Re: Revised Computing Randomness & the UD
Hi Hal, >Unfortunately I still miss some of your posts because of the absence of a >date stamp, my mailer puts them way at the top of my list since I save all >of some categories of e-mail. I'm really sorry. I *will* come back to Eudora one day >Now I consider myself in favor of the idea that some input from a random >oracle is required in a SAS supporting universe so I am posing my arguments >in an effort to defeat determinism in such a setting. Determinism has been defeated in the comp setting through the UDA, IMO. To be sure, oracle appears too. That is, the UD generates and executes all programs, including the one which are using (Turing) oracle. Bruno
Re: Revised Computing Randomness & the UD
Hi Hal, >These are some of the things I want to explore as a result of formulating >my question re the UD. > >To start: > >1) Are you saying that the UD contains all other computations as data? No. The UD is a program without data. It generates and executes all computations. It "dovetails", i.e. it executes all programs by little pieces, so it does not stuck itself in a non stopping computation. >... But for sake of covering all bases I thought >"compute" = "prove" was established by Turing. The work of Post, Church, and Turing (and others) shows the contrary. It shows the abyssal difference between provability and computability. Godel realises that his own incompleteness protect Church thesis and gives to the notion of computability a miraculous absolute nature. At the same time it shows that provability is always relative to a choosen formal system. That is what makes the Universal Dovetailer really universal. It computes everything computable, but is does not makes proof. If you look at it as a theorem prover, most formal description of the UD will give a not really powerful theory. >3) If in order to work the UD contains all other computations >as data then the UD is a highly complex program. ... I wrote one in LISP. It makes more or less 100 lines. You can write very little one in PROLOG. It is not a highly complex program. But, please, the computations are not the data, they are the output. >Random strings are complex and I thought "all Theorems" was a very low >complexity object. Or is it random? It depends how you code it. "All theorems" is an ambiguous expression. Never use the word "theorem" without mentionning the theory. "All computations" is not ambiguous (it does not depend on the formal system in use). In some sense Chaintin OMEGA number codes all computations in a terrible compressed way. It is "random". But then there exists nice natural way to encode all computation in a non random, but deep (in Bennett sense) manner. >4) Why not run a highly parallel computer rather than a UD? [That is rather >like a part of my model.] First, the meaning of quasi-geometrical word like "parallel" are among the things I want explain. I cannot take such meaning for granted. Actually, with some terrestrial sense for parallel, it will not change a thing in the limit (see the UDA argument which explain why). Also, The UD has a property I never mentionned, it generates by itself an infinity of variant of itself which on almost all inputs (almost all = all with a finite number of exceptions) is "quasi-infinitely" more quick. The UD is self-speedable. But that is not interesting, for the UD does not need to go quickly. Conscience and matter are inside view of UD*, the complete (infinite) work of the UD; belonging to Plato's heavens. >5) My meaning for "knowing" is at first order like proving. Chaitin is >actually talking about the complexity of the FAS's theorem checker as the >complexity of the FAS. The theorem checker "knows" all theorems of the >FAS. If a proof is an elegant proof by default then the theorem checker >also knows this proof or it would not know how to check the output for >theorem hood. It can prove this proof is elegant because there is no other >choice. OK. But be careful. In my work knowing is quite different of proving. The modalities are not the same. This will be discussed in my explanation to George Levy. In fact I will define at some point "knowing p" by proving p and p is true. knowing and proving will appear equivalent only for the guardian angel, but the consistent machine cannot know that, nor prove it! Bruno
Re: Revised Computing Randomness & the UD
Dear Bruno: Unfortunately I still miss some of your posts because of the absence of a date stamp, my mailer puts them way at the top of my list since I save all of some categories of e-mail. I found one from last week which I will respond to soon. However, here is an addition to my latest response. 6) The UD as I see it is a FAS in its own right. It is generating a proof "in" itself. There is no "out" or we would be discussing just a piece of the Everything. I would call this proof a deterministic cascade since there are no other proofs of any of its outputs [no "in coming branches] and it is unique at each step [no outgoing branches]. Now I consider myself in favor of the idea that some input from a random oracle is required in a SAS supporting universe so I am posing my arguments in an effort to defeat determinism in such a setting. Hal
Re: Revised Computing Randomness & the UD
Dear Bruno: These are some of the things I want to explore as a result of formulating my question re the UD. To start: 1) Are you saying that the UD contains all other computations as data? 2) By your comment "You confuse computability and provability." I think you mean the UD indeed contains all computations as data. But for sake of covering all bases I thought "compute" = "prove" was established by Turing. Though I know some want to expand beyond this. Regardless, IMO even if it has all this data it is still an expression plus data plus self-delimiter that outputs [proves] a value. That is the "sense" I am using. 3) If in order to work the UD contains all other computations as data then the UD is a highly complex program. A small expression with a lot of data and a very large self-delimiter. If you have to put the data string in to get it out that is a sign you are working with a random string. Random strings are complex and I thought "all Theorems" was a very low complexity object. Or is it random? 4) Why not run a highly parallel computer rather than a UD? [That is rather like a part of my model.] 5) My meaning for "knowing" is at first order like proving. Chaitin is actually talking about the complexity of the FAS's theorem checker as the complexity of the FAS. The theorem checker "knows" all theorems of the FAS. If a proof is an elegant proof by default then the theorem checker also knows this proof or it would not know how to check the output for theorem hood. It can prove this proof is elegant because there is no other choice. Hal
Re: Revised Computing Randomness & the UD
Hal Ruhl wrote: >1) The UD proof of the object "all theorems" is complex because each step >is a unique slice of progress towards some sub component of the target >object thus all steps are different and there are a great many of them. > >2) The UD knows its proof is complex and since it is the only way it has to >the target object it knows it is elegant. >The UD [a simple FAS] - uses an incredibly complex proof [known by the UD >it to be elegant and incredibly complex] - to exhibit "all theorems" [an >object of extremely low complexity]. You confuse computability and provability. The UD does not generates formal proof in a formal system, it executes all possible computations. There is a sort of absolute sense (with Church's thesis) in saying that the UD generates all the proof of \Sigma_1 sentences, though. A sentence which is provably equivalent to ExP(x) (It exist a x such that P(x) with P recursive). And there are surely a lot of other senses, but then you need to define the one you are using. How do you define "knowing"? In what sense does the UD know something? It can be shown that the set of *all theorems* of a reasonable first order arithmetic theory is not so low; with reasonable logical complexity definition. Bruno
Re: Revised Computing Randomness & the UD
Dear Russell: As simple as I can get the idea: The UD [a simple FAS] - uses an incredibly complex proof [known by the UD it to be elegant and incredibly complex] - to exhibit "all theorems" [an object of extremely low complexity]. This does not work. Hal
Re: Revised Computing Randomness & the UD
Dear Russell: I think you miss what I am saying. At 4/20/01, you wrote: >I disagree. The UD will have a particular way of generating (or >enumerating) the theorems of the FAS, such that it doesn't generate >the same theorem twice. The UD is [so it is said] generating all theorems. Some of these are no doubt not elegant. It does not generate proofs twice but may generate more than one proof of a given object. All this is not germane to the fact that it is doing all this by the only path [proof] - {its "particular way" as you say} that it has available. That makes this path elegant by default. Thus we have an algorithm - the generating FAS - generating a very simple object [so it is said] by an incredibly complex and elegant "particular way". Incredibly complex elegant proofs do not end in very simple objects. Hal
Re: Revised Computing Randomness & the UD
I disagree. The UD will have a particular way of generating (or enumerating) the theorems of the FAS, such that it doesn't generate the same theorem twice. However, that is not to say that the proofs it generates are elegant, as other proof algorithms exist, which may generate shorter proofs. I would think that moreover it is impossible to ascertain in general whether a particular proof output by the UD is elegant or not, as this appears to be equivalent to the problem of evaluating the K-complexity of a given program, which is known to be uncomputable. Cheers Hal Ruhl wrote: > > Dear Russell: > > To condense the idea the UD is generating the collection of strings by the > only path it has so the proof is automatically elegant. > > It is also extremely complex. > > Hal > Dr. Russell Standish Director High Performance Computing Support Unit, Phone 9385 6967 UNSW SYDNEY 2052 Fax 9385 6965 Australia[EMAIL PROTECTED] Room 2075, Red Centrehttp://parallel.hpc.unsw.edu.au/rks
Re: Revised Computing Randomness & the UD
Dear Russell: To condense the idea the UD is generating the collection of strings by the only path it has so the proof is automatically elegant. It is also extremely complex. Hal
Re: Revised Computing Randomness & the UD
Dear Russell: The idea I am trying to exploit in my latest posts is that a deterministic cascade means one that uses every ounce of inference in the FAS in every possible way on all the current data at every single step of the cascade. Each step is individually elegant and since deterministic is taken to mean no other proof of any string of the cascade exists then I called it everywhere elegant. Thus the cascade is known to be elegant by the FAS. Known elegant proofs can not be more complex than the FAS itself plus a constant. Cascades can not stop. Thus a contradiction. The UD as I understand it as described by various sources I have read it is applying all of its inference in every possible way to the current data of each step. It is a single track machine. See below: At 4/20/01, you wrote: >But if P(A)=B and P'(B)=C are elegant proofs, it is very unlikely for >P'(P(A))=C to be an elegant proof. That is not as the UD is described as I have read it. It executes one sub component at each step in a determined sequence. It has no choice in this sequence. The fact that it may produce C more than once is not an issue since this is not its goal. Its goal is to produce some collection of strings and that just might contain C more than once but the overall proof of that collection is still the only one available thus it is elegant. snip >But whilst it is necessary for each step of >an elegant proof to be elegant, it is not sufficient I add the constraint of "deterministic" cascade. Hal
Re: Revised Computing Randomness & the UD
But if P(A)=B and P'(B)=C are elegant proofs, it is very unlikely for P'(P(A))=C to be an elegant proof. This is what the dovetailer is constructing - it is not possible to know whether the any particular proof output by the UD is elegant, only that it must contain elegant proofs since it is comprehensive. What exactly do you mean by everywhere elegant proof? That each step of the proof is elegant? But whilst it is necessary for each step of an elegant proof to be elegant, it is not sufficient Cheers Hal Ruhl wrote: > > > Let me correct one little issue which I think helps to clarify what I am > saying. I add a comment on the universal dove-tailer. > > 2) A universal dove-tailer generating all strings using a fixed algorithm > every part of which applies to all the current data in the same way at each > step seems an odd thing. > > A dove-tailer is not directly generating the "whole" ensemble. What it is > doing is selecting by fixed rules a particular string out of the ensemble > and adding some quantity of bits, putting that back in and selecting > another and adding some quantity of bits to it etc., etc. That is a very > selective and complex process on a step by step basis. You wind up > constructing this incredibly complex everywhere elegant proof of what must > then be an incredibly complex object that is nevertheless considered to be > a very low complexity object. > > If I have the ideas of a UD, elegant proof, and AIT complexity correct the > UD appears to be a contradiction. > > The contradiction again is that we have a FAS that constructs a proof it > knows is elegant that is nevertheless far more complex than any proof it > can know is elegant. > > Hal > Dr. Russell Standish Director High Performance Computing Support Unit, Phone 9385 6967 UNSW SYDNEY 2052 Fax 9385 6965 Australia[EMAIL PROTECTED] Room 2075, Red Centrehttp://parallel.hpc.unsw.edu.au/rks
Re: Revised Computing Randomness & the UD
Let me correct one little issue which I think helps to clarify what I am saying. I add a comment on the universal dove-tailer. 1) Yesterday I said that the cascade 1+1=2, 2+1=3, 3+1=4, etc. was not everywhere elegant. But I went outside my identification process for "determinism" = "everywhere elegant proof" to do so. The error was to slip into full number theory and think: "84" one of the strings this cascade will eventually reach for example has more than one proof. That is true because number theory is richer than just "data + 1 =". But here the working definition of "deterministic" I use is that all of the selected set of rules of a cascade act at each step on all the data. You should not have within the idea of "deterministic" some of the rules active today and some others active tomorrow, some of the data regions exempt from some rules today and other data regions partially exempt tomorrow unless that was itself in Rj. Deterministic as I understand it = all the rules of physics always apply to the entire state of the universe. So for the above cascade we have selected "data + 1 =" as the exclusive expression - the entire rule set of the operative FAS - for which we are seeking the cascade of values given some start data [effectively the axiom of that cascade] in this case "1". The fact that this rule may also belong to a different and richer FAS is not germane to the cascade viewed as an attempted deterministic sequence. The operative FAS contains just this one rule as its Rj and applies it to all the data at each step. That makes this cascade everywhere elegant because there is no other proof of any of the output strings available in the operative FAS. Thus hits the complexity wall established by the complexity of the operative FAS. 2) A universal dove-tailer generating all strings using a fixed algorithm every part of which applies to all the current data in the same way at each step seems an odd thing. A dove-tailer is not directly generating the "whole" ensemble. What it is doing is selecting by fixed rules a particular string out of the ensemble and adding some quantity of bits, putting that back in and selecting another and adding some quantity of bits to it etc., etc. That is a very selective and complex process on a step by step basis. You wind up constructing this incredibly complex everywhere elegant proof of what must then be an incredibly complex object that is nevertheless considered to be a very low complexity object. If I have the ideas of a UD, elegant proof, and AIT complexity correct the UD appears to be a contradiction. The contradiction again is that we have a FAS that constructs a proof it knows is elegant that is nevertheless far more complex than any proof it can know is elegant. Hal
Re: Revised Computing Randomness
More detail: Start with an axiom Aj, use it as data for a LISP expression Rj. A program that computes the value of this expression "B" in AIT has length: Expression + data + self-delimiter or in this case Rj + Aj + Self-delimiter. Like this: P = {Expression(data) + self-delimiter} computes [produces value] "B" If P is the shortest program to do so then it is called elegant and its length is the complexity of "B". In an everywhere elegant cascade each P(i) is longer than P(i -1) because the data for P(i) alone has the same complexity as P(i - 1) then you add the complexity of the expression and the self-delimiter. What is left is to 1) identify "everywhere elegant cascade" = deterministic evolution of the isomorphism and 2) find an upper bound on the complexity of the P(i). Chaitin did the latter by putting an upper limit on the complexity an N-bit FAS could prove an elegant program has i.e. it own complexity plus a constant. But a deterministic cascade is already known to be an everywhere elegant program. A contradiction is established unless the cascade stops but it can not stop. The only way out is a spontaneous increase in the complexity of the FAS. Hal
Re: Revised Computing Randomness
Dear Russell: Also since 1+ 1 = 2, 2 + 1 = 3, ... etc. is not everywhere elegant each successive string is not necessarily more complex than its preimage. However, in a cascade that is everywhere elegant each successive string is more complex than its preimage by definition of elegant proof. Hal
Re: Revised Computing Randomness
Dear Russell: Here is the cascade rewritten: Pj(i) = {Rj(Uj(i - 1)) + Self-delimiterj(i)} computes Uj(i) with the added constraint that the cascade is everywhere elegant. It is similar to: 1+ 1 = 2, 2 + 1 = 3, ... which eventually produces strings of arbitrary complexity despite its mechanical repetition. The difference is that 1 + 1 = 2, 2 + 1 = 3 etc is not everywhere elegant so does not meet a complexity wall. Hal Uj(i) is more complex than Uj(i -1) At 4/19/01, you wrote: >This last statement is surely incorrect. Because the map Rj is a >mechanical application of rules, the complexity of B is no greater >than that of Aj - you can only get out what you put in.
Re: Revised Computing Randomness
Hal Ruhl wrote: > > Here is a revised version of my comments on this subject. I think it fixes > several aspects of what I have had to say earlier. > > Standalone deterministic evolving universes: > > Such a universe is describable as a concatenation of single output programs > of the form: > > Rj(Aj) -> B; Rj(B) -> C; ... Rj(F) -> G; Rj(G) -> H; Rj(H) -> I; > > > "Rj" is a subset of the rules of an N-bit FAS. "Aj" is an axiom of that > FAS. There are no potential branches in or out. It is an elegant proof > and defines the complexity of each of its objects B,C,...F,G,H,I ... > > Because the proof is everywhere elegant each successive object is more > complex than its preimage. This is the type of theorem cascade I was > trying to identify. > This last statement is surely incorrect. Because the map Rj is a mechanical application of rules, the complexity of B is no greater than that of Aj - you can only get out what you put in. Cheers Dr. Russell Standish Director High Performance Computing Support Unit, Phone 9385 6967 UNSW SYDNEY 2052 Fax 9385 6965 Australia[EMAIL PROTECTED] Room 2075, Red Centrehttp://parallel.hpc.unsw.edu.au/rks
Revised Computing Randomness
Here is a revised version of my comments on this subject. I think it fixes several aspects of what I have had to say earlier. Standalone deterministic evolving universes: Such a universe is describable as a concatenation of single output programs of the form: Rj(Aj) -> B; Rj(B) -> C; ... Rj(F) -> G; Rj(G) -> H; Rj(H) -> I; "Rj" is a subset of the rules of an N-bit FAS. "Aj" is an axiom of that FAS. There are no potential branches in or out. It is an elegant proof and defines the complexity of each of its objects B,C,...F,G,H,I ... Because the proof is everywhere elegant each successive object is more complex than its preimage. This is the type of theorem cascade I was trying to identify. It has a problem: It will - if it does not stop - eventually outrun the complexity of its N-bit FAS. Since the proof is known to be elegant this is [as I understand it] a contradiction of Chaitin's result. However, if it stopped at complexity N + c as required by Chaitin [again I think] the end state would be highly but finitely complex. Highly but finitely complex states are going to have some consequent under the rules of the cascade. It can not be both "provably highly but finitely complex" and "absent operative correlations" under the same set of finite rules. It might reach an absence of operative correlations when it becomes infinitely complex and the domains of any correlations fall below the resolution of the still finite Rj of the cascade. So the cascade can not stop prior to an infinity complex end state. This seems to be a contradiction. It can be cured if the FAS is able to spontaneously take in additional complexity from an outside source. I call such a source a Superverse. The new information would act to replace the latest object of the cascade with a new axiom acceptable to the Rj and the cascade proceeds. I use this to argue in favor of non deterministic standalone universes. Now this universe is actually an isomorphism linked to the successive strings [objects] B, C etc. The isomorphism has the same Rj as its laws of physics. This should continue to hold even if one shifts to a universal dovetailer running all universes simultaneously. Individual isomorphisms would still jump to the string currently containing as the latest sub string [string structure: ;sub string;sub string;latest substring] that is the successor state of that isomorphism under its individual Rj. Thus each individual isomorphism remains a single output program or elegant proof under its Rj. A non deterministic result is sustained. Random histories are just isomorphisms with Rj = "do not care" and they jump as easily as any isomorphism. Well that is the current state of the argument. Hal