You can infer a regular course of action for constructing those parts of the system that are indeed patterned, yes. But a representational approach leaves room for incomplete representation, whereas an algorithmic approach does not.
In other words, the inability of a Turing machine to determine whether an arbitrary algorithm will ever stop without running that algorithm forever does not apply to representational systems because we don't insist on applying the strict ordering of operations to representations. We can add to a representation in any order we see fit, and work with partial representations as long as they cover what we need them to cover. And so we don't have this ridiculous issue with stopping conditions that formal algorithms have, because we drop the completeness constraint that isn't vital to the functioning of intelligence. As for Copycat, you are making the argument that because the aproach used hasn't worked, it won't. Aside from that argument being fallacious, I pointed out that software as an example of the different paradigm I've been trying to explain, not as proof in hand. On Aug 23, 2012 3:36 PM, Mike Tintner <[email protected]> wrote: Explain how Copycat can be applied to diverse patterns [it’s a terribly unimaginative program that in no way solved the problem Hofstadter set himself]. Patterns *are* essentially algorithms.[This is one area where Ben and I may well agree – though he may want to be cussed tonight]. If you have a regular relationship of parts – as in a pattern – then you can immediately infer a regular course of action – a regular way of physically combining those parts and constructing that pattern,, i.e. an algorithm.Conversely, an algorithm is a way of constructing a pattern, i.e. set of variations on a pattern. From: [email protected] Sent: Thursday, August 23, 2012 9:22 PM To: AGI Subject: Re: [agi] Boris Explains His Theory You should read up on Douglas Hofstadter's Copycat software, along with his other related projects. While it may be true that there are fundamentally different kinds of patterns that can't be generalized, some approaches to manipulating those patterns *can* be generalized. And there's a key difference between an algorithm that produces all algorithms vs. a pattern that encompasses all patterns. Patterns are *representational*, not algorithmic. This means that they need to be complete to be useful, and it is permissible to leave some parts unrepresented or under-represented, unlike with algorithms. This is core to how the human mind processes things. It's why we can see the truth of the self-referential statements used in Gödel's incompleteness theorems, but no algorithm can *prove* their truth. We function on representational principles, not algorithmic ones, and the algorithms we implement merely allow us to extend those representations within certain subdomains of reliable deducibility. On Aug 23, 2012 2:56 PM, Mike Tintner <[email protected]> wrote: Ben: that's easy, these [[all patterns]] are all obviously susceptible to lossy compression using algorithms native to the brain... Total shameless waffle. You haven’t the slightest idea of what you’re talking about, any more than when you claimed a program could produce all Da Vinci’s paintings. Explain how a single algorithm can produce the first three patterns AND then any example of a cellular automaton AND then any patterns which will be produced for your algorithm AFTER you have defined it. If you have a concept of “pattern”, that’s what you must be able to do – embrace not only known patterns but also all patterns yet to be created. How IOW are you proposing that a single algorithm can identify/produce *any* kind of elements in *any* kind of regular relationships? And while you’re at it, you might as well explain how a single algorithm can identify/produce **any and all** algorithms – because that is essentially the same claim you are making. If there’s a pattern for all patterns, there’s an algorithm for all algorithms. And there’s a formula for all formulae. Total cobblers. Reality: patterns, algorithms and formulae are specialist through and through, with no AGI powers of generalization whatsoever. From: Ben Goertzel Sent: Thursday, August 23, 2012 1:37 PM To: AGI Subject: Re: [agi] Boris Explains His Theory If you want to put that mathematically, take a whole set of diverse patterns – Koch curve, Mandelbrot, herringbone, cellular automaton etc . etc. – and explain how the brain is able to abstract from *all of them together* and recognize them collectively as “patterns” (and not just as Koch curves/herringbones etc. etc). Where’s the pattern in a set of diverse patterns, B & B? And where’s the complexity, Jim? that's easy, these are all obviously susceptible to lossy compression using algorithms native to the brain... ben AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription AGI | Archives | Modify Your Subscription ------------------------------------------- AGI Archives: https://www.listbox.com/member/archive/303/=now RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393 Modify Your Subscription: https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968 Powered by Listbox: http://www.listbox.com
