Just wanted to add something, to bring it back to feasibility of embodied/unembodied approaches. Using the definition of embodiment I described, it needs to be said that it is impossible to specify the goals of the agent, because in so doing, you'd be passing it information in an unembodied way. In other words, a fully-embodied agent must completely structure internally (self-organize) its model of the world, such as it is. Goals must be structured as well. Evolutionary approaches are the only means at our disposal for shaping the goal systems of fully-embodied agents, by providing in-built biases towards modeling the world in a way that is in alignment with our goals. That said, Friendly AI is impossible to guarantee for fully-embodied agents.
The question then becomes, is it necessary to implement full embodiment, in the sense I have described, to arrive at AGI. I think most in this forum will say that it's not. Most here say that embodiment (at least partial embodiment) would be useful but not necessary. OpenCog involves a partially embodied approach, for example, which I suppose is an attempt to get the best of both worlds - the experiential aspect of embodied senses combined with the precise specification of goals and knowledge, not to mention additional components that aim to provide things like natural language processing. The part I have difficulty understanding is how a system like OpenCog could hope to marry the information from each domain - the self-organized, emergent domain of embodied knowledge, and the externally-organized, given domain of specified knowledge. These two domains must necessarily involve different knowledge representations, since one emerges (self-organizes) at runtime. How does the cognitive architecture that processes the specified goals and knowledge dovetail with the constructions that emerge from the embodied senses? Ben, any thoughts on that? Terren --- On Sat, 8/23/08, Terren Suydam <[EMAIL PROTECTED]> wrote: > Yeah, that's where the misunderstanding is... "low > level input" is too fuzzy a concept. > > I don't know if this is the accepted mainstream > definition of embodiment, but this is how I see it. The > thing that distinguishes an embodied agent from an > unembodied one is whether the agent is given pre-structured > input - that is, whether information outside the agent is > directly available to the agent. A fully embodied agent does > not have any access at all to its environment. It only has > access to the outputs of its sensory apparatus. > > Obviously animal nervous systems are the inspiration here. > For example, we have thermo-receptors in our skin that fire > at different rates depending on the temperature. The > interesting thing to note is that these receptors can be > stimulated by things other than temperature, like the > capsaicin in hot peppers. The reason that's important is > because our experience of hotness is present only to the > extent that our thermo-receptors fire, without regard to how > they're stimulated. Likewise for the patterns we see > when we rub our eyes for long enough - we're using > physical pressure to stimulate photo-receptors. > > What all that reveals is that there is a boundary between > the environment and the agent, and at that boundary, > information does not cross. The interaction between the > environment and sensory apparatus results in *perturbations* > in the agent. The agent constructs its models based solely > on the perturbations on its sensory apparatus. It > doesn't know what the environment is and in fact has no > access to it whatsoever. > > This is a key idea behind autopoiesis > (http://en.wikipedia.org/wiki/Autopoiesis), which is a way > to characterize the difference between living and non-living > systems. > > So all text-based I/O fails this test of embodiment because > the agent is not structuring the input. That modality is > based on the premise that you can directly import knowledge > into the agent, and that is an unembodied approach. > > Terren > > --- On Fri, 8/22/08, Vladimir Nesov > <[EMAIL PROTECTED]> wrote: > > > On Fri, Aug 22, 2008 at 5:49 PM, Vladimir Nesov > > <[EMAIL PROTECTED]> wrote: > > > On Fri, Aug 22, 2008 at 5:35 PM, Terren Suydam > > <[EMAIL PROTECTED]> wrote: > > >> > > >> She's not asking about the kind of > embodiment, > > she's asking what's the use of a non-embodied > AGI. > > Your quotation, dealing as it does with low-level > input, is > > about embodied AGI. > > >> > > > > > > I believe "non-embodied" meant to refer > to > > I/O fundamentally different > > > from our own (especially considering a context of > > previous message in > > > this conversation). What is a non-embodied AGI? > AGI > > that doesn't > > > exist? > > > > > > > On second thought, maybe the term "low-level > > input" was confusing. I > > include things like text-only terminal or 3D vector > > graphics input or > > Internet connection or whatever other kind of > interaction > > with the > > world in this concept. Low-level is relative to a > model in > > the mind, > > it is a point where non-mind environment directly > interacts > > with the > > model, on which additional levels of representation > are > > grown within > > the mind, making that transition point the lowest > level. I > > didn't mean > > to imply that input needs to be something like a noisy > > video stream or > > sense of touch (although I suspect it'll be > helpful > > developmentally). > > > > -- > > Vladimir Nesov > > [EMAIL PROTECTED] > > http://causalityrelay.wordpress.com/ > > > > > > ------------------------------------------- > > agi > > Archives: > https://www.listbox.com/member/archive/303/=now > > RSS Feed: > https://www.listbox.com/member/archive/rss/303/ > > Modify Your Subscription: > > https://www.listbox.com/member/?&; > > Powered by Listbox: http://www.listbox.com > > > > > > ------------------------------------------- > agi > Archives: https://www.listbox.com/member/archive/303/=now > RSS Feed: https://www.listbox.com/member/archive/rss/303/ > Modify Your Subscription: > https://www.listbox.com/member/?&; > Powered by Listbox: http://www.listbox.com ------------------------------------------- agi Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/ Modify Your Subscription: https://www.listbox.com/member/?member_id=8660244&id_secret=111637683-c8fa51 Powered by Listbox: http://www.listbox.com
