On Wed, Feb 12, 2014 at 12:37 AM, omd <[email protected]> wrote: > On Wed, Feb 12, 2014 at 1:00 AM, Nicholas Evans <[email protected]> wrote: > > [An attempt at very simply tightening the restraints against attempting > to > > register things that cannot utilize language to communicate their own > ideas. > > While originating and communicating imply intentionality, they're hard to > > disprove when given a sufficiently clever sentence generator. Rephrasing > > implies understanding of the semantics of a language, and recursion > implies > > a cognitive awareness of a language. Combined, these are unlikely to be > > spoofed.] > > I'm somewhat skeptical of the word "organism". > > If it has to be biological, then just say "human", there's no > guarantee hypothetical space aliens or AIs would be considered > organisms anyway. :) > > If not, then you have to deal with the issue that personhood is > currently an all or nothing passive affair, while the aforementioned > hypothetical entities might not have such rigid boundaries. For > example, while it is reasonable (though perhaps unkind) to require Deb > and Bob to exist as separate players, because humans tend to be > capable of operating with distinct personalities, Vernor Vinge's Tines > might not appreciate the sentiment despite having some ability to be > split up. This could be mostly solved by only requiring that players > be disjoint, but that would require putting restrictions on > registration rather than merely tweaking the definition of a person. > (Though for an AI which could fork itself on a whim, the concept of > "disjoint" might itself be murky.) > > Finally, I think that the new parts of the proposed wording are not > actually that hard for an actual AI to satisfy. After all, automated > rephrasing would be a similar problem to machine translation - not > 100% accurate but able to produce readable outputs for most inputs. > (Well, it can be accomplished in a roundabout way using the latter - > do what translationparty.com does for a language easier to translate > than Japanese and there's your rephrasing.) "Recursively referring > to" is harder to pin down, but probably doable to some extent. > > /me wonders whether speculating about Tines is too silly for the New > Agora. Or is that the Old Agora? >
I agree that AI can imitate rephrasing, to an extremely limited extent. I also agree that they can imitate recursion to an even more limited extent. In fact, if it can rephrase then you could give it the set expression: "<Phrase A> is identical in meaning to <Rephrasing of A>." I highly doubt the AIs ability to reliably rephrase that. A human could do so, then recursively refer to both versions, and continue to go down this rabbit hole to the limit of their working memory, and even moreso if they took the time to write every version down. Additionally, http://translationparty.com/#11235473: "I am a green man" is equivalent in meaning to "I am both green and a man." 'I'm green man' is ' I'm green and people ' is semantically the same. ' ''Is green man I green and others ', is the same in meaning. ' ' The green man I green and others ', is the same in meaning. If this isn't satisfactory evidence, I could add more constraints. AI and Linguistics is my area of focus.
