Solved the Maze?

Start at phils house. At first, you can only make right turns through the maze. Each
time you cross the red zigzag sign (under arls auto repair!, the direction in "hich
you turn changes. So, after the first time you cross that sign, you can then only
make left turns# after the second time, you s"itch $ack to right turns only, etc. %o"
can arls auto repair $e reached?
&nductive 'easoning
(
%ypothesis )esting
*ogical 'easoning and %uman +ature
, %istorically, many researchers $elieved that logical reasoning is an
essential part of human nature
- Aristotle
- .oole (/012!. A $ook on logical calculus
3An investigation of the la"s of thought4
,
'ational $ehavior 5 logical thinking
(deductive reasoning!
,
%o"ever, humans are not natural logical reasoners
6eductively valid?
7remise8 All cars have "heels
7remise8 All "heels are round
onclusion8 All cars have round "heels
99
7remise8 & have a diamond
7remise8 Most diamonds are shiny
onclusion8 My diamond is shiny
999
7remise8 :ohn is ;<
onclusion8 :ohn "ill not do a dou$le $ack flip today
&nductive vs. 6eductive 'easoning
, 6eductive reasoning8
- conclusion follo"s logically from premises
,
&nductive reasoning8
- conclusion is likely $ased on premises.
- involves a degree of uncertainty
,
'easoning in real9"orld is often $ased on induction
&nductive 'easoning
,
'eason from o$serva$le information to uno$serva$le and uncertain
information
,
)he =oogle sets program is an e>ample of inductive reasoning8
http8??la$s.google.com?sets
'eal "orld inductive inferences
,
Medical diagnosis8
- Symptoms, test outcomes (o$serva$le! 6iseases (uno$serva$le!
,
Scientific reasoning8
- E>perimental data (o$serva$le! %ypotheses (uno$serva$le!
,
*a"8
- facts (o$serva$le! guilt (uno$serva$le, uncertain!
'easoning under @ncertainty
,
.ayes rule tells us ho" to optimally reason "ith uncertainty.
, Allo"s us to say ho" "e $elieve something to $e true $ased on
prior $eliefs and ne" availa$le evidence
)homas .ayes (/ABC9/AD/!
.ayes 'ule
7rior pro$a$ility
Evidence
7osterior 7ro$a$ility
Bayes rule tells us ho" the availa$le evidence
should alter our $elief in something $eing true
6o people reason like .ayes rule?
,
7ro$lems understanding conditional probability
- 6octors need to calculate the pro$a$ility of disease given the o$served
symptoms8 7( disease E symptoms !
- Sometimes 7( symptoms E disease ! is used incorrectly "hen reasoning
a$out the likelihood of a disease
, Fhy is this "rong?
,
)o get 7( disease E symptom !, you need to kno" a$out 7( symptom
E disease ! and also the base rate 99 prevalence of the disease
$efore you have seen patient
,
More intuitive e>ample8
- "hat is the pro$a$ility of $eing tall given you are player in the +.A?
- "hat is the pro$a$ility of $eing a player in the +.A given that you are
tall?
7( +.A player E tall ! G 7( tall E +.A player !
)he $ase rate is important
'easoning "ith $ase rates
,
Suppose there is a disease that affects
/ out of /BB people
,
)here is a diagnostic test "ith the follo"ing properties8
- &f the person has the disease, the test "ill $e positive ;0H of the time
- if the person does not have the disease, the test "ill $e positive /H of
the time
,
A person tests positive, "hat is the pro$a$ility that this person has
the disease?
- IreJuent ans"er 5 .;0
- orrect ans"er K .1B
Are "e really that $ad in Ludging pro$a$ilities?
According to some researchers (e.g., =igerenzer!, it
matters ho" the information is presented and processed.
7rocessing freJuencies is more intuitive than pro$a$ilities
A counting heuristic (in tree form!
/B,BBB people
/BB have disease
;,;BB do not
;0 test positive
C test negative
;; test positive
;0B/ test negative
7( disease E test positive ! 5 ;0 ? ( ;0 M ;; ! K .1B
)he same thing in "ords ...
,
*ets take /B,BBB people.
,
Nn average, /BB out of /B,BBB actually have the disease and ;0 of
those "ill test positive (;0H true positive rate!
,
Among the ;,;BB "ho do not have the disease, the test "ill falsely
identify /H as having it. /H of ;,;BB 5 ;;
,
Nn average, out of /B,BBB people8
;0 test positive and they have the disease
;; test positive and they do not have the disease.
,
)herefore, a positive test outcome implies a ;0?(;0M;;!K1BH chance
of having the disease
hange the e>ample
, Fhat no" if the disease affects only
/ out of /B,BBB people?
, Assume same diagnosticity of test
(;0H true positive rate, /H false positive rate!
, A person tests positive, "hat no" is the pro$a$ility that this person
has the disease?
A counting heuristic (in tree form!
/,BBB,BBB people
/BB have disease
;;;,;BB do not have the disease
;0 test positive
C test negative ;;;; test positive ;0;;B/ test negative
7( disease E test positive ! 5 ;0 ? ( ;0 M ;;;; ! 5 .BB;A (smaller than /H!
.ayes 'ule
, )he previous e>ample essentially is a simple "ay to apply .ayes
rule8
( )
( ) ( )
| ( )
( | )
| ( ) | ( )
P positive disease P disease
P disease positive
P positive disease P disease P positive not disease P not disease
=
+
( ) | .98 P positive disease =
( ) | .01 P positive not disease =
( ) .0001 P disease =
( ) | .0097 P disease positive =
+ormative Model
,
.ayes rule tells you ho" you should reason "ith pro$a$ilities - it is
a prescriptive (i.e., normative! model
,
.ut do people reason like .ayes?
&n certain circumstances, "e o$serve base rate neglect
)he )a>i 7ro$lem8 version /
,
A "itness sees a crime involving a ta>i in ar$orough. )he "itness
says that the ta>i is blue. &t is kno"n from previous research that
"itnesses are correct 80% of the time "hen making such statements.
,
Fhat is the pro$a$ility that a $lue ta>i "as involved in the crime?
Oahneman ( )versky (/;A<!
)he )a>i 7ro$lem8 version C
,
A "itness sees a crime involving a ta>i in ar$orough. )he "itness
says that the ta>i is blue. &t is kno"n from previous research that
"itnesses are correct 80% of the time "hen making such statements.
,
)he police also kno" that /1H of the ta>is in ar$orough are $lue,
the other 01H $eing green.
,
Fhat is the pro$a$ility that a $lue ta>i "as involved in the crime?
.ase 'ate +eglect8 )he )a>i 7ro$lem
,
Iailure to take prior pro$a$ilities (i.e., $ase rates! into account
,
&n the ta>i story, the addition of8
3)he police also kno" that /1H of the ta>is in ar$orough are $lue,
the other 01H $eing green.4
has little influence on rated pro$a$ility
.ase 'ate +eglect (C!
,
Oahneman ( )versky (/;A<!. 7laced cards "ith names of people and their
occupations in a $o"l

.o"l A8 AB engineers and <B la"yers
.o"l .8 <B engineers and AB la"yers
,
Fhat is pro$a$ility of picking an engineer from $o"l A and .? Su$Lects
can do this P
7rovide some evidence P
,
3:ack is a 21 year9old man. %e is married and has four children. %e is
generally conservative, careful, and am$itious. %e sho"s no interest in
political and social issues and spends most of his free time on his many
ho$$ies, "hich include home carpentry, sailing, and mathematical
puzzles4
,
Fhat no" is pro$a$ility :ack is an engineer?
,
Estimates for $oth $o"l A and . "as 7 5 .;
%ypothesis )esting
Fason Selection )ask
3&f a card has a vo"el on one side, then it has an even num$er on the
other side4
Fhich cards do you need to turn over to test the correctness of the rule?
E O 2 A
oncrete e>amples are much easier
,
&f a person is drinking $eer, then the person must $e over C/. %o"
to test "hether some$ody is a$iding $y this rule?
6rinking
$eer
6rinking
oke
/D years
of age
CC years
of age
'esult8 A2H ans"ered correctly
onclusion from Fason Selection )ask
,
Irom a pure deductive point of vie", su$Lects often fail to reason
appropriately "ith a$stract pro$lems
,
%o"ever, from an inductive point of vie", su$Lects choices are Juite
reasona$le under certain assumptions:
- Rules such as If Cause then Effect are interpreted probabilistically
- Causes are rare
- Effects are rare
(Naksford ( hater!
%ypothesis )esting
, C9290 is a set of num$ers that conforms to a rule.
,
6iscover the rule $y Juerying "ith any set of three num$ers and &ll
give feed$ack "hether it is a positive or negative e>ample.
(task from Fason, /;DB!
onfirmation .ias
,
Fason (/;DB!8 su$Lects test hypotheses $y generating positive
rather than negative e>amples
,
7opper (/;1;!8 confirmatory strategies provide am$iguous
information. )he hypothesis may $e correct or another hypothesis
may $e correct scientists should try to falsify their theories
(%o"ever, in many cases, it might make more sense to confirm
hypotheses, and not to attempt falsification!
)hank Qou