I think this is basically correct; intelligent systems by default have the property we call consciousness.
Stated loosely: there is going to be something complicated going on in their mind; and they accurately know that there’s something complicated going on. Thus, they will accurately report that it’s “like” something to be them.
There are a bunch of caveats, particularly “okay how intelligent/complex before they’re really conscious” which I think is a red herring.
That’s a very quick summary, and probably not well-stated, because I’ve never gotten around to writing up my thinking on this, even though I’ve been thinking about it alongside studying brain function since the turn of the century.
I apologize for dropping in my $.02 before deeply processing everything you’ve written; I wanted to leave a comment to boost visibility, even though I’m rushing on my own work right now.
I don’t think this is a necessary property, since you could create an intelligent system, then block its ability to think about itself or its own cognition (and therefore all introspection). That system would not be conscious in most senses of the word.
And specifying the sense in which you’re addressing consciousness is always crucial. The main point I make about consciousness is that a lot of the disagreement and confusion is coming from people using the same word to mean different things.
I subscribe to Dennett’s “heterophenomenological” thesis on approach: once you’ve explained why someone (or an AI) gives the reports they do about their own experience, you’ve explained all there is to explain. It’s very weird to claim that there’s something extra beyond the mechanisms causing you to say (and think) things about your experience/self.
And I think a pretty complete explanation is available for humans. Data on the nature of perceptual representations matches human reports quite nicely. Mechanisms for accurately (and inaccurately) introspecting are more theoretical/speculative, but I don’t think any strange/novel theories are needed; the same general principles of information processing and interpretation of sense data account for higher-order interpretations of the nature of one’s representations and how they impact the whole system—what we call qualia and experience.
Thank you for the engagement! I do think that blocking a system to think about itself or its own cognition would actually reduce its task performance drastically, as it can’t effectively plan. This is why I think the whole capability --> {world models, etc} direction in Arrow 1 is important, because world models enable you to plan, and they implicitly entail a self-model, because you need to model yourself to be able to be a good estimator of the transition kernel , where are the actions you as an agent take in an environment.
I think this is basically correct; intelligent systems by default have the property we call consciousness.
Stated loosely: there is going to be something complicated going on in their mind; and they accurately know that there’s something complicated going on. Thus, they will accurately report that it’s “like” something to be them.
There are a bunch of caveats, particularly “okay how intelligent/complex before they’re really conscious” which I think is a red herring.
That’s a very quick summary, and probably not well-stated, because I’ve never gotten around to writing up my thinking on this, even though I’ve been thinking about it alongside studying brain function since the turn of the century.
I apologize for dropping in my $.02 before deeply processing everything you’ve written; I wanted to leave a comment to boost visibility, even though I’m rushing on my own work right now.
I don’t think this is a necessary property, since you could create an intelligent system, then block its ability to think about itself or its own cognition (and therefore all introspection). That system would not be conscious in most senses of the word.
And specifying the sense in which you’re addressing consciousness is always crucial. The main point I make about consciousness is that a lot of the disagreement and confusion is coming from people using the same word to mean different things.
I subscribe to Dennett’s “heterophenomenological” thesis on approach: once you’ve explained why someone (or an AI) gives the reports they do about their own experience, you’ve explained all there is to explain. It’s very weird to claim that there’s something extra beyond the mechanisms causing you to say (and think) things about your experience/self.
And I think a pretty complete explanation is available for humans. Data on the nature of perceptual representations matches human reports quite nicely. Mechanisms for accurately (and inaccurately) introspecting are more theoretical/speculative, but I don’t think any strange/novel theories are needed; the same general principles of information processing and interpretation of sense data account for higher-order interpretations of the nature of one’s representations and how they impact the whole system—what we call qualia and experience.
Thank you for the engagement! I do think that blocking a system to think about itself or its own cognition would actually reduce its task performance drastically, as it can’t effectively plan. This is why I think the whole capability --> {world models, etc} direction in Arrow 1 is important, because world models enable you to plan, and they implicitly entail a self-model, because you need to model yourself to be able to be a good estimator of the transition kernel