Quantum Models of Cognition and Decision

Catarina Moreira
Instituto SuperiorTécnico,Technical University of Lisbon / GAIPS
7th November 2012

 Motivation Example
 Classical Probability vs Quantum Probability
 Violations of Classical Probability
 Order of Effects
 Conjunction Errors
 The SureThing Principle
 The Double Slit Experiment
 Implications and Research Questions

CLASSICALTHEORY QUANTUMTHEORY
Andrei Kolmogorov (1933) John von Neumann (1932)

 Suppose you are a juror trying to judge
whether a defendant is Guilty or Innocent
 What beliefs do you experience?

 Classical Information Processing
 Single path trajectory. Jump between states.
 At each moment favors Guilty and another
moment favors Innocent
G GI I

 Quantum Information Processing
 Beliefs don’t jump between each other.They
are in a superposition!
G G
I I
G G
I I

 Quantum Information Processing
 We experience a feeling of ambiguity,
confusion or uncertainty about all states
simultaneously

Q1. Do you generally think that
Passos Coelho is honest and
trustworthy?
Q2. How about Ramalho Eanes?
Ramalho Eanes is honest and
trustworthy?
Q2. How about Passos Coelho?
A Gallup Poll question in 1997 (N = 1002, split sample)
(Politician’s names differ from the original work)
Moore, D.W. (2002). Measuring new types of question order effects. Public Opinion
Quaterly, 66, 80-91

trustworthy? (50%)
(60%)
trustworthy? (68%)
(57%)
Moore, D.W. (2002). Measuring new types of question order effects. PublicOpinion
Quaterly, 66, 80-91

trustworthy? (50%)
(60%)
trustworthy? (68%)
(57%)
Moore, D.W. (2002). Measuring new types of question order effects. PublicOpinion
Quaterly, 66, 80-91
18%
3 %

 Classical Probability cannot explain order of
effects, because events are represented as
sets and are commutative!
Sample Space Sample Space
A B
P( A ∩ B ) = P( B ∩ A )
B A

 Order of effects are responsible for
introducing uncertainty into a person’s
judgments.
Judgment 1
Judgment 2

 Events
 System State
 State Revision
 Compatible Events
 Incompatible Events (quantum only)

 Sample space (Ω). Contains a finite number of
points N, Ω = { Guilty, Innocent }.
 Events are mutually exclusive and sample
space is exhaustive.
 Combining events obey to logic of set theory
(conjunction and disjunction operations) and to
the distributive axiom of set theory.

 Hilbert Space (H). Contains a (in)finite number
of basis vectors , 𝑉 = |𝐺𝑢𝑖𝑙𝑡𝑦 , |𝐼𝑛𝑛𝑜𝑐𝑒𝑛𝑡 .
Allows complex numbers!
 Basis vectors are orthonormal (i.e, mutual
exclusive)
 Events are defined by subspaces. Combining
events obey the logic of subspaces. Does NOT
OBEY the DISTRIBUTIVE AXIOM!

|𝑆 =
1
2
|𝐺𝑢𝑖𝑙𝑡𝑦 +
1
2
|𝐼𝑛𝑛𝑜𝑐𝑒𝑛𝑡

 State is a probability function, denoted by Pr(.)
 Function directly maps elementary events into
probabilities. Pr 𝐺𝑢𝑖𝑙𝑡𝑦 =
1
2
 Empty set receives probability zero.
 Sample space receives probability one.

 State is a unit-length vector in the N-
dimensional vector space, defined by |𝑆 , used
to map events into probabilities
 The state is projected onto the subspaces
corresponding to an event, and the squared
length of this projection equals the event
probability.

Pr 𝐺𝑢𝑖𝑙𝑡𝑦 = 𝑃𝑔
2
=
1
2
2
=
1
2

 An event is observed and want to determine
other probabilities after observing this fact.
 Uses conditional probability function.
Pr 𝐼𝑛𝑛𝑜𝑐𝑒𝑛𝑡 𝐺𝑢𝑖𝑙𝑡𝑦 =
Pr⁡( 𝐼𝑛𝑛𝑜𝑐𝑒𝑛𝑡 ∩ 𝐺𝑢𝑖𝑙𝑡𝑦)
Pr⁡( 𝐺𝑢𝑖𝑙𝑡𝑦)
Pr 𝐼𝑛𝑛𝑜𝑐𝑒𝑛𝑡 𝐺𝑢𝑖𝑙𝑡𝑦 = 0

 Changes the original state vector by projecting
the original state onto the subspace
representing the observed event.
 The lenght of the projection is used as
normalization factor
|𝑆 𝑔 =
𝑃𝑔|𝑆
𝑃𝑔|𝑆

 Can all events be described within a single
sample space?
 Classical theory:YES! Unicity assumption!
 Quantum theory: Only if they share a common
basis!

 There is only one sample space.All events are
contained in this single sample space.
 A single probability function is sufficient to
assign probabilities to all events. Principle of
Unicity.
 Conjunction and disjunction operations are
well defined

 There is only one Hilbert Space where all
events are contained in.
 For a single fixed basis, the intersection and
union of two events spanned by a common set
of basis vectors is always well defined.
 A probability function assigns probabilities to
all events defined with respect to the basis.

 Event 𝐴 is spanned by 𝑉 = {|𝑉𝑖 , 𝑖 = 1, … , 𝑁}, such
that 𝑉𝐴 ⊂ 𝑉
 Event 𝐵 byW= {|𝑊𝑖 , 𝑖 = 1, … , 𝑁}, such that
𝑊𝐵 ⊂ 𝑊
 Then the intersection and union of these events is
not defined.
 Probabilities are assigned to sequences of events
using Luders rule. Distributive axiom does not
hold!

 Luders Rule: Compute the probability of the
sequence of events A followed by B.
Pr 𝐴 = 𝑃𝐴|𝑆 2
the revised state is |𝑆 =
𝑃 𝐴|𝑆
𝑃 𝐴|𝑆
 The probability of B has to be conditioned on the
first event , that is Pr 𝐵 𝑆𝐴 = Pr 𝐴 . Pr⁡( 𝐵|𝐴)
Pr 𝐴 . Pr 𝐵 𝐴 = 𝑃𝐴|𝑆 2
. 𝑃𝐵|𝑆𝐴
2

Pr 𝐴 . Pr 𝐵 𝐴 = 𝑃𝐴|𝑆 2
. 𝑃𝐵|𝑆𝐴
2
= 𝑃𝐴|𝑆 2
. 𝑃𝐵
𝑃𝐴|𝑆
𝑃𝐴|𝑆
2
=⁡ 𝑃𝐴|𝑆 2
.
1
𝑃𝐴|𝑆 2
𝑃𝐵 𝑃𝐴|𝑆 2
=⁡ 𝑃𝐵 𝑃𝐴|𝑆 2
≠ 𝑃𝐴 𝑃𝐵|𝑆 2

Cx Axis: Passos Coelho
Gx Axis: General
Ramalho Eanes

 Coelho-Eanes  Eanes-Coelho
 Using a quantum model, the probability of responses
differ when asked first vs. when asked second

 Passos Coelho is a honest person
|𝑆 = 0.8367|𝑃 + 0.5477𝑃
 General Eanes is a honest person
|𝑆 = 0.9789|𝐺 − 0.2043𝐺
 Analysis of first question – Passos Coelho
Pr 𝐶𝑦 = 𝑃𝐶|𝑆 2
= 0.8367 2
= 0.70
Pr 𝐶𝑛 = 𝑃𝐶|𝑆 2 = 0.5477 2 = 0.30

 Passos Coelho is a honest person
|𝑆 = 0.8367|𝑃 + 0.5477𝑃
 General Eanes is a honest person
|𝑆 = 0.9789|𝐺 − 0.2043𝐺
 Analysis of first question – General Eanes
Pr 𝐺𝑦 ⁡= 𝑃𝐺|𝑆 2
= 0.9789 2
= 0.9582
Pr 𝐺𝑛 = 𝑃𝐺|𝑆 2 = −0.2043 2 = 0.0417

 Analysis of the second question
 The probability of saying “yes” to Passos Coelho is the
probability of saying “yes” to General Eanes and then “yes”
to Passos Coelho plus the probability of saying “no” to
General Eanes and then “yes ” to Passos Coelho
Pr 𝐶𝑦 = 0.96 . 0.50 + 0.04 . (0.50)
= 0.50
Pr 𝐺𝑦 = 0.70 . 0.50 + 0.30 . (0.50)
= 0.50

 According to this simplified two dimensional model:
 Large difference between the agreement rates for
two politicians in non-comparative context: 70% for
Passos Coelho and 96% for General Eanes
 There is no difference in the comparative context:
50% for both.

 Effects on question order
 Human Probability Judgment Errors
 The SureThing Principle
 The Double Slit Experiment

“ Linda is 31 years old, single, outspoken and very
bright. She majored in philosophy. As a student, she
was deeply concerned with issues of discrimination
and social justice, and also participated in anti-
nuclear demonstrations.”
Choose what Linda is more likely to be:
(a) BankTeller;
(b) Active in the Feminist Movement and BankTeller
Morier, D.M. & Borgida, E. (1984).The conjunction fallacy: a task specific
phenomena? Personality and Social Psychology Bulletin, 10, 243-252

 90% of people answered option (b) over option (a).
 People judge Linda to be: “Active in the feminist
movement and a bank teller” over being a “Bank
Teller”
Pr 𝐹𝑒𝑚𝑖𝑛𝑖𝑠𝑡⁡ ∩ 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟 ≥ Pr⁡( 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟)
Morier, D.M. & Borgida, E. (1984).The conjunction fallacy: a task specific

 According to mathematics and logic, we were
expecting to find:
Pr⁡( 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟) ≥ Pr⁡( 𝐹𝑒𝑚𝑖𝑛𝑖𝑠𝑡⁡ ∩ 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟)
 Even if we considered:
⁡⁡Pr 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟 = 0.03⁡⁡⁡ Pr 𝐹𝑒𝑚𝑖𝑛𝑖𝑠𝑡 = 0.95
⁡⁡⁡Pr 𝐹𝑒𝑚𝑖𝑛𝑖𝑠𝑡⁡ ∩ 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟 = 0.03 × 0.95
⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡⁡= 0.0285 ≤ Pr⁡( 𝐵𝑎𝑛𝑘⁡𝑇𝑒𝑙𝑙𝑒𝑟)

The SureThing Principle:
“ If under state of the world X, people prefer
action A over action B and in state of the world
~X prefer action A over B, then if the state of
the world in unknown, a person should always
prefer action A over B” (Savage, 1954)
Savage, L.J. (1954).The Foundations of Statistics. JohnWiley & Sons.

 At each stage, the decision was wether or not
to play a gamble that has an equal chance of
winning $2.00 or loosing $1.00.
 Three conditions for participants:
 Informed they won the first gamble
 Informed they lost the first gamble
 Did not know the outcome of the first gamble
Tversky, A. & Shafir, E. (1992).The disjunction effect in choice under uncertainty.
PsichologicalScience, 3, 305-309

 Results:
 If participants knew they won the first gamble,
(68%) chose to play again.
 If participants knew they lost the first gamble,

 Results:
 If participants knew they won the first gamble,
 If participants knew they lost the first gamble,
 If participants did not know the outcome of the
first gamble, (64%) chose not to play.

U
W: Believe
win 1st play
L: Believe
loose 1st play
N
G
N: Decides NOT
to play 2nd game
G: Decides to
GAMBLE again
Unobserved
𝑾|𝑼
𝑳|𝑼
𝑵|𝑾
𝑮|𝑳
𝑮|𝑾
𝑵|𝑳

 Classical theory - Law of total probability
Pr 𝐺 𝑈 = 𝑃𝑟 𝑊 𝑈 . 𝑃𝑟 𝐺 𝑊 + Pr 𝐿 𝑈 . Pr⁡( 𝐺|𝐿)
 From this law, one would expect:
Pr 𝐺 𝑊 = 0.69 > Pr 𝐺 𝑈 > Pr 𝐺 𝐿 = 0.59
 Tversky & Shafir (1992) found that
Pr 𝐺 𝑈 = 0.36 < Pr 𝐺 𝐿 = 0.59 < Pr 𝐺 𝑊
= 0.69

 Quantum theory - Law of total amplitude
Pr 𝐺|𝑈 = | 𝑊|𝑈 𝐺|𝑊 + 𝐿|𝑈 𝐺|𝐿 |2
= 𝑊|𝑈 𝐺|𝑊 2
+ 𝐿|𝑈 𝐺|𝐿 2
+
+2. 𝑅𝑒[ 𝑊 𝑈 𝐺 𝑊 𝐿 𝑈 𝐺 𝐿 . 𝐶𝑜𝑠⁡𝜃]
 To account forTversky and Shafir results, 𝜃 must be
chosen such that
2. 𝑅𝑒 𝑊 𝑈 𝐺 𝑊 𝐿 𝑈 𝐺 𝐿 . 𝐶𝑜𝑠⁡𝜃 < 0

 A single electron is dispersed from a light
source.
 The electron is split into one of two channels
(C1 or C2) from which it can reach one of the
two detectors (D1 or D2).

 Two conditions are examined:
 The channel through which the electron passes is
observed.
 The channel through which the electron passes in
not observed.

Quantum Models of Cognition and Decision

𝑃𝑟 𝑐1 = 0.5 0
0.5 0.5
0.5 0.5
= 0.25 0.25
Pr 𝑐2 = 0 0.5
0.5 0.5
0.5 0.5
= 0.25 0.25
Pr 𝑐1⁡𝑜𝑟⁡𝑐2 = Pr 𝑐1 + Pr 𝑐2 = 0.5 0.5
0.5
0.5
0.5
0.5
0.5 0.5

𝑐1 = 1 2 𝑖 0
1 2 𝑖 1 2 𝑖
1 2 𝑖 − 1 2 𝑖
= −0.5 −0.5
𝑐2 = 0 1 2 𝑖
1 2 𝑖 1 2 𝑖
1 2 𝑖 − 1 2 𝑖
= −0.5 −0.5
Pr 𝑐1⁡𝑜𝑟⁡𝑐2 = 𝑐12
+ 𝑐22
= 0.5 0.5
1 2𝑖
1 2𝑖
−1 2𝑖
1 2𝑖 1 2𝑖
1 2𝑖

𝑐12 = 1 2 𝑖 1 2 𝑖
1 2 𝑖 1 2 𝑖
1 2 𝑖 − 1 2 𝑖
= −1 0
Pr 𝑐12 = 𝑐122 = 1 0
1 2𝑖
1 2𝑖
−1 2𝑖
1 2𝑖 1 2𝑖
1 2𝑖

 If we do not observe the system
 Then, we cannot assume that one of only two
possible paths are taken
 In quantum theory, the state is superposed
between the two possible paths!

 What are the implications of quantum
probabilities in Computer Science models?
 Bayesian Networks, Markov Networks
A B
C D
E
Unobserved Nodes
Observed Nodes

 What are the implications of quantum
probabilities in Decision Making?
 Decision trees, utility functions, risk management...

 What are the implications of quantum theory
in machine learning?
 A couple of works in the literature state that it is
possible!

 Busemeyer, J., Wang, Z., Townsend, J. (2006). Quantum
Dynamics of Human Decision-Making, Journal of Mathematical
Psychology, 50, 220-241
 Pothos, E., Busemeyer, J. (2009). A Quantum Probability
Explanation for Violations of Rational Decision Theory,
Proceedings of the Royal Society B
 Busemeyer, J., Wang, Z., Lambert-Mogiliansky, A. (2009).
Empirical Comparison of Markov and Quantum Models of
Decision Making, Journal of Mathematical Psychology, 53, 423-
433
 Busemeyer, J., Pothos, E., Franco, R., Trueblood, J. (2011). A
Quantum Theoretical Explanation for Probability Judgment
Errors, Psychological Review, 118, 193-218
 Trueblood, J., Busemeyer, J.(2011). A Quantum Probability
Account of Order Effects in Inference, Cognitive Science, 35,
1518-1552

 Busemeyer, J., Bruza, P. (2012), Quantum Models of Decision and
Cognition, Cambridge Press
 Khrennikov, A.(2010). Ubiquitous Quantum Structure: From Psychology
to Finance. Springer.
 Griffiths, R. (2003). Consistent Quantum Theory. Cambridge University
Press.
 Moore, D.W. (2002). Measuring new types of question order effects.
Public Opinion Quaterly, 66, 80-91
 Morier, D.M., Borgida, E. (1984). The conjunction fallacy: a task specific
 Tversky, A. & Shafir, E. (1992). The disjunction effect in choice under
uncertainty. Psichological Science, 3, 305-309
 Savage, L.J. (1954). The Foundations of Statistics. John Wiley & Sons.

Quantum Models of Cognition and Decision

More Related Content

Similar to Quantum Models of Cognition and Decision (20)

Recently uploaded (20)

Quantum Models of Cognition and Decision