1. Cognition and Decision-Making
Read Montague
Baylor College of Medicine
Houston, TX

2. Fundamental difference between brains and computers

4. 8 dedicated chess processors per node (total 256)
DEEP BLUE
32 nodes
8 dedicated chess processors per node (total 256)
1.4 tons (lots of AC)
200,000,000 positions per second

5. Deep Blue was wildly inefficient
Kasparov stayed warm to the touch
Kasparov’s brain is freakishly efficient
~100 watts 
brain uses ~20 watts

6. functional MRI (fMRI) …

7. Behavioral or perceptual
Relating brain responses to behavior & perception
HEAT EQUIVALENTS
THEORIES
Physical measurement
(many tools here)
Behavioral or perceptual
measurement (model)
expressed trust + -
change in $ sent back
Current work is moving away from psychological categories to
hypothesized computational functions

8. Why should brands influence tastes?

9. Counterfactual experience
Goal-directed choice
Select goal
Sustain goal
Pursue goal
Actual experience
Counterfactual experience
‘what could have been’
guidance
signals

10. Sharks don’t go on hunger strikes
But humans do
Ideas gain behavioral potency
of primary rewards like food and water
How?

11. Valuation Dopamine neurons
Goal pursuit requires guidance signals  dopamine
Valuation
fMRI
experiments
Computational
models
Dopamine neurons

12. Relating valuation to choice
Response to
stimulus
(art, brand name)
Internal
Valuation
Make movement
or reveal choice
In scanner
Outside scanner ?

13. The brain has guidance signals related to ongoing valuation

14. Counterfactual experience
Goal-directed choice
Select goal
Sustain goal
Pursue goal
Actual experience
Counterfactual experience
‘what could have been’
guidance
signals

15. Midbrain dopamine neurons
Pause, burst, and ‘no change’ responses represent
reward prediction errors – ongoing emission of information R
time
naive
burst
pause R
time
After learning
time
After learning
(catch trial)
ERROR SIGNAL = current reward + g next prediction - current prediction

16. Can we detect a reward prediction error signal in a human subject?

17. Neural correlates of these error signals have been observed in conditioning and decision tasks
Passive conditioning tasks
Instrumental conditioning tasks
Sequential decision tasks
Social and economic exchange tasks

18. J J Induce reward prediction error in
a passive conditioning task  measure brain response 6s J
train 6s 4s J
test

19. J + - 6s 4s Hemodynamic response in ventral putamen
0.4
0.8
0.0
-0.4
% signal change 2 6 10 12
sec + -
Hemodynamic response in ventral putamen 6s 4s J
“BETTER THAN EXPECTED”
Juice not expected but delivered
Juice expected but not delivered
“WORSE THAN EXPECTED”
McClure et al., Neuron, 2003

20. Midbrain dopamine neurons
burst and pause responses encode reward prediction errors
ERROR SIGNAL = current reward + g next prediction - current prediction
“I want to solve Fermat’s last theorem”
Can these systems be re-deployed for abstractly defined rewards (ideas)?

21. What about the something more abstract like the expression of trust?

22. Must involve risk (uncertainty)
Trust
Modeling
Must involve risk (uncertainty)
Social lubricant

23. TRUST A B
I love you
Let’s do lunch B A

24. Simplifying and quantifying Trust (Berg et al
Simplifying and quantifying Trust (Berg et al., 1995; Weigelt and Camerer, 1988)
Trust is the amount of money a sender sends to a receiver
without external enforcement.
Agent 2
Agent 1 I R

25. A dynamic version of the Trust game (10 rounds)
X 3
$20
Investor
Trustee

26. Scan all brains during the interaction
information:
software download:
client
Hyperscan server
and database
client
Houston (BCM)
California (Caltech)
New Jersey (Princeton)
Georgia (Emory)
Germany (Uni-Ulm)
China (HKUST, Guangzho)

27. Structure of a round

28. Reciprocity = TIT-FOR-TAT
What is the behavioral signal that most strongly influences changes in trust (money sent) ?
Agent 2
Agent 1 I R
Reciprocity = TIT-FOR-TAT

29. Reciprocity = TIT-FOR-TAT+
Benevolent
signal
money sent
to partner
Neutral
signal -
Malevolent
signal

30. ‘intention to increase trust’ shifts with reputation building
Trustee Brain:
‘intention to increase trust’ shifts with reputation building
submit
reveal *
increases or decreases in future trust by trustee
-0.2
0.2
time (sec)
Trustee will increase
trust on next move
Trustee will decrease
trust on next move
Reputation develops *
time (sec) -8 10
-0.2
0.2
Signal is now
anticipating the
outcome

31. Temporal shift resembles value transfer in reward learning experiments
time
naive
burst
pause R
time
After learning
time
After learning
(catch trial)

32. What about something more culturally dependent like a brand image?

33. Taste and preference are perceptual constructs Subject to ‘illusions’
Numerous variables
Subject to ‘illusions’

34. Top is
visible
condensation
orientation

35. Anonymous taste test (outside scanner)P
3 pairs Or
15 pairs
Panel B is the one-way t-test of surprising delivery of Pepsi-Coke contrast in the Double blind test.

36. No correlation between
stated preference
and
revealed preference
n1 = 18
n2 = 18 8 6
Ave. # Coke Selections
(out of 15) 4 2
Prefer
Coke
Prefer
Pepsi
Carbonated

37. Imaging part of experiment
outside scanner
inside scanner 6s P C P 6s C
train P C 6s 4s P P C 6s 4s C
test
Contrast surprising Coke delivery to surprising Pepsi delivery
Analogous to Smiling – Crying contrast from before

38. Response that correlates with anonymous behavioral preference
% signal change
(coke – pepsi)
t score 6
# selections to Coke
n=15
-0.2
0.0
0.2
0.4 1 2 3
Ventro-medial
prefrontal cortex
% signal change
(coke – pepsi)
# selections to Coke 5 10 15
-0.6
-0.4
-0.2
0.0
0.2
0.4
t score 6
n=17

39. “Coke must taste different when you think it’s coke,
do you think we could image that?”
Latane’ Montague
high school student
What does the idea of the brand taste like?

40. Same systems engaged by visual art during passive viewing

41. Simply look at art in the scanner
4-14s (random) 5s
4-14s (random)

42. Liking and familiarity ratings outside scanner-4 4

43. Common valuation response?
% signal change
(coke – pepsi)
# selections to Coke 5 10 15
-0.6
-0.4
-0.2
0.0
0.2
0.4
t score 6
n=17
Art Familiarity
Art Preference
Coke Preference
same region
Use idiosyncratic ratings for preference and familiarity
for each individual

44. 1. All decisions derive from biological mechanisms and will have some kind of associated brain responses…

45. 2. Abstractions that become goals appear to re-deploy reward-harvesting circuitry that we share with all vertebrates

46. 3. Relationship between choice and brain response is statistical, never deterministic
However, choices can be systematically predictable using brain imaging data
Soda, art, etc.

47. Collaborators Baylor College of Medicine Emory University Pearl Chiu
Amin Kayali
Brooks King-Casas
Terry Lohrenz
Sam McClure (Princeton)
Read Montague
Damon Tomlin
Caltech
Cedric Anen
Colin Camerer
Steve Quartz
UCL
Peter Dayan
Nathaniel Daw
Salk Institute
Terry Sejnowski
Princeton University
Jon Cohen
Emory University
Greg Berns
University of Alabama
Laura Klinger
Mark Klinger
Families of autistic subjects
Funding Sources
NIDA
NIMH
Dana Foundation
Kane Family Foundation
Angel Williamson Imaging Center
Institute for Advanced Study, Princeton NJ