INTROD UC TION T O

IND UC TIVE LOG IC

Logic. Lecture 7.
1. I N D U C TI O N : TH E WH Y? AND T HE WH AT?

• Deduction is surely coherent and powerful reasoning pattern that capacitates one to move
from tacit knowledge or assumed propositions to other propositions which they entail.
However, in search for truths from which the reasoning process is initiated, this inferential
move is insufficient.

• As David Hume puts it, reasoning process usually begins with some “matters of fact” —
accepted truths. Knowledge, into which it arrives, consists of, and is derived, from both
inferential and non-inferential compounds, as Wilfrid Sellars has it. To establish such
matters of fact one uses inductive reasoning.

2
‘
’
‘
’
1. I N D U C TI O N : TH E WH Y? AND T HE WH AT?
• Inductive arguments conceived as object of logical investigations may be outlined as such
that their conclusions are claimed to follow from the premises only with probability, but
without certainty, and the probability itself here is a matter of degree and scaling,
dependent on what else may be the case (both as something [other] or as nothing at all).

• Therefore, inductive logics, even in formalised form and notation, do not work as other
formal logics do: although reminding fuzzy and many-valued logics in their valuations, they
are not identical or even equivalent, since inductive reasoning rather borrows then invents
its own apparatus of such an evaluation of the arguments from the two sources, (a)
probability theory (where numerical calculation is the most efficient and applicable at all),
namely, basic frequentist P(A) = 0 ≤ n ≤1; (b) informal reasoning (since inductive reasoning
is tightly ingrained into the fabric of many activities and practices of crucial importance,
such as law, politics, science & media).

3
‘
’
‘
’
 2 . IND UC TION — D E D UC TION DI F F ER E NC E R E V IS IT E D

• There’re many principal differences in structure, functions, functionality, scope, use of

deductive and inductive reasoning, I will take only a part for the sake of demonstration of
their incommensurability with regards to reduction and unification:

1. Relation between the premises and conclusion: in deductive argument the claim is made
that conclusions follow with certainty from the premises which is itself a relation of logical
necessity; in inductive argument, the claim of certainty is not made, therefore, the
conclusions / arguments are not assessed on being neither true / false, nor valid / invalid —
the arguments in support of inductive inferences are relied on as merely probable or
probably true — at most.

4
’
’
 2 . IND UC TION — D E D UC TION DI F F ER E NC E R E V IS IT E D
2. In logical structure of science, while deductive pattern is utilised for scientific reductions,
explanations and generally speaking oriented on theorising and unification of
[propositional] knowledge, induction deals with methodology, contents of states of affairs,
generalisation of what is to become, as a further stage a matter of deduction; anticipation
(prediction, prognosis, assessment of future states of affairs), as well as assessment /
evaluation of singular particular cases in science (as well as everyday reasoning), heavily
rely on formal and informal methodology and patterns of inductive or induction-based
reasoning.

3. Some semiformal inductive logics have their peculiar versions of proof systems (akin, say, to
truth tables but for probabilistic calculus of propositional logical expressions), but no such
system is decidable (and the very notion of decidability requires revisiting since the system
would simply lack properties defining classical logical decidability).

4. In both modes of reasoning there are, surely, qualitative and quantitative dimensions, but
in deduction, qualitative aspects (meaning, content, truth) would always prevail, while for
inductive methods and arguments quantity is either equipotent to quality (e.g., the more
cases observed with such-and-such outcome, the stronger is the argument in favour of
general rule concerning it), or even decisive over quality.
5
’
’
3 . I N D U C T I VE A RG U M E N T S :
AN ALOGY
6
 3 .1. A N A L O G I C A L A RG U M E N T S
• The most common type of inductive argument relies on analogy.

• Analogy — a parallel drawn between 2 ≤ n entities or two different collections of entities

(making up situations, states of affairs, behavioural dispositions or anything comparable)
which is made through indication of one or more respects in which they are similar or
invoke symbolic reciprocity.

• It’s main utility is explanation (making x at least to some extent more ostensibly thinkable
by comparison to y, which is presumably more familiar to recipient, by explicating
similarities and resemblances between x and y in certain domains and features).

• Secondary ‘commonsense-level’ utility of analogies is also of nonargumentative nature: for

the purpose of lively descriptions. It is with this intention, such tropes of fiction and non-
fiction literature as metaphors or compound epithets are utilised; all these techniques are
basically structured as analogies.

7
 3 .1. A N A L O G I C A L A RG U M E N T S
• Analogical argument is an inductive argument where from the fact that 2 ≤ n entities (or
their collections) in some respect(s) / domain(s) are alike, it is inferred that they are also
alike in some other respect(s) [without explicit assessment or attesting the latter’s facticity
prior to the drawn conclusion].

• All analogical arguments share the common structure and pattern, varying only numerically
in premises, the amount of entities within the frame of reference, and the level of their
complexity (which do not affect ‘bare’ or ‘formal’ properties of the argument with which
logic deals with, particularly). Every analogical argument proceeds from the similarity of 2 ≤
n entities in 1 ≤ m domains / features / etc. as premises to reach the conclusion where
similarity of those entities in some further respect is postulated.

8
 3 .1. A N A L O G I C A L A RG U M E N T S
• The informal description of the argument: let {a, b, c, d} be a collection of either simple or
compound entities, F, G and H — some their predicates (any attributes or ‘respects’), then:
as a matter of fact, let a, b, c, d all be F and G;
let it also be true, that a, b and c are H as well;
Therefore, it is probable that d also has the property H.

• Formal notation with augmented *FOPLp syntax* (adding the notation and terms from
classical probability theory):
Let D be the domain of discourse D(x) = {a, b, c, d}; then,
∀x(Fx ⋀ Gx) ⋀ (Ra ⋀ (Rb ⋀ Rc))
⊢ P(Rd) > 0.

9
3.2. SIX CRITERIA OF ASSESSMENT

• Analogical arguments, as, actually, any probabilistic inference, are assessed as better or
worse, stronger or weaker with respect to the degree of probability assigned to them —
the one with which, relying on the premises they put forward, their conclusions may be
affirmed. As a meta-scientific (epistemological) tool for applied and detailed assessment or
analysis, it is possible to use several such tools; we will deal with one of them, among the
most widely used.

10
3.2. SIX CRITERIA OF ASSESSMENT

(1) Number of entities. Several “successful*” cases (of experience, things, states of affairs,
instances etc.) of the same kind support the conclusion more than one such case.
Generally speaking (postulating it as a criterion): the larger the number of entities - the
stronger the argument. [‘Strength’ should not be mixed with probability here: here it is a
parametric property, affecting the quantitative stats of the conclusion, taken as a singular
events, but we do not compute probability of premises.]

11
3.2. SIX CRITERIA OF ASSESSMENT
(2) Variability of the instances in the premises. The more dissimilar are generally speaking
entities under assessment in the argument, mentioned only in the premises, the stronger is
the argument itself. The rule is not formalisable, but can be instantiated as follows. Let us
have a set S comprised of complex entities (subsets) A, B, C, each of which is breakable
down to atomics A = {a, a’, a’’, …}, etc., and A ⋂ (B ⋂ C) = ∅. It is asserted that ∀xPx or that
for the most part of x’s Px, then set S can be characterised as almost certainly being ℙ.
Under series of observations, P is uncovered, successively, like: O₁: {a, c’’’}, O₂: {b’, b’’’, a’’’’},
O₃: {c, a’’’’’’’, b’’, c’}, etc. all be P; with each observation the argument’s claim goes stronger,
although it is not converted into deduction.
e.g. Claim on property: ‘Air transport is perhaps the safest form of travelling today’. We
take different air companies, different models of planes, timeframe : accidents number
ratio, do the same for other forms of transport, compare the data and observe that,
although differ greatly in many details, it is the technology itself inherent to aircrafts makes
them safer and nothing beside.

12
3.2. SIX CRITERIA OF ASSESSMENT

(3) Number of similar respects. All the respects in and extent to which the entities in the
premises are akin to one another, and, at the same time, to the instance / entity in
conclusion, increase the probability that the conclusive entity will possess the attribute at
which the argument is aimed. In other words, the greater the number of respects in which
the entity in the conclusion is similar to those from premises, the more probable becomes
the conclusion.

13
3.2. SIX CRITERIA OF ASSESSMENT
(4) Relevance criterion. Similarities (cr.3) add to the strength of the argument when they are
relevant — and a single factor of high relevance would itself contribute more to the
argument than an array of irrelevant similarities. One attribute is said to be relevant to
another when they are connected in a way of causal relation or reciprocal implication
between them (that is why establishing and identification of causal nexus is often
instrumental in determining the strength of the argument or probability of its conclusion).
Analogical arguments can be probable even when the attribute from premises is neither
the cause nor the effect of the conclusion’s attribute, but when both are the effects of one
single cause. Causal connections can be discovered only empirically, beyond logic, by
observation, experiment, investigation etc.

14
3.2. SIX CRITERIA OF ASSESSMENT
(5) Disanalogies identi cation. A point of difference, a respect in which the case we are
reasoning about in conclusion is distinguishable from the cases on which the argument is
based is called disanalogy. Disanalogies undermine analogical arguments when the points
of difference that are identified are relevant, that is, causally connected to the outcome
that we desire and are looking for to demonstrate. Since disanalogies weaken analogical
arguments, they are used critically — to attack or refute an analogical argument
completely or partially. If the differences are indeed ostensibly great — and disanalogy is
therefore critical and cohent — the latter aim is achieved; conversely, whatever holds any
potential disanalogy at bay will contribute to the argument’s defence. NB! With respect to
the principle (2), it should be said that, while in (2) we consider differences that are among
the entities in the premises only, while in (5) a difference is drawn between the cases with
which we have experienced and the one about which is a claim in conclusion being made.
If the latter is the case, the analogy does not hold or is too strained to make up an
argument.

15
fi
3.2. SIX CRITERIA OF ASSESSMENT

(6) A claim of the conclusion. Generally speaking, the more assertions the conclusion
contains, the greater is the burden of defence of the whole claim, and the lesser is its
probability as a joint event. So, we say that the ‘modesty’ of conclusion with regard to
premises determines the coherence of the whole inference and its soundness degree
(although, unlike in classical logic, there is no 0/1 meaning of soundness-as-property).
Therefore, analogical arguments can be strengthened by reduction of the claim made on
the basis of the premises affirmed, as well as by retaining it unchanged (without adding to
the claim), providing it with additional premises, relevant and powerful.

16
3 . 3 . L O G I C A L A N A L O GY A S R E FU TAT I O N
• In addition to proofs by analogical argument, in the domain of our interest it is also a
relatively effective means of refutations: we often seek to demonstrate the weakness of
one’s position / claim by construing a counterargument (deliberately erroneous argument
in this case) of the same logical form as that which we seek to refute.

• Arguments, claims, conjectures, dispositions in science, society, economy and politics, not
holding for deductive, can be countered precisely through presenting other arguments
which have very similar designs, such that their conclusions are known to be false or are
conventionally regarded as being highly or almost certainly improbable.

• So, dealing with an inductive argument one wishes to refute, one can build another
inductive argument of the same form, however, a priori flawed, with a doubtful conclusion;
by means of it, one casts a similar doubt on the conclusion and coherence of the argument
under attack.

17
4 . I N D U C T I VE A RG U M E N T S :
CAUSALIT Y AND CAUSAL
R E AS O NING
18
4 .1. C A U S E & E FFE C T A S CO NC E I VE D BY L O G I C
• Anything that actually (in fact) and / or potentially (in principle) conditionalises and initiates
some particular emergence, occurrence or event is called cause; conversely (but without
vicious circle since cause itself was defined independently), anything that is / was actually
and / or may be potentially conditioned by external entity, and, in this regard, cannot
emerge, be initiated, occur or take place, and does so only after effectuation of this
external entity is said to be the effect of some cause.

• When one knows (or believes that one knows), as the starting point, that X is the cause of Y,
or that Y is effect of X (the difference is in givenness of knowledge to serve as a premise),
one can reason from cause to effect or vice versa, and, if the relations between X and Y
were considered and established correctly and their correlation is indeed a matter of fact,
the reasoning based on their nexus is very fruitful, and potential arguments would be very
strong. Causal reasoning is a form of inductive reasoning in which an effect is inferred from
what is claimed / asserted / proven to be its cause; or some cause is inferred from
something that is claimed / asserted / proven to be effect of it.

19
4 .1. C A U S E & E FFE C T A S CO NC E I VE D BY L O G I C

• Let’s now examine the middle term ‘to conditionalise’, used in definitions of cause and
effect since it is also related to the point since events take place under certain conditions,
and it is customary to demarcate the necessary from sufficient conditions in events
occurrences. A necessary condition is a circumstance / their set in the absence of which the
event cannot occur (like complex amino acids being one of the conditions of living to
emerge, at the same time not necessarily presupposing life, given them alone, like recent
findings of them on Enceladus). A sufficient condition is a circumstance, given it be realised
(actualised), the event must occur (when the temperature is t° ≤ 0°, the water freezes
anyway).

20
4 .1. C A U S E & E FFE C T A S CO NC E I VE D BY L O G I C
• Let’s now examine the middle term ‘to conditionalise’, used in definitions of cause and
effect since it is also related to the point since events take place under certain conditions,
and it is customary to demarcate the necessary from sufficient conditions in events
occurrences.

• A Necessary Condition is a circumstance / their set in the absence of which the event
cannot occur (like complex amino acids being one of the conditions of living to emerge, at
the same time not necessarily presupposing life, given them alone, like recent findings of
them on Enceladus). A Sufficient Condition is a circumstance, given it be realised
(actualised), the event must occur (when the temperature is t° ≤ 0°, the water freezes
anyway). Necessary and Sufficient Condition is the conjunction of necessary conditions for
the occurrence of an event, this conjunction being all that is needed for event’s occurrence
necessitation. So it is basically the sense in which the word cause is used in the concept
‘causal reasoning’.

21
4 . 2 . C A U S A L R E A S O N I N G M E T H O D O L O GY
• When there is a one-directional, linear successive series of occurrences conditioning one
another with one-to-one correspondence (like ‘A → (B → (C → D))’), then it is said that the
case is a causal sequence or causal chain, to which each pair in succession (<A, B>; <B, C>)
is a link. What should be minded here is also how to generally attest the relations between
any of the elements of the chains of a kind.

• Remote cause in any chain of effects and causes is an event distant from the effect for
which explanation is sought. Proximate cause is an event nearest to the event whose
explanation is sought.

• Every occurrence of what may be regarded as a cause producing some effect is an instance
or example of the generality postulated as causal law, a generalisation claiming that such-
and-such circumstances are always accompanied by such-and-such entity. Causal law is a
descriptive law that postulates or asserts a necessity of the nexus between a set of events
which is the cause and another set which is the effect.

22
4 . 2 .1. I N D U C T I O N BY E N U M E R AT I O N

• The process and algorithms of a movement of thought to arrive at universal propositions

expressing some general causal relations, inferred from the particular facts derived from
experience and empirical knowledge as such is called inductive generalisation. Suppose
each observation of an event e₁ is confirmed to always be followed by a separate e₂. One
then may draw a general conclusion about what will happen at every instance of e₁, which
is itself an inductive generalisation.

23
4 . 2 .1. I N D U C T I O N BY E N U M E R AT I O N
• Inductive generalisation which takes such form, as

Observation-1 of e is conditioned by c.
Observation-1 of e is conditioned by c.
Observation-1 of e is conditioned by c.
…
Observation-1 of e is conditioned by c.
⊢ [Therefore,] every instance of e is conditioned by c.

is an induction by simple enumeration.

• Generally speaking, induction by simple enumeration is a tupe of inductive generalisation

in which the premises are instances where phenomena of two kinds repeatedly accompany
one another in certain conditions, from which it is concluded that phenomena of those two
kinds always accompany one another in such conditions.

24
4 . 2 .1. I N D U C T I O N BY E N U M E R AT I O N

• The method’s principal setback is that it takes no (and cannot take) account of exceptions
to the posited general causal law. Exceptions, such as negative instances,
counterexamples, anomalies, are a realisers of the law disproving; they are themselves the
cases found in alleged conclusion not followed by presupposed effects, or one in those
where the effect is encountered with the absence of what is posited as a cause, remote or
proximate. So, despite their fruitfulness, value and utility scopes in suggesting general
causal laws, induction by enumeration is not at all suitable for their testing and verification.

25
 4 . 2 . 2 . FI VE J SM -M E TH OD S

• In their canonic form, five methods central to induction reasoning were comprehensively
postulated by British philosopher John Stuart Mill in A System of Logic. Today they are, as
a century ago, known among the logicians as ‘Mill’s methods of induction’. Mill’s naming
and formulations retained, although their estimation and uses has changed. Many of
contemporary discoveries in physics, biology, geology, social sciences commonly owe the
methodology they deployed for these discoveries as being one or another variant or
combination of these methods of inductive inference: the method of agreement; the
method of difference; the joint method of agreement and difference; the method of
residues; the method of concomitant variation.

26
‘
’
 4 . 2 . 2 .1. M E T H O D O F AG R E E M E N T

• Method of agreement is a pattern of inductive inference in which it is concluded that, if two

or more instances of a given phenomenon have only one condition in common, that one
common condition is the cause (or effect) of the phenomena. From this follows that the
method seeks not only for discovery of the repeated causal nexus, but also to fix the only
condition, the sole condition, that is invariably associated with the effect / phenomenon
which is of interest.

27
 4 . 2 . 2 .1. M E T H O D O F AG R E E M E N T
• Semi-formal postulation:

circumstances A, B, C and D trigger occurrences a, b, c and d [((A ⋀ B) ⋀ (C ⋀ D)) → ((a ⋀ b) ⋀ (c ⋀ d))]

circumstances D, E, F and G trigger occurrences d, e, f and g [((D ⋀ E) ⋀ (F ⋀ G)) → ((d ⋀ e) ⋀ (f ⋀ g))]
⊢ D is the cause of d [D d]

• The method is particularly useful in identifying a kind of phenomenon or a range (scope) of

conditions and circumstances, the investigation of which is a subject of scientific research; the second
occurs when / if one finds a single condition which is common to all the observed and explored
instances of a phenomenon (at least, in that case the region is considered to be located
approximately).

28
4 . 2 . 2 . 2 . M E T H O D O F D I FFE R E NC E
• Method of difference is a pattern of inductive reasoning in which, in cases where a given
phenomenon occurs differ in only one condition, circumstance of aspect from the cases
where the phenomenon is absent, that condition / circumstance is inferred to have causal
connection to the phenomenon. Hence, the pattern focuses not on the common cause, so
to say, but on what is different of essentials between those cases where the effect is
produced and those in which it is not.

• Wherever it is possible to identify and highlight a singular conditioning factor that creates
the critical difference on the effect (with everything else remaining the same as ‘normal’
conditions — elimination factor for the phenomenon under investigation, if the factor is
removed, or, conversely, the factor which produces the phenomenon after its re-
introduction (inputting) — the cause is then have been identified, or what is identified is at
least what John Stuart Mill calls indispensable part of the cause (a part of compound entity
which is the cause of some effect that cannot be neglected, omitted or not realised in
order for effect to take place) of the phenomenon-effect.

29
4 . 2 . 2 . 2 . M E T H O D O F D I FFE R E NC E

• Semi-formal notation of the method is schematised as follows:

((A ⋀ B) ⋀ (C ⋀ D)) ((a ⋀ b) ⋀ (c ⋀ d))

B ⋀ (C ⋀ D) b ⋀ (c ⋀ d)
⊢ (A a) ∨ (((¬A ⋀ B) ⋀ (C ⋀ D)) ¬a) (we read it as ‘therefore, either A is itself a cause of
a, or A is an indispensable part of the cause of a’, i.e., given all the other potential causes
for a, given the absence of A, a disappears from the output)

30
4 . 2 . 2 . 3 . J O I N T M E T H O D [ O F AG R E E M E N T & D I FFE R E NC E ]

• Joint method of agreement and difference is an inductive pattern where the methods of
agreement and difference are used simultaneously as a means of attempt of increasing the
probability of conclusion. Mill himself viewed it as an independent method in itself, yet
today it is clearly shown to be merely auxiliary technique, an algorithm of use rather than a
method autonomous and self-containing. This may be achieved because such use renders
a disjunction at the output (whereas conjunction of probabilities makes the whole
compound claim on probability of event as series of events less probable, the disjoining of
a compound event increases the probability of only one of them to occur).

31
4 . 2 . 2 . 3 . J O I N T M E T H O D [ O F AG R E E M E N T & D I FFE R E NC E ]

• Semi-formal notation is like this:

Let (A) be the use of method on agreement on the case C, and (D) — that of difference
on the same case C with the same data. Then, we have:

(A): ((A ⋀ (B ⋀ C)) (a ⋀ (b ⋀ c))) ⋀ ((C ⋀ (D ⋀ E)) (c ⋀ (d ⋀ e)));

(D): ((A ⋀ (B ⋀ C)) (a ⋀ (b ⋀ c))) ⋀ ((D ⋀ E) (d ⋀ e));
⊢ (A a) ∨ ((¬A ⋀ (B ⋀ C)) ¬a)

32
 4.2.2.4. METHOD OF RESIDUES

• Method of residues is an inductive pattern where, when particular parts or portions of an

investigated occurrence of phenomenon / event are known to be the effects of certain
conditions identified previously, it is plausible to conclude that the remaining portion of the
occurrence resulted from the remaining conditions. To put it simply, having known all the
causal nexuses within a compound causal chain, deduce all them to get one-to-one, n-to-
one or one-to-m correspondence which is the causal nexus under investigation, or the
cause in search or both.

33
 4.2.2.4. METHOD OF RESIDUES

• It differs from the previous three methods in that it can be used to investigate one case,
whereas others would need at least two cases. It also heavily relies on the established
ancedently general causal laws from which the previous methods are independent.

• In reliance thus way on, what it actually relies on is ‘common truths’ on which deductive
reasoning relies as well (a subset of truths discharged in deductive arguments); however, it
is not a deductive, but an inductive method due to its conclusions: it generates only
probable conclusions that cannot be validly deduced from premises.

34
 4.2.2.4. METHOD OF RESIDUES

• Semi-formal notation of the method may be like this:

(A ⋀ (B ⋀ C)) (a ⋀ (b ⋀ c))
(B → b) ⋀ (C → c)
⊢A→a

35
 • Until now, we dealt with instances of methods known as inferences derived by eliminative
induction. Let’s summarise them in this respect in a short table:

METHOD ELIMINATED CONTENT / ENTITY

[of] Agreement Conditions without which the occurrence can nevertheless take place.

[of] Difference [Some of] the possible causes by removing antecedent factor proven to be crucial.

Joint Both (1) and (2)

[of] Residues Conditions and causes, the effect of which has been previously inductively conducted.
4 . 2 . 2 . 5 . M E T H O D O F CO NCO M I TA N T VA R I AT I O N

• … the thing is, that, in many situations, none of these methods is applicable: there may
simply be nothing to eliminate in principle — from medicine to social sciences. The last
inductive method is one of the responses to this methodological complication.

• Method of concomitant variation is an inductive pattern in which it is concluded that, when

one phenomenon varies consistently with some other phenomenon in equivalent or
identical manner (co-absence, co-emergence), then there is at least causal correlation
between them or, at most, a full-blood causal nexus. The method identifies this connection
between two conditions by matching variations in one condition with variations in another.
According to one formulation, increases are matched with increases (co-emergence) and
decreases with decreases (co-absence).

37
4 . 2 . 2 . 5 . M E T H O D O F CO NCO M I TA N T VA R I AT I O N
• When the increase in degree, emergence or presence of one phenomenon (in
antecedents) parallels the increase in degree, emergence or presence of another, the
phenomena vary directly with regard each other.

• In other subset of cases, we may have that increase in degree, emergence or presence of
the antecedent leads to decrease in degree, gradual or abrupt disappearance or absence.
In that case, it is asserted that we have the case of causal nexus in which cause and effect
vary inversely.

• While the previous four methods were dealing with qualitative aspects, in all respects,
method of concomitant variation is essentially a quantitative one, because it enables to
adduce changes in the degrees of conditions and phenomena presence, as evidences.

• Due to the latter, its use presupposes having at hand methods of estimation, valuation and
measurement, scaled and framed, appropriate to what is explored.

38
 4 . 2 . 2 . 5 . M E T H O D O F CO NCO M I TA N T VA R I AT I O N
• Semi-formal notation of the method of concomitant variation is made for two principally distinct
variations and syntax extension (represented by two binary operators ‘✚’ standing for ‘added to
premises / increasing’ and ‘⊖’ for ‘subtracted from premises / decreasing):

• For Concomitant variation for direct ‘cause — effect’ variation:

(A ⋀ (B ⋀ C)) (a ⋀ (b ⋀ c))
((A ⋀ B) ✚ C)) ((a ⋀ b) ✚ c) (given that they were absent when occurrence was triggered.)
((A ⋀ B) ⊖ C) ((a ⋀ b) ⊖ c) (given that they were present when occurrence was triggered.)
⊢ C is causally related (connected) to c.

• For Concomitant variation for inverse ‘cause — effect’ variation:

(A ⋀ (B ⋀ C)) (a ⋀ (b ⋀ c))
((A ⋀ B) ✚ C)) ((a ⋀ b) ⊖ c) (given that C was added / started increasing, and c was present.)
((A ⋀ B) ⊖ C) ((a ⋀ b) ✚ c) (given that c was absent / emerged / started increasing, and C absent /
started decreasing, etc.)
⊢ C is causally related (connected) to c.

39
5 . FI N A L R E M A R K S : O N S O M E L I M I T S O F I N D U C T I VE R E A S O N I N G
• John Stuart Mill saw inductive reasoning as a collection of instruments which would allow
us to discover laws of nature in forms of causal connections, as well as being canons with
which causal connections can be proved. In these postulates, he overestimated their
capabilities and significance, ignoring some of its constraints and setbacks.

1. In postulating the methods, Mill presumed that one can identify cases by either “having
only one circumstance in common”, or “having every circumstance in common save one”.
But this should not be understood literally: any two objects may have many common
predicates of various nature and scope, and no pair of entities would differ in only one
respect.

2. In no way all the possible instances could ever he examined. What the scientists have in
mind is not all the cases, but the sets or collections of relevant circumstances and entities,
sampled by particular principle (randomised, classified, categorised, taxonomy etc.).
Application of Mill’s induction requires possessing tacit knowledge (:= they alone are
neither sufficient, nor leading or basic in the process of scientific discovery).

40
5 . FI N A L R E M A R K S : O N S O M E L I M I T S O F I N D U C T I VE R E A S O N I N G
3. The methods themselves are also unable constitute rules for proof. Their application is
based on hypotheses (as antecedents) about causal factors and conditions, and since
conditions and circumstances cannot be quantified to ‘∀x’ extent in principle, as was said
in 1.

4. The application of the inductive methodology depends on observed correlations which,

given that they are true, sound and accurate, may still be incomplete, therefore being
deceptive in one way or another. The greater the number, the higher is the likelihood that
we indeed deal with some general causal law, but no matter how many of them are at
hand, one still cannot legitimately infer a causal connection for the instances which haven’t
been observed.

5. A valid argument following deductive inferential pattern constitutes a proof, or

demonstration, but for inductive inferential patterns, as it was asserted instantly, the case is
only high probability of the conclusion content, at best; and, even though, inductive
argument is never demonstrative (in technical sense, as deductive).

41
5 . FI N A L R E M A R K S : O N S O M E L I M I T S O F I N D U C T I VE R E A S O N I N G

6. ⊢ Given 1.-5., Mill’s claims about his canons as ‘proof methods’ or ‘methods of discovery’
must be rejected.

7. Despite that, inductive reasoning plays great role in scientific research today, and many
very powerful arguments, explanative hypotheses and theories are construed with their
use. Due to the uncertainty with circumstances quantification, the application of methods
must always suppose causal hypotheses concerning the research subject. These
hypotheses are themselves follow abductive pattern of reasoning, nonmonotonic and
revisable; they also compete with one another for the right to be a theory accepted by the
community of scientists. Each such hypothesis and their array are the subjects of testing, as
the former, and comparative (meta-)analysis, as the latter. Mill’s methods enable us to do
the following thing in this respect: IF the analysis of antecedent conditions is true, one of
them can’t be (or, conversely, must be) the whole cause or its part, for the effect
undergoing the research.

42
5 . FI N A L R E M A R K S : O N S O M E L I M I T S O F I N D U C T I VE R E A S O N I N G

8. Inductive methods can give reliable results if and only if the hypothesis to which they
appeal correctly identifies relevant circumstances, and the deduction of results is possible
when the hypothesis has been assumed as a premise of the argument.

So, perhaps, a closure on methods would be just as simple as this. Inductive methods aren’t
paths of discovery or proof. They are cognitive instruments for testing hypotheses. Taken
jointly, the statements and postulates of inductive methods, describe the generalised and
abstracted version of controlled experiment, one of the most vital and indeed indispensable
tools of contemporary science, especially in our century with new modalities of its conduction,
such as simulation and modelling with AI.

43