The Grice Club


The Grice Club

The club for all those whose members have no (other) club.

Is Grice the greatest philosopher that ever lived?

Search This Blog

Wednesday, February 24, 2016

Griceian Aspects of Reason -- and Reasoning!


Grice knew how to lecture.

When invited to deliver the Kant lectures across the bay, at Stanford, he chose the right topic: "Aspects of reason and reasoning."

When invited to deliver the Locke lectures across the pond, at Oxford, he chose the right topic -- for surely Locke predates Kant and understands English idioms like 'to reason' better: "Aspects of reason and reasoning".

Reasoning is related to problem solving, because people trying to solve a 
reasoning task have a definite goal and the solution is not obvious.
However,  problem solving and reasoning are typically treated separately. Reasoning
problems differ from other kinds of problems in that they often owe their 
origins to systems of formal logic, as symbolised by Frege and 'laughed' 
metaphorically by Grice in "Logic and Conversation".

There are clear overlaps between the two areas, which may differ less than 
one might initially suppose. Inductive reasoning involves making a
generalised  conclusion from premises referring to particular instances. Hypotheses
can never  be shown to be logically true by simply generalising from
confirming instances  (i.e., induction). Generalisations provide no certainty for
future events.  Deductive reasoning allows us to draw conclusions that are
definitely valid  provided that other statements are assumed to be true. For
example, if we assume

i. Grice is taller than Popper
ii. Popper is taller than Johnson-Laird
the conclusion

iii. Grice is taller than Johnson-Laird.

is necessarily true.

It is well known that Popper argues for a distinction between confirmation 
and falsification. Confirmation involves obtaining evidence to confirm the 
correctness of one’s hypothesis. Falsification involves attempting to
falsify  hypotheses by experimental tests.
Popper argued that it is impossible to  achieve confirmation via hypothesis
testing. Rather, scientists should focus on  falsification.

When Johnson-Laird and Wason devised their tests, such as "the 2–4–6
task",  in which participants have to discover a relational rule underlying a set
of  three numbers, they found performance was poor on the task because
people tended  to show confirmation bias – they generated numbers conforming to
their original  hypothesis rather than trying hypothesis disconfirmation. A
positive test is  when numbers produced are an instance of your hypothesis.
A negative test is  when numbers produced do not conform to your hypothesis.

Wason’s theoretical position predicts that people should perform better 
when instructed to engage in disconfirmatory testing.

The evidence was mixed.

Cowley and Byrne argue that people show confirmation bias because they are 
loath to abandon their own initial hypothesis.

Tweney finds that performance on the 2–4–6 task was enhanced when 
participants were told to discover two rules, one the complement of the other.

Gale and Ball argue that it was important for participants to identify the 
crucial dimensions of ascending vs. descending numbers.

Performance on the Johnson-Laird's and Wason's 2–4–6 task involves 
separable processes of:
 hypothesis generation;
 hypothesis  testing.

Cherubini (not the author of "Medea") argues that participants try to 
preserve as much of the information contained in the example triple (i.e.,  2–4–
6) as possible in their initial hypothesis. As a result, this hypothesis is
typically much
more specific than the correct rule. Most hypotheses are  sparse or narrow
in that they apply to less than half the possible entities in  any given
domain (vide Navarro & Perfors).

The 2–4–6 problem is a valuable source of information about inductive 
reasoning. The findings from the 2–4–6 task may not be generalisable because,
in  the real world, positive testing is not penalised. Additional factors
come  into play in the real world. Hypothesis testing in simulated and
real-world  settings.

There is a popular view that "scientific discovery is the result of genius,
inspiration, and sudden insight" (Trickett & Trafton).

That view is largely incorrect.

Scientists (that Grice never revere -- 'we philosophers are into 
hypostasis; while mere scientists can only grasp hypothesis') typically use what 
Klahr and Simon describe as weak methods.

Kulkarni and Simon found scientists make extensive use of the unusualness 
heuristic, or rule of thumb.

This involves focusing on unusual or unexpected findings and then using 
them to guide future theorising and research.

Trickett and Trafton argue that scientists make much use of “what if” 
reasoning in which they work out what would happen in various imaginary 

Dunbar uses a simulated research environment. He found that  participants
who simply tried to find data consistent with their hypothesis  failed to
solve the problem.

It is believed that scientists should focus on falsifying their hypotheses.
However this does not tend to happen.

Nearly all reasoning in everyday life is inductive rather than deductive. 
Hypothesis testing is a form of inductive reasoning.

It is well known that Popper argued that it is impossible to confirm a 
hypothesis via hypothesis testing. Rather, scientists should focus on 
falsification. However, it is now accepted that Popper’s views were  oversimplified
and confirmation is often appropriate in real scientific  research.

When Johnson-Laird and Wason devised stuff like the 2–4–6 task,  they
found people tended to show confirmation bias, producing sequences  that
confirmed their hypotheses rather than seeking negative evidence. However,  later
studies demonstrated that people’s behaviour is often more accurately 
described as confirmatory or positive testing.

"What if", or conditional reasoning is basically reasoning with “if”.

It has been studied to decide if human reasoning is logical.

In propositional logic, meanings are different from those in natural 

There are different types of logical reasoning statements:

"Affirmation of the consequent":
Premise (if P then Q), (Q), Conclusion  (P).
Invalid form of argument.

"Denial of the antecedent":
Premise (if P then Q), (not P), Conclusion  (not P).
Invalid form of argument.

"Modus tollens":
Premise (If P, then Q), (not Q), Conclusion (not  P).
Valid form of argument.

"Modus ponens"
Premise (If P, the Q), (P), Conclusion (Q).
Valid form  of argument.

Invalid inferences (denial of the consequent, affirmation of  the
consequent) are accepted much of the time, the former typically more often  (Evans).

De Neys finds evidence that conditional reasoning is strongly  influenced
by the availability of knowledge in the form of counterexamples  appearing to
invalidate a given conclusion. He also found performance on  conditional
reasoning tasks depends on
individual differences.

Bonnefon argues that reasoners draw inferences when presented with 
conditional reasoning problems.

According to Markovits, there are two strategies people can use with 
problems: a statistical strategy and a counterexample strategy.

Various findings suggest many people fail to think logically on conditional
reasoning tasks.

Conditional reasoning is closer to decision making than to classical logic 

The Johnson-Laird/Wason selection task has four cards, each with a  number
on one side and a letter on the other.

are told a rule and asked to select only those cards that  must be turned
over to decide if the rule is correct.

The correct answer is only given by 5–10% of those who are engaged in the 

Many attempts have been made to account for performance on this task.

Evans identifies matching bias as an important factor.

This is the tendency for participants to select cards matching items named 
in the rule regardless of whether the matched items are correct.

Stenning and van Lambalgen argue that people have difficulties interpreting
precisely what the selection problem is all about.

Oaksford argues that the logical answer to the Johnson-Laird/Wason 
selection task conflicts with what typically makes most sense in everyday  life.

Performance in the Johnson-Laird/Wason selection task can be improved by 
making the underlying structure of the problem more explicit (Girotto) or by 
motivating participants to disprove the rule (Dawson).

A syllogism consists of two premises or statements followed by a 
conclusion. The validity of the conclusion depends only on whether it follows 
logically from the premises. Belief bias is when people accept  believable
conclusions and reject unbelievable conclusions, irrespective of  their
logical validity or invalidity. Klauer finds various biases in  syllogistic
reasoning, including a base-rate effect, in which performance is  influenced by
the perceived probability of syllogisms being valid. Stupple and  Ball find
with syllogistic reasoning that people took longer to process  unbelievable
premises than believable ones. Stupple finds participants were more  likely
to accept conclusions that matched the premises in surface features  than
those not matching.

Conditional reasoning has its origins in a system of logic known as 
propositional logic. Performance on conditional reasoning problems is typically 
better for the modus ponens inference than for other inferences (e.g., modus 
tollens). Conditional reasoning is influenced by context effects (e.g., the
inclusion of additional premises). Performance on the Wason selection task
is  generally very poor, but is markedly better when the rule is deontic
rather than  indicative. Performance on syllogistic reasoning tasks is
affected by various  biases, including belief bias and the base rate. The fact that
performance on  deductive reasoning tasks is prone to error and bias
suggests people often fail  to reason logically.

The mental models approach is one of the most influential approaches and 
was proposed by Johnson-Laird.

A mental model represents a possibility, capturing what is common to the 
different ways in which the possibility could occur. People use the
information  contained in the premises to construct a mental model.
Here are the main  assumptions of mental model theory. A mental model
describing the given  situation is constructed and the conclusions that follow
are generated. The  model is iconic (its structure corresponds to what it
represents). An attempt is  made to construct alternative models to falsify the
conclusion by finding  counterexamples to the conclusion. If a
counterexample model is not found, the  conclusion is assumed to be valid.

The construction of mental models involves the limited resources of working
memory. Reasoning problems requiring the construction of several mental
models  are harder to solve than those requiring only one mental model because
of  increased demands on working memory. The principle of truth states that
individuals minimise the load on working memory by tending to construct
mental  models that represent explicitly only what is true, and not what is
false  (Johnson-Laird).

Successful thinking results from the use of appropriate mental models. 
Unsuccessful thinking occurs when we use inappropriate mental models.
Knauff  finds deductive reasoning was slower when it involved visual
imagery. Copeland  and Radvansky test this assumption. They find a moderate
correlation between  working memory capacity and syllogistic reasoning. They also
found that problems  requiring more mental models had longer response times.
Legrenzi tested the  principle of truth. He found performance was high on
problems when  adherence to the principle of truth was sufficient. In
contrast, there were  illusory inferences when the principle of truth did not
permit correct  inferences to be drawn. People are less susceptible to such
inferences if  explicitly instructed to falsify the premises of reasoning
problems (Newsome  & Johnson-Laird).

Newstead, Eysenck, and Keane also found participants consistently failed to
produce more mental models for multiple-model syllogisms than for
single-model  ones.

Most predictions of mental model theory have been confirmed experimentally.
In particular, evidence shows that people make errors by using the
principle of  truth and ignoring what is false. Limitations with the theory are
that it  assumes that people engage in deductive reasoning to a greater extent
than is  actually the case. The processes involved in forming mental models
are  underspecified.

There are two process involved in human reasoning. One system involves 
unconscious processes and parallel processing, and is independent of 
intelligence. The other system involves conscious processes and rule-based  serial
processing, has limited capacity and is linked to intelligence. Evans 
proposes the heuristic–analytic theory of reasoning, which distinguishes between 
heuristic processes (System 1) and analytic processes (System 2). Initially, 
heuristic processes use task features and
knowledge to construct a single  mental model. Later, effortful analytic
processes may intervene to revise this  model. This is more likely when task
instructions tell participants to use  abstract or logical reasoning;
participants are highly intelligent; sufficient  time is available for effortful
analytic processing; or participants need to  justify their reasoning.
Human reasoning is based on the use of three  principles: the Singularity
principle, the Relevance principle (not to be  confused with Grice's
conversational category of Relatio, after Kant), and the  Satisficing principle.
In contrast to the mental model theory, the  heuristic–analytic theory
predicts that people initially use their world  knowledge and immediate context
in reasoning. Deductive reasoning is regarded as  less important.
Belief bias is a useful phenomenon for distinguishing between  heuristic
and analytic processes. Evans finds less evidence of belief bias when 
instructions emphasised logical reasoning. Stupple compares groups of  participants
who showed much evidence of belief bias and those showing little  belief
bias. Those with high levels of belief bias responded faster on  syllogistic
reasoning problems. De Neys finds high working memory capacity  was an
advantage only on problems requiring the use of analytic processes. A  secondary
task impaired performance only on problems requiring analytic  processes.
Evans and Curtis-Holmes find belief bias was stronger when time  was strictly
Thompson suggests two processes are used in syllogistic  reasoning:
Participants provided an intuitive answer. This is followed by  an assessment of
that answer’s correctness (feeling of rightness). After which,  participants
have unlimited time to reconsider their initial answer and provide  a final
answer (analytic or deliberate answer). Thompson argues that we possess  a
monitoring system (assessed by the feeling-of-rightness ratings) that 
evaluates the output of heuristic or intuitive processes. Evidence that people  are
more responsive to the logical structure of reasoning problems than 
suggested by performance accuracy was reported by De Neys. The  heuristic–analytic
theory of reasoning has several successes: the notion that  cognitive
processes used by individuals to solve reasoning problems are the same  as those
used in other cognitive tasks seems correct. Evidence supports the  notion
that thinking is based on singularity, relevance and satisficing  principles.
There is convincing evidence for the distinction between heuristic  and
analytic processes. The theory accounts for individual differences, for 
example in working memory capacity.

Limitations with the approach are that it is an oversimplification to 
distinguish between implicit heuristic and explicit analytic processes. Also, 
the distinction between heuristic and analytic processes poses the problem of
working out exactly how
these two different kinds of processes interact. It  is not very clear
precisely what the analytic processes are or how individuals  decide which ones
to use. Logical processing can involve heuristic or intuitive  processes
occurring below the conscious level.

The assumption that heuristic processing is followed by analytic processing
in a serial fashion may not be entirely correct.

According to mental model theory, people construct one or more mental 
models, mainly representing explicitly what is true. Mental model theory fails 
to specify in detail how the initial mental models are constructed, and
people  often form fewer mental models than expected. Dual-system theories
answer the  two main limitations of most other research into human reasoning
because they  take account of individual differences in performance and
processes. There is  now convincing evidence for a distinction between relatively
automatic,  heuristic-based processes and more effortful analytic-based
processes. However,  it is unlikely that we can capture all the richness of human
reasoning simply by  assuming the existence of two cognitive systems.
 Prado finds the brain  system for deductive reasoning is centred in the
left hemisphere involving  frontal and parietal areas. Specific brain areas
activated during deductive  reasoning included: inferior frontal gyrus;
middle frontal gyrus; medial  frontal gyrus; precentral gyrus; basal ganglia.
Goel studies patients having  damage to left or right parietal cortex. Those
with left-side damage perform  worse than those with right-side damage on
reasoning tasks in which complete  information is provided. Prado finds the
precise brain areas associated  with deductive reasoning depended to a large
extent on the nature of the task.  Prado also finds that the left inferior
frontal gyrus (BA9/44) is more  activated during the processing of categorical
arguments. Prado finds found  the left precentral gyrus (BA6) was more
activated with propositional reasoning  than with categorical or relational

Language seems to play little or no role in processing of reasoning tasks 
post-reading (Monti & Osherson). Reverberi identifies three strategies used 
in categorical reasoning: sensitivity to the logical form of problems (the
left  inferior lateral frontal (BA44/45) and superior medial frontal (BA6/8)
areas);  sensitivity to the validity of conclusions (i.e., accurate
performance) -- the  left ventro-lateral frontal (BA47) area, use of heuristic
strategies, no  specific pattern of brain activation. More intelligent
individuals exhibit less  belief bias because they make more use of analytic
processing strategies (De  Neys). Individual differences in performance accuracy
(and thus low belief bias)  were strongly associated with activation in the
right inferior frontal cortex  under low and high cognitive load conditions
(Tsujii &  Watanabe).
Fangmeier uses mental model theory as the basis for assuming the  existence
of three stages of processing in relational reasoning. Different brain 
areas were associated with each stage: Premise processing: temporo-occipital 
activation reflecting the use of visuo-spatial processing. Then there's
Premise  integration: anterior prefrontal cortex (e.g., BA10), an area associated
with  executive processing. Finally, there is Validation: the posterior
parietal  cortex was activated, as were areas within the prefrontal cortex
(BA6, BA8) and  the dorsal cingulate cortex.
Bonnefond studies the brain processes associated  with conditional
reasoning focusing on modus ponens: There is enhanced brain  activity when premises
and conclusions do not match and anticipatory processing  before the second
premise occurs when they match. Limited progress has been made  in
identifying the brain systems involved in deductive reasoning. This is  because of
simple task differences and individual differences that affect the  results.

Informal reasoning is a form of reasoning based on one’s knowledge and 
experience. People make extensive use of informal reasoning processes such as 
heuristics in formal deductive reasoning tasks. However, there are also 
differences between processes in formal and informal reasoning: content; 
contextual factors; informal reasoning concerns probabilities; and  motivation.
Ricco identifies common informal fallacies: Irrelevance (seeking  to
support a claim with an irrelevant reason); Slippery slope.The myside bias is  the
tendency to evaluate statements with respect to one’s own beliefs rather 
than solely on
their merits (Stanovich & West). Support for the  probabilistic approach
was reported by Hahn and Oaksford.They identify several  factors influencing
the perceived strength of a conclusion: degree of previous  conviction or
belief; positive arguments have more impact than negative  arguments; and
strength of the evidence.

Hahn and Oaksford find a Bayesian model predicted informal reasoning 
performance very well. However, Bowers and Davis argue that the Bayesian 
approach is too flexible and thus hard to falsify. Sá finds unsophisticated 
reasoning was more common among those of lower cognitive ability. Informal 
reasoning is more important in everyday life than deductive reasoning. However, 
most reasoning research is far removed from everyday life. Hahn and  Oaksford
propose a framework for research on informal reasoning based on 
probabilistic principles. There is reasonable support for their model,  particularly
for the role of prior belief and new evidence on strength of  argument. In
future, it will be important to establish the similarities and  differences
in processes underlying performance on informal and deductive  reasoning

Are humans rational? Much evidence seems to indicate that our thinking and 
reasoning are often inadequate, suggesting that we are not rational, even 
if Grice thought he was. Human performance on deductive reasoning tasks 
does seem very prone to error.
Most people cope well with problems in  everyday life, yet seem irrational
and illogical when given reasoning problems  in the laboratory. However, it
may well be that our everyday thinking is less  rational than we believe.
Heuristics allow us to make rapid, reasonably  accurate, judgements and
decisions, as Maule and Hodgkinson point out.  Laboratory research findings
suggest people can think rationally when problems  are presented in a readily
understandable form. Many of the apparent "errors" on  deductive reasoning tasks
may also be less serious than they seem. There is  reasonable support for
the notion that factors such as participants’  misinterpretation of problems,
or lack of motivation, explain only a fraction of  errors in thinking and
(e.g., Camerer & Hogarth). Individual  differences in intelligence and
working memory also influence performance on  conditional reasoning tasks. Some
researchers have found inadequacies in  performance even when steps are
taken to ensure that participants fully  understand the problem (e.g., Tversky &
Kahneman’s conjunction fallacy  study). Interestingly, those who are
incompetent have little insight into their  reasoning failures; this is the Dunning
–Kruger effect (Dunning). Deciding  whether humans are rational depends on
how we define “rationality”. Sternberg  points out that few problems of
consequence in our lives had a deductive or even  any meaningful kind of ‘
correct’ solution. Normativism “is the idea that human  thinking reflects a
normative system one conforming to norms or standards  against which it should
be measured and judged. (Elqayam & Evans).

An alternative  approach is that human rationality involves effective  use
of probabilities rather than logic. Oaksford and Chater put forward an 
influential probabilistic approach to human reasoning. Simon suggests the notion
of bounded rationality should be considered in human reasoning. This means
an  individual’s informal reasoning is rational if it achieves his/her goal
of  arguing persuasively. Many “errors” in human thinking are due to
limited  processing capacity rather than irrationality. Toplak reports a
correlation of  +0.32 between cognitive ability and performance across 15 judgement
and decision  tasks. Stanovich (2012) developed the tripartite model with
two levels of  processing: Type 1 processing (e.g., use of heuristics) within
the autonomous  mind is rapid and fairly automatic. Type 2 processing (also
called System 2  processing), which is slow and effortful.

There are three different reasons why individuals produce incorrect 
responses when confronted by problems: the individual lacks the mindware (e.g., 
rules, strategies) to override the heuristic response; or the individual has
the  necessary mindware but fails to realise the need to override the
heuristic  response; or the individual has the necessary mindware and realises
that the  heuristic response should be overridden, but doesn’t have sufficient
decoupling  capacity.

Stanovich uses the hybrid term (that scared Grice) dysrationalia to refer 
to "the inability to think and behave rationally despite having adequate 
intelligence". Most people (including those with high IQs) are cognitive
misers,  preferring to solve problems with fast, easy strategies than with more
accurate  effortful ones. Humans can be considered rational because errors
are caused by  limited processing capacity (Simon). Classical logic is almost
totally  irrelevant to our everyday lives because it deals in certainties.
Our thinking  and reasoning are rational when used to achieve our goals.
However, humans can  be considered irrational because many humans are cognitive
misers. There is a  widespread tendency on judgement tasks to de-emphasise
base-rate information.  They fail to think rationally because they are
unaware of limitations and errors  in their thinking. Apparently poor performance
by most people on deductive  reasoning tasks does not mean we are illogical
and irrational because of the  existence of the normative system problem,
the interpretation problem and the  external validity problem.

Yet, when Grandy and Warner decided for a festschrift for P. Grice, they 
came up with "Philosopical Grounds of Rationality: Intentions, Categories, 
Ends", but then it's an acronym: PGRICE ("and Clarendon didn't notice!") 

No comments:

Post a Comment