Jump to content
Objectivism Online Forum

SpookyKitty

Regulars
  • Content Count

    420
  • Joined

  • Last visited

  • Days Won

    7

SpookyKitty last won the day on April 24 2018

SpookyKitty had the most liked content!

1 Follower

About SpookyKitty

  • Rank
    Member

Previous Fields

  • Country
    United States
  • Interested in meeting
    No
  • Relationship status
    Single
  • Sexual orientation
    Straight
  • Experience with Objectivism
    Atlas Shrugged, Fountainhead, ITOE, Objectivism The Philosophy of Ayn Rand and various articles

Profile Information

  • Gender
    Female

Recent Profile Visitors

2243 profile views
  1. Yes, that seems like a very important distinction to keep in mind. The only thing I would change is to avoid naming this characteristic "constructible" and the values "0" and "1", and instead use the terms "hypothetical" and "actual". An entity with the value "hypothetical" denotes an entity that has merely been imagined, whereas the value "actual" denotes an entity that has actually been observed. I'm sure there's a fancy word for this kind of distinction, but I forgot what it was. Is this what you have in mind? Yeah, it would be absurdly hard, since then you would have to reproduce a human-like perceptual system. Instead, the point here is to represent entities as they are represented in conscious reasoning, and, during the course of conscious reasoning, you cannot conceive of a specific entity apart from its characteristics.
  2. Two characteriatics A and B are commensurable if and only if they are identical as types. For example, Char1 is commensurable with Char1 and nothing else. Color is commensurable with color and nothing else.
  3. Characteristics are supposed to be anything you can directly perceive. For example color, texture, shape, etc. Their terms are the values of those characteristics like red, rough, round, etc. I Will make a more detailed submission soon.
  4. EUREKA! After several cups of coffee and another sleepless night, I now have an example of a second-order concept. Basically, this concept is a concept about first-order concepts. First-order concepts are those that are directly about entities at the perceptual level, such as the concept C above. The second-order concept I've cooked up, called Valid_1_Conc determines whether a given first-order concept is actually valid or not according to known Objectivist Epistemology. Let's take a look at it: Inductive Valid_1_Conc : (Entity -> Prop) -> Prop := | is_valid (F : Entity -> Prop) (H : exists (e1 : Entity) (e2 : Entity), F e1 /\ F e2 /\ ~(e1 = e2) /\ (forall (x x' : Char1) (y y' : Char2) (z z' : Char3), (e1 = Kind1 x y z /\ e2 = Kind1 x' y' z') -> (x = x' \/ y = y' \/ z = z'))) : Valid_1_Conc F. First, note its type declaration, (Entity -> Prop) -> Prop. This concept takes abstract functions (of which valid first-order concept are a sub-type) and outputs propositions about them. Its constructor requires two inputs. First, a term of type Entity -> Prop. Second a proof of the condition H, which, in English, says that "The concept F must (1) Accept entities e1 and e2 (2) The entities e1 and e2 cannot be equal (that is, there must be at least two distinct entities that the concept refers to) (3) The two entities must share a value of at least one characteristic in common, (that is, arbitrary collections of entities are not concepts according to Objectivism as it currently stands)." The beauty of this thing is that by proving propositions such as (Valid_1_Conc C) as we are about to do, we are precisely validating a first order concept. Here is the theorem: Theorem : Valid_1_Conc C Proof. In order to prove Valid_1_Conc C we use the only constructor available, that being is_valid. "is_valid" requires an F of type Entity -> Prop, but that is already given as C. The next thing we need is a proof of H. To prove an "exists " statement we must actually produce an entity with the required properties. We see that we actually require two such entities. Let e1 = Kind1 a d f and e2 = Kind1 a e f. Now, we have to prove a total of four propositions (1) C (Kind1 a d f) (2) C (Kind1 a e f) (3) (Kind1 a d f) =/= (Kind1 a e f) (4) (forall (x x' : Char1) (y y' : Char2) (z z' : Char3), (e1 = Kind1 x y z /\ e2 = Kind1 x' y' z') -> (x = x' \/ y = y' \/ z = z')) The proof of (1) is just the theorem in the previous post, so that's taken care of. The proof of (2) is just the proof of (1) with "e" chosen instead of "f" at the appropriate step. To prove (3) we assume that (Kind1 a d f) = (Kind1 a e f) and derive a contradiction. We do so by noting that this works only if a = a, d = e, and f = f. Since the second of these is a contradiction, we're done here. Proving (4) is not as hard as it seems. Let x, x', y, y', z, and z' be any arbitrary values. The statement now reduces to "if (e1 = Kind1 x y z /\ e2 = Kind1 x' y' z'), then (x = x' \/ y = y' \/ z = z')" Note that e1 = Kind1 a d f and e2 = Kind1 a e f from the "exists" steps so that the hypotheses now say Kind1 x y z = Kind1 a d f and Kind1 x' y' z' = Kind1 a e f From this we can immediately derive the six equations: x = a x' = a y = d y' = e z = f z' = f and we only need to prove one of the disjuncts of x = x' \/ y = y' \/ z = z'. We choose the first one (though the third would work just as well and the second would be impossible). Since x = a = x', it follows that x = x' and the proof is complete. The automatically generated proof of this theorem is MONSTROUS, but I'll post it for the curious: (is_valid C (ex_intro (fun e1 : Entity => exists e2 : Entity, C e1 /\ C e2 /\ e1 <> e2 /\ (forall (x x' : Char1) (y y' : Char2) (z z' : Char3), e1 = Kind1 x y z /\ e2 = Kind1 x' y' z' -> x = x' \/ y = y' \/ z = z')) (Kind1 a d f) (ex_intro (fun e2 : Entity => C (Kind1 a d f) /\ C e2 /\ Kind1 a d f <> e2 /\ (forall (x x' : Char1) (y y' : Char2) (z z' : Char3), Kind1 a d f = Kind1 x y z /\ e2 = Kind1 x' y' z' -> x = x' \/ y = y' \/ z = z')) (Kind1 a e f) (conj (_is_a_C (Kind1 a d f) d f eq_refl) (conj (_is_a_C (Kind1 a e f) e f eq_refl) (conj (fun H : Kind1 a d f = Kind1 a e f => let H0 := eq_ind (Kind1 a d f) (fun e : Entity => match e with | Kind1 _ d _ => True | Kind1 _ e _ => False | Kind2 _ _ => False | Kind3 _ => False | Kind4 _ => False end) I (Kind1 a e f) H : False in False_ind False H0) (fun (x x' : Char1) (y y' : Char2) (z z' : Char3) (H : Kind1 a d f = Kind1 x y z /\ Kind1 a e f = Kind1 x' y' z') => match H with | conj H0 H1 => let H2 := match H0 in (_ = y0) return (y0 = Kind1 x y z -> x = x' \/ y = y' \/ z = z') with | eq_refl => fun H2 : Kind1 a d f = Kind1 x y z => (fun H3 : Kind1 a d f = Kind1 x y z => let H4 := f_equal (fun e : Entity => match e with | Kind1 _ _ c2 => c2 | Kind2 _ _ => f | Kind3 _ => f | Kind4 _ => f end) H3 : f = z in (let H5 := f_equal (fun e : Entity => match e with | Kind1 _ c1 _ => c1 | Kind2 _ _ => d | Kind3 _ => d | Kind4 _ => d end) H3 : d = y in (let H6 := f_equal (fun e : Entity => match e with | Kind1 c0 _ _ => c0 | Kind2 _ _ => a | Kind3 _ => a | Kind4 _ => a end) H3 : a = x in (fun H7 : a = x => let H8 := H7 : a = x in eq_ind a (fun c : Char1 => d = y -> f = z -> c = x' \/ y = y' \/ z = z') (fun H9 : d = y => let H10 := H9 : d = y in eq_ind d (fun c : Char2 => f = z -> a = x' \/ c = y' \/ z = z') (fun H11 : f = z => let H12 := H11 : f = z in eq_ind f (fun c : Char3 => a = x' \/ d = y' \/ c = z') (let H13 := match H1 in (_ = y0) return (y0 = Kind1 x' y' z' -> a = x' \/ d = y' \/ f = z') with | eq_refl => fun H13 : Kind1 a e f = Kind1 x' y' z' => (fun H14 : Kind1 a e f = Kind1 x' y' z' => let H15 := f_equal (fun e : Entity => match e with | Kind1 _ _ c2 => c2 | Kind2 _ _ => f | Kind3 _ => f | Kind4 _ => f end) H14 : f = z' in (let H16 := f_equal (fun e0 : Entity => match e0 with | Kind1 _ c1 _ => c1 | Kind2 _ _ => e | Kind3 _ => e | Kind4 _ => e end) H14 : e = y' in (let H17 := f_equal (fun e : Entity => match e with | Kind1 c0 _ _ => c0 | Kind2 _ _ => a | Kind3 _ => a | Kind4 _ => a end) H14 : a = x' in (fun H18 : a = x' => let H19 := H18 : a = x' in eq_ind a (fun c : Char1 => e = y' -> f = z' -> a = c \/ d = y' \/ f = z') (fun H20 : e = y' => let H21 := H20 : e = y' in eq_ind e (fun c : Char2 => f = z' -> a = a \/ d = c \/ f = z') (fun H22 : f = z' => let H23 := H22 : f = z' in eq_ind f (fun c : Char3 => a = a \/ d = e \/ f = c) (or_introl eq_refl) z' H23) y' H21) x' H19) H17) H16) H15) H13 end : Kind1 x' y' z' = Kind1 x' y' z' -> a = x' \/ d = y' \/ f = z' in H13 eq_refl) z H12) y H10) x H8) H6) H5) H4) H2 end : Kind1 x y z = Kind1 x y z -> x = x' \/ y = y' \/ z = z' in H2 eq_refl end)))))))
  5. I've been working on an idea I had a while back that Objectivist Epistemology is equivalent to a fragment of Intuitionistic Type Theory. From the The Stanford Encyclopedia of Philosophy: I am not the fist to notice that the Objectivist account of concepts is somehow linked to programming. Furthermore, through type theory we can go further and link up the Objectivist notion of concepts with logic in a formal, constructive, and algorithmic way. I hope to use type theory to eventually give a detailed account of higher-order concepts and logic consistent with Objectivism. I will now explain what I've discovered so far. According to Objectivism, there are things called "entities" which can have "characteristics" or "attributes". The characteristics themselves can take on several different (finitely many) values which we observe through perception. We can represent characteristics as types quite simply in type theory. Imagine that we live in a very simple universe where there are only three possible characteristics called "Char1" "Char2" and "Char3". Each one can further have one of a small number of values unique to that characteristic. Inductive Char1 : Type := | a : Char1 | b : Char1 | c : Char1. Inductive Char2 : Type := | d : Char2 | e : Char2. Inductive Char3 : Type := | f : Char3 | g : Char3 | h : Char3 | i : Char3. The first definition above says "Char1 : Type" which is read as "Char1 is of type Type" (more accurately, "Char1 is of sort Type", but don't worry about that for now). It further says that "a : Char1", i.e. "the term 'a' is of type 'Char1'". The term "a" represents the value "a" of the characteristic corresponding to the type "Char1". Similarly for the rest. Now, an entity can be represented as simply a list listing the values of all of its characteristics. However, some entities are not commensurable, i.e., they don't have the same characteristics. To take that into account, we divide entities into several different kinds according to their characteristics. Inductive Entity : Type := | Kind1 : Char1 -> Char2 -> Char3 -> Entity | Kind2 : Char2 -> Char3 -> Entity | Kind3 : Char1 -> Entity | Kind4 : Char2 -> Entity. The arrow notation above in "Kind1 : Char1 -> Char2 -> Char3 -> Entity" says that in order to produce an Entity of Kind1, we need to list one term of Char1, then on term of Char2, then one term of Char3. Thus, the terms of the type "Entity" all look like the following examples: Kind1 a d f Kind2 e g Kind3 b ... and so on. I assume that concepts can only be formed from entities that are fully commensurable, i.e. are all of the same kind. The reason is, though I have not proven it yet, that mixing entities of different kinds within the same concept can lead to inconsistency. Here is an example of a concept: Inductive C : Entity -> Prop := | is_a_C (e : Entity) (y : Char2) (z : Char3) (H : e = Kind1 a y z) : C e. The symbol "C" is the name of the concept. Its type is "Entity -> Prop", and so a concept is an abstract function which takes entities as inputs and produces propositions (propositions are types in the sort "Prop") as outputs. The propositions that are produced are described by the last part "C e". It says, essentially "e is a C" (for a concrete example, "Socrates (an entity) is a Man (a concept)"). Since "e" must be an entity, then, concretely, the terms of the type C all look like the following examples: C (Kind1 a d f) -> "The entity 'Kind1 a d f' is a C" C (Kind2 e g) C (Kind1 b d f) ... Since "C (Kind1 a d f)" is also a proposition, and propositions are just types, what are the terms of "C (Kind1 a d f)"? The terms of the type "C (Kind1 a d f)" are the proofs of the proposition "C (Kind1 a d f)". The stuff to the left of the "C e" tells us how to construct the proofs of the proposition "C e". Specifically, it tells us that we need four things. First, an entity e, then a value y of Char2, then a value z of Char3, and finally a proof H of the proposition "e = Kind1 a y z" , that is, a proof that the entity e is identical to the entity (Kind1 a y z), where a is the value of Char1 (i.e., a constant) from before. The important thing to take away from all this is that a concept contains lots of non-trivial logical as well as computational structure. Let's look at an example of how a concept may be used. To do that, we prove the following theorem: Theorem : C (Kind1 a d f) which, in English, says that "The entity of kind1 whose first characteristic has value a, whose second characteristic has value d, and whose third characteristic has value f is a C". We will prove this using a "backwards style proof", one where we start with the conclusion and then work our way backwards to see what we would have to prove before we can prove it. During the course of the proof, we build up a "proof object", i.e., a term of the type C (Kind1 a d f). Proof: Our goal is to prove that C (Kind1 a d f). To do that, we refer to the definition of the concept C. It says that the only way to construct a proof of the proposition C (Kind1 a d f), is to use the "is_a_C" constructor. The "is_a_C" constructor says that we first need an e of type Entity. It is already given to us as (Kind1 a d f), so our proof object so far is is "is_a_C (Kind1 a d f)". The next thing we need is a y of type Char2. For this proof, we will use the value "d" as it is the only one that makes sense. Hence, so far we have "is_a_C (Kind1 a d f) d". We now need a z of type Char3. We choose the value "f" for the same reason as before to get "is_a_C (Kind1 a d f) d f". Finally, we need H a proof of the proposition "e = Kind1 a y z". Since e is just (Kind1 a d f), and y is d and z is f (note that if we had chosen different values for y and z we would not be able to complete the proof), this amounts to saying that we need a proof of the proposition "Kind1 a d f = Kind1 a d f". But this is just true by the law of identity. The Coq implementation of the Calculus of Inductive Constructions provides a term of the law of identity as "eq_refl" (reflexive property of equality) . Hence, this is a valid proof of the proposition H. Our completed proof object is now (is_a_C (Kind1 a d f) d f eq_refl). There are a few lessons to learn from this proof. First, it tells us how, exactly, a concept refers to entities. It does so by only allowing entities of the form (Kind1 a * *) through its proof procedure. If we had instead been trying to prove that C (Kind1 a e f) we could have simply chosen the value e for y and we would be able to complete the proof. Thus, this would prove that the entity (Kind1 a e f) is also a C, and we can say that the concept C refers to it. If, on the other hand we had been trying to prove that C (Kind1 b d f) there would be no way to complete the proof whatsoever at the final step since we would never be able to prove the contradiction "Kind1 b d f = Kind1 a d f". The same is true of entities like (Kind2 ...) since the proof would fail right at the start. Therefore, the concept C refers to the entities (Kind1 a * *) where the * represent characteristics whose measurements have been omitted. Second, it tells us that true propositions such as "C (Kind1 a d f)" will have at least one term in them, while false propositions such as "C (Kind1 b d f)" will be empty. Third, we can think of the constructor "is_a_C" as a program whose execution terminates if the inputs to it correspond to an entity e that C accepts and evidence y, z, and H that can together prove the proposition (C e), and which never terminates otherwise. We can even prove theorems like "~ C (Kind1 b d f)" just with what we have so far using the type of reasoning above where we showed why C cannot accept the entity (Kind1 b d f), but I won't be doing so formally here as the proof term that Coq provides is ridiculously complicated (though the proof procedure itself is dead simple). The machinery we have so far can even be extended to differentiation. For example, we can define a sub-concept of C whose values in its third characteric are restricted to the range of values i and h as follows: Inductive C_diff : Entity -> Prop := | is_a_C_diff (e : Entity) (y : Char2) (z : Char3) (H0 : e = Kind1 a y z) (H1 : z = h \/ z = i) : C_diff e. The last part (H1 : z = h \/ z = i) says that, in addition to the stuff we needed before, we also need a proof of the proposition that the value z is equal to either h or i. Here is a sample theorem: Theorem : C_diff (Kind1 a d h). Proof. Proceeding analogously as before, we construct (is_a_C_diff (Kind1 a d h) d h eq_refl (--something--)). To complete the proof we need to provide a proof of H1. We note that z = h, and therefore, H1 becomes (h = h \/ h = i). In order to prove an "or-statement" in the Calculus of Inductive Constructions we need to make a choice of either proving the left or right disjunct. Here we will choose the left one which we can prove easily by using eq_refl, and definitely not the right one which is a contradiction. So our completed proof object is now (is_a_C_diff (Kind1 a d h) d h eq_refl (or_introl eq_refl)).
  6. Thank you for your intelligent and thoughtful contributions. You two are truly the bearers of a deep and enlightening philosophy. I will immediately file away these incredible insights alongside the deep wisdom I acquired from flat-earthers, creationists, and post-modernists. As a more direct answer to William O from a more traditional Objectivist perspective, some concepts are formed solely through introspection. These are what Rand called "concepts of consciousness", and modus ponens is one of them. I am currently writing up a short paper which explains how the various concepts of logic ("and" "or" "not" "implies") are derived from introspection, and I will hopefully have something posted sometime tonight. I can't guarantee that I will get to modus ponens, but I can guarantee that I will get to conjunctive and disjunctive introduction and elimination rules, the rule of assumption, and ex falso quodlibet.
  7. I am currently working on a theory which aims to show that the formal theory underlying the Objectivist process of concept formation is something very similar to Per Martin-Lof's Intuitionistic Type Theory. If we understand Objectivist concepts as types, then a statement like A -> B says that there is a computable function which transforms any proof/construction of the concept/type A into a proof/construction of the concept/type B. The rule of modus ponens is then simply function application. If f : A -> B, then the term f a , where a is a proof/construction of concept/type A, is a proof/construction of the concept/type B. Hence, from a proof of A and a proof of A -> B we derive a proof of B. One could then argue that the rule of modus ponens is somehow inherent in any process of computation. This is just what it means for a concept to be "axiomatic" in Objectivist terminology.
  8. I'm sorry but this seems to just be a very clever restatement of General Relativity.
  9. It's the same with sex! Moral of the story: Chess is neither work nor war. It's just sex.
  10. This is one of my main disagreements with Objectivism. The description of human cognition that Rand and Peikoff give us sounds good on paper, but then when you actually try to go and use it... you can't. It simply doesn't work. To this date, I have never seen even a single detailed step-by-step account of an interesting and/or useful idea being formed and validated using the Objectivist method. Formal logic and probability theory are the way to go.
  11. DavidOdden I would like to see your attempts to solve the above problems. Since you seem to know more about concept formation than anyone else, a conceptual breakdown of the process you used to tackle the problems would be very valuable.
  12. What are you even hoping to achieve by insulting me? Read the thread title and apply yourself to the given problem.
  13. Suppose you had an infinite amount of time on your hands. How exactly would you go about searching through every possible function? How would you know when you've found the best one possible? What do you imagine distinguishes it from all the rest?
  14. @Nicky and @Doug Morris Is it possible to measure the "simplicity" of a provided answer? Also, just drop this side argument about what an open-ended question is and isn't. It's just fucking stupid and not even remotely productive.
×
×
  • Create New...