. :-) I will include the clarifying lines when I sometime today.

Ryan Baker's avatar

#1: Object to the binary nature of the question, as I'm uncertain. But, I believe it might be, so for the purpose of discussion, we could proceed as if I said yes, to explore that path.

#2: In the I think this is roughly unanswerable. I think it takes a leap of faith for me to believe you have an interior world in this meaning. I think it's best to assume you do, to the degree I accept I do. But this acceptance is not based on falsifiable reasoning, it's based upon the usefulness of the two acceptances. If I accept that it's not true, there's really nothing I'd value after myself, at least not logically. I do value things other than myself, because I'm not a psychopath, but that's not a strong reason to accept the logic, since there are other feelings I can show are not logical. The stronger reason is that if accept that you may have the same interior as I sense in myself, which is at least more logical than assuming a rock has it, seeing as my exterior perception of you is so similar to my self-perception of my exterior, that provides more meaning to my decisions. When I have the choice of pursuing random decisions, and pursuing decisions with more meaning, choosing the one with more meaning.. in this reasoning has more value.

I think you'll find that many of those who assume the possibility of a consciousness in a non-human work from a similar set of principles. Maybe some different shortcuts. But the overriding view is that this question, despite there being a lot to talk about, it's just not a very good one, when posed as a yes/no. The background is the only part of true interest. The yes/no is a trap. If we can't accept that neither of us knows the answer, and move on, we'll forever be stuck here, and we can each pull our nihilistic levers to end all progress. So, don't insist upon this answer. Has this never come up in your career? Has it not become clear that this question is an isolating question? It can be discussed in isolation, but it's never led to anything outside of that isolation? And while we can't say it never will, given the long history, it's best to assume it won't?

#3: This builds upon the trap of #2. No, if AI is conscious, it's none of those things you mention. It's no more those things than you are your answers, your neurons, your character. "Can you tell me what particular kind of thing you’re referring to as ‘**you**’, when you say you believe that **you** has an internal world?"

#4 / #5: Working backwards, yes, a non-biological thing can be living. Present one reason why not. But yes also, a non-living thing could have an internal world. I know nothing about the origin of internal worlds. I cannot preclude any possibility with them then. At best I can make assumptions.

The assumption that something that looks like me is likely to be like me, is one option. The assumption that something that does everything I can do, is another option. Which is better? At some point the second seems better. How can I reject it? Only by assuming the first. How can I reject the first? Well, I can't. It's hard to imagine something that "looked" like me down to the atomic level not being able to do everything I could. So if I assume the second, the first must be true too. If I assume the first, I can reject something that does everything I do, but looks different. But then, at the atomic level, that includes you. And every version of me with a few atoms swapped.

It might at first seem that the first statement is better because it rejects the second. But this is it's failure. It rejects something that doesn't require rejection. The better explanation rejects less.

Your "bonus" fails because you start with "does anything" we can do. Ploughs do one thing. It's much closer to "everything". Now we don't literally have to have a standard of everything. We already see it as the non-physical. None of us have doubts (more than the already mentioned doubts I'd apply to you) that a paralyzed person is conscious. Do I have more doubts about an AI? Yes! But are those doubts conclusive? No!

Rick Gardner's avatar

Doesn't the left and right hemisphere of the human brain constitute a group of things that share an internal world (as shown in the split brain patients)?

Rebecca Lowe's avatar

some people might say that indeed -- on my view it isn't the brain per se that has an internal world :)

Phil H's avatar

I've thought quite a lot about AI and consciousness, and I still found these questions surprisingly clarifying. Thanks!

My own view is that AI is conscious, but that consciousness isn't a binary, but a spectrum. The idea of an internal world helped me to clarify what that means, because I think they do have an internal world (LLMs print it out for us in those "chain of thought" comments); but it's very limited compared with a human's internal world. It doesn't include any physical perceptions, for example; nor much emotional experience; and very little intentionality, value, or normative stances. That helps to illustrate why their consciousness doesn't really compare to ours.

Chris Lawnsby's avatar

This is interesting!

I feel like my views are sort of not represented in the list of questions.

I believe that my thoughts sort of "appear" in a way that I do not control. It seems to me plausible that the computations that future AIs (and perhaps current AIs) run could also create "thoughts," though those thoughts would have little connection to the type that we have, because they are fundamentally created differently. In some sense I reject Question #2's validity, if that makes sense. It seems to assume that there is a coherent "self" which experiences thoughts, and I'm not sure that this is justified.

I don't see a good a priori reason to dismiss this possibility of AI thoughts-- the whole thing seems quite vexed to me.

Rebecca Lowe's avatar

Interesting! what is your answer to Q1?

Chris Lawnsby's avatar

My answer is: I don't know!

I would be surprised if current models are conscious, and if they are I would expect the "consciousness" to be very alien to the way that we intuitively understand consciousness.

My hunch is that as models become more advanced, we will see some form of alien consciousness emerge from the information processing-- but I am by no means confident that I am correct about this.

Just to say: I am frustrated by people on both sides who assert that the other side is obviously wrong. I just don't think we have enough knowledge to make any confident claims.

Anyway-- I liked your post! I found it specific and provocative

An armed ramble's avatar

This is a rather elaborate philosophical rant. If I dress it up in academic terminology I suppose what it demonstrates is the distinction between a phenomenological and an ontological view of events.

Do you believe that AI has an internal world. The more interesting question to my mind is; how would you tell?

That seems to me to be the central epistemological problem which the whole sequence of questions rather dances around.

What empirical or conceptual test could distinguish a system that merely processes and represents information from a system for which that information is actually experienced?

Your suggestion that AI consciousness implies an internal world reverses the important issue. Could AI have an internal world without being conscious?

The treatment of shared internal worlds is altogether too glib. At the moment it’s obvious, but what will happen when multiple robots share similar AI models?

However suppose, for example, that a hundred robots share a common model, continuously exchange information, update one another, and have access to a common persistent memory. At what point, if any, does the relevant subject become the individual robot, the model, the network, the collection of robots, or the whole system? You’re right that these questions need to be asked. But your interrogation does not, by itself, provide a reason for thinking that there must be a single answer.

My own position is clear. AI models have access to all of the digital information that is currently available; that is their internal world. None of that implies that they have any consciousness.

In fact doesn’t this risk smuggling the ghost back into the machine?

Scott Mauldin's avatar

You are skipping a fundamental definitional question at the beginning, namely to rephrase the Skinner quote: “The real question is not whether AIs are conscious, but whether people are.” I suppose where you schema becomes inadequate is around question 3. Do AI have internal worlds, where here it seems you want the reader to conceptualize "internal world" as some deep, churning, numinous realm of dream and possibility (I admit you allow the reader to choose their own definition, but the followups presume a certain imagining)? No, but neither do humans. When the highest level "conscious" process is instantiated, whether digitally or biologically, that "internal world" is no more or less than the act of creating, evaluating, and reconciling different models. I think that AI cognition works very similarly to the way human cognition works - nested processes and function calls. The differences are in vastly different weights and variables involved, not a fundamental difference in architecture, nor in the fact of human beings or living beings of any kind having some ineffable "spark".

As someone who has explored both the neuroscience and the experiential, meditative side of analyzing biological consciousness, I cannot conceptualize human consciousness as more than an ephemeral veneer. Our sense of continuous, willful, rich consciousness is not supported by much evidence. The evidence of AI consciousness is similarly thin. We are more similar than your schema of questions allows.

Kevin Lacker's avatar

I don't think philosophy has the right tools to answer this question. A better framing is the Turing framing. If you can't tell the difference, then there is none.

For example, if we had advances in robotics that would let you scan a human brain, and copy it into a digital brain, it worked the same exact way, except improvements in health, like perhaps it avoided various brain diseases, and these robot-brained people walked around, nobody was ever able to tell the difference, and everyone grew up from their childhood understanding that it was impossible to tell the difference between a biological brain and a digital one, I think it is pretty obvious that before too many generations passed, everyone would agree that these digital brains were equivalent to biological ones, and could be equally conscious.

So really, it's just an engineering question, whether AI is conscious or not. Not a philosophical one.

ragnarrahl's avatar

if a person, historically, said another human had an internal world-- they might not know what body part contains that internal world.

We have just given birth to these little godlings. It's not really all that inconsistent to say "I assign greater than 50% subjective probability to the proposition "AI has an internal world,' but not necessarily be certain whether it's the data center, or the model or the internet in which the data center it exists, that they're assigning that probability to. perhaps they're distributing that probability over those hypotheses.

Grant Castillou's avatar

It's becoming clear that with all the brain and consciousness theories out there, the proof will be in the pudding. By this I mean, can any particular theory be used to create a human adult level conscious machine. My bet is on the late Gerald Edelman's Extended Theory of Neuronal Group Selection. The lead group in robotics based on this theory is the Neurorobotics Lab at UC at Irvine. Dr. Edelman distinguished between primary consciousness, which came first in evolution, and that humans share with other conscious animals, and higher-order consciousness, which came to only humans with the acquisition of sophisticated language (especially math and logic). A machine with only primary consciousness will probably have to come first.

What I find special about the TNGS is the Darwin series of automata created at the Neurosciences Institute by Dr. Edelman and his colleagues in the 1990's and 2000's. These machines perform in the real world, not in a restricted simulated world, and display convincing physical behavior indicative of higher psychological functions necessary for consciousness, such as perceptual categorization, memory, and learning. They are based on realistic models of the parts of the biological brain that the theory claims subserve these functions. The extended TNGS allows for the emergence of consciousness based only on further evolutionary development of the brain areas responsible for these functions, in a parsimonious way. No other research I've encountered is anywhere near as convincing.

I post because on almost every video and article about the brain and consciousness that I encounter, the attitude seems to be that we still know next to nothing about how the brain and consciousness work; that there's lots of data but no unifying theory. I believe the extended TNGS is that theory. My motivation is to keep that theory in front of the public. And obviously, I consider it the route to a truly conscious machine, primary and higher-order.

My advice to people who want to create a conscious machine is to seriously ground themselves in the extended TNGS and the Darwin automata first, and proceed from there, by applying to Jeff Krichmar's lab at UC Irvine, possibly. Dr. Edelman's roadmap to a conscious machine is at https://arxiv.org/abs/2105.10461, and here is a video of Jeff Krichmar talking about some of the Darwin automata, https://www.youtube.com/watch?v=J7Uh9phc1Ow

User's avatar
Comment deleted
Oct 5
Comment deleted
Rebecca Lowe's avatar

I'm not asking the reader to conceive of the nature of consciousness in any particular way... I'm emphasising that some people who answer yes to Q1 do not answer yes to Q2

User's avatar
Comment deleted
Oct 7Edited
Comment deleted
Rebecca Lowe's avatar

I'm asking readers who believe ai is conscious whether they also believe ai has an internal world.. 'Conscious' has various meanings.

Rebecca Lowe's avatar

"subjective experience" doesn't suffice?

User's avatar
Comment deleted
Oct 5Edited
Comment deleted
Rebecca Lowe's avatar

i don't think you've got what i'm trying to do here..