NEUROBIOLOGY OF DECISION-MAKING, CSHL, May 2005
Choice, decision and action investigated with visually guided
saccades.
Jeffrey D. SchallWith
Leanne Boucher, Gordon Logan & Tom Palmeri
• Choice – action in the context of alternatives to satisfy a goal, desire or preference• Action – movements with consequences that can be explained by referring to preferences, goals and beliefs• Decision – deliberation when alternatives are vague, payoffs are unclear and habits are reversed
Definitions
“I look forward to playing and hopefully I can get to that point where I can make that decision.” — Michael Jordan on his anticipated return to professional basketball. Associated Press, 19 July 2001.
"I feel that way right now. Ask me in two or three months and I may change. I don't think I will. I'm pretty sure that's my decision." — Michael Jordan on his retirement from professional basketball. Associated Press, 17 July 1998.
• Distinguish two uses of “decision”• As characteristic of behavior (e.g., Decision Theory)
• But measures of outcome do not specify mechanism• As process producing behavior
• Mechanism with particular architecture
• Decision as process has two distinct meanings• Decide to -- Alternative actions (can be identified with choosing)• Decide that -- Alternative categories (not identified with choosing)
Further defining “decision”
• The properties of neurons do not reveal function
• Formal (computational) theories of performance explain function
• But distinct models cannot be distinguished from behavior testing, e.g., diffusion or race
• Properties of neurons might provide constraints to distinguish between models …
• … if and only if the neural activity measured is the instantiation of the cognitive process in question, which constitutes a linking proposition
Necessity of formal linking propositions
Teller DY. 1984. Vision Research 24:1233-1246Schall JD. 2004. Ann Rev Psychol 55:23-50
Linking propositions for decision making
Time from stimulus (sec)A
ctiv
atio
n0.0 0.1 0.2
Hanes & Schall (1996) described neural activity that looked like an accumulator.
They identified this activity with form of sequential sampling models.
Time from stimulus (sec)
Act
ivat
ion
0.0 0.1 0.2
Linking propositions for decision makingRT = Decision time + Residual timeResidual time = Encoding time + Preparation time
Stimulus encoding Sequential sampling Response preparation
Time from stimulus (sec)
Act
ivat
ion
0.0 0.1 0.2
Countermanding task
NO STOP SIGNAL Trials
Reaction Time
STOP SIGNAL Trials
Stop Signal Delay
CANCELLED
NON-CANCELLED
Countermanding performance
Countermanding paradigm: Race model
Logan, G.D. & Cowan, W.B. (1984) On the ability to inhibit thought and action: A theory of an act of control. Psychological Review 91:295-327. Hanes DP and Schall JD (1995) Countermanding saccades in macaque.Visual Neuroscience 12:929-937
Reaction Time
Stop Signal DelayCANCELLED
“GO”
“GO”
“STOP”
NON-CANCELLED
“GO”
“STOP”
Visual Cortex
LGN
RF
Saccade
Thalamus
Cerebellum
SCsSCi
Frontal cortex
(DLPFC, ACC, SEF)Parietal
Cortex (LIP)
Retina
Temporal Cortex (TEO)
FEF
Basal Ganglia
Munoz DP, Schall JD (2003) Concurrent distributed control of saccade initiation in the frontal eye field and superior colliculus. In The Oculomotor System: New Approaches for Studying Sensorimotor Integration. Edited by WC Hall, AK Moschovakis. CRC Press, Boca Raton, FL. Pages 55-82.
Saccades are produced by a distributed network
Countermanding physiology
No stop trials
Non-canceled trials
No stop trials
Canceled trials
STOP SSRT STOP SSRT
Hanes, D.P., W.F. Patterson, J.D. Schall (1998) The role of frontal eye field in countermanding saccades: Visual, movement and fixation activity. Journal of Neurophysiology 79:817-834.
Pare M, Hanes DP (2003) Controlled movement processing: superior colliculus activity associated with countermanded saccades. Journal of Neuroscience 23:6480-6489.
1 - The race model of countermanding performance assumes that the GO and the STOP processes have independent finish times (Logan & Cowan, 1984).
Mapping the race model onto neural processes
2 – Saccades are produced by a network of interacting neurons.
Paradox – How can a network of interacting neurons produce behavior that looks like the outcome of race between independent processes?
Explore properties of simple network of GO and STOP units.Mapping the race model onto neural processes
Constrained by the characteristics of countermanding behavior and by the form of activation of neurons
L.Boucher, G.D.Logan, T.J.Palmeri, J.D.Schall. An interactive race model of countermanding saccades. Program No. 72.10. 2003 Abstract Viewer/Itinerary Planner.
GOSTOPSTOPGOGO
dta
dtda
STOPGOGOSTOPSTOP
dta
dtda
STOP
GO
GO, GO
ST
OP
D STOP
GO
DGO
STOP, STOP
STOPGO
a
a
STOP processGO process
0.0
0.5
1.0
0 100 200 300
a
Complete independenceSTOPGO
L.Boucher, G.D.Logan, T.J.Palmeri, J.D.Schall. An interactive race model of countermanding saccades. Program No. 72.10. 2003 Abstract Viewer/Itinerary Planner.
Complete independenceReproduces countermanding behavior…
STOPGO
L.Boucher, G.D.Logan, T.J.Palmeri, J.D.Schall. An interactive race model of countermanding saccades. Program No. 72.10. 2003 Abstract Viewer/Itinerary Planner.
Stop signal delay (ms)
pro
bab
ility
(no
n-c
ance
lled
)
50 100 150 200 250
0.0
0.5
1.0
Observed
Model
b
Reaction time (ms)
200 300 400
0%
50%
100%c
L.Boucher, G.D.Logan, T.J.Palmeri, J.D.Schall. An interactive race model of countermanding saccades. Program No. 72.10. 2003 Abstract Viewer/Itinerary Planner.
Complete independence
Stop Signal SSRT
Stop Signal SSRT
… but does not produce correct activations.
The GO process isnever interrupted!
STOPGO
Key insight – the inhibition of STOP on GO cannot be uniform and instantaneous; it must be late and potent
STOPGO
Δt
Delayed potent STOPSTOPGO
Δt
0 100 200 3000.0
0.5
1.0
a
Time from stimulus (ms)
Ac
tiva
tion
STOP processGO process
STOP SSRT
Delayed potent STOPReproduces countermanding behavior…
Stop signal delay (ms)
Pro
bab
ility
(n
on
can
cell
ed)
50 100 150 200 2500.0
0.5
1.0
ObservedModel
b
Reaction time (ms)200 300 400
0%
100%
50%
c
STOPGO
Δt
Delayed potent STOP… and reproduces neural activation
The GO process is not modulated in non-canceled trials
The GO process is modulated within SSRTin canceled trials
0 100 200 3000 100 200 300
Activa
tion
Time from stimulus(ms)
0.0
1.0
0.5
STOPGO
Δt
Countermanding performance is produced by pool of neurons the prepare movements (GO process) and pool of neurons that interrupt preparation (STOP process).
The STOP process is composed of an early (afferent) stochastic stage and a late potent interruption stage.
Specific conclusions
Target Stop signal
D GO D STOP
0
Time from stimulus(ms)
SSRT
50 250200150100 300
Redundant but distinct models cannot be distinguished based on behavior data (Moore, 1956, in Automata Studies, ed. CE Shannon, J McCarthy. Princeton Univ. Press)
Properties of neurons can distinguish between alternative architectures… but only if neurons instantiate the processes in question.
GO process identified with pool of “movement” neurons.STOP process identified with pool of “movement inhibition” neurons.
General conclusions
Stochastic response preparation process necessary to explain countermanding performance.
If so, response preparation must be more or less stochastic during all tasks.
Therefore, the proper form of response preparation variability must be incorporated into sequential sampling models of perceptual or memory decisions.
This and much other evidence indicates that RT is the expression of at least two distinct but not necessarily discrete stages of processing – encoding+categorization (decide that) and response preparation (decide to).
General conclusions continued
General conclusions continued
It is possible now to determine the duration of intermediate stages with invasive measures of neural states.
However, this depends on proper linking propositions.
Information about process durations and transitions is necessary to elucidate how stimulus ambiguity, prior probability and reward history influence choices.
"[Since] we cannot break up the reaction into successive acts and obtain the time of each act, of what use is the reaction time?" – R.S. Woodworth (1938) in Experimental Psychology [quoted in Luce (1986)]
Rees et al. Nature Neuroscience 3, 716 - 723 (2000)
An empirical basis for distinguishing between choosing and deciding
It is deciding when anterior cingulate cortex is engaged.
Area MT
fMR
I am
plit
ude
0% 100%
Motion strength
Anterior cingulate cortex
0% 100%
Motion strength
Monkey C Monkey A1 Monkey A2 Parameter Independent Interactive Independent Interactive Independent Interactive
μGO 4.62 4.64 5.49 5.42 5.12 4.99
σGO 18.53 18.22 23.11 22.81 22.91 22.84
μSTOP 12.11 11.56 27.42 21.45 27.60 23.08
σSTOP 13.22 18.22 140.60 140.97 140.90 141.07
βGO 0.00 0.00429 0.00 0.000265 0.00 0.0399
βSTOP 0.00 0.00694 0.00 0.00561 0.00 0.0399
DSTOP 1 71 27 15 27 50
χ2 22.85 24.23 89.74 91.16 106.41 96.91 cancel time — -18 — -25 — -23
STOPinterrupt — 11 — 35 — 10 SSRT 103 102 77 77 76 85
5 6 7 8 9 10 11 12 1330
40
50
60
70
80
90
Rate
D
(ms
)
a
b
- 17
1.0 1.2 1.4 1.6 1.8 2.0 2.2 2.4 2.6 2.8
c
m
/ mmSTOP
STOP
GO
a
0 100 200 300
c
1000 200 300
50 100 150 200 250
X = 2.332
150 200 250 300 350 400
3.990.746.812.33
b
X =2
Stop signal delay (ms)
Reaction time (ms)
50 100 150 200 25050 100 150 200 2500.0
0.5
1.0
0%
50%
100%
Pro
bab
ility
(n
on
can
cell
ed)
Ac
tiva
tio
n
1.0
0.0
0%
50%
100%
0.0
0.5
1.0
0.0
0.5
1.0
0%
50%
100%
1.0
0.0
Time from stimulus (ms)Time from stimulus (ms)
Reaction time (ms)Reaction time (ms)150 200 250 300 350 400150 200 250 300 350 400
Stop signal delay (ms)Stop signal delay (ms)
X =2X =
2
X = 2.492X = 2.35
2
6.240.747.535.24
4.590.747.575.24
100 200 300Time from stimulus (ms)
0
1.0
0.0
0.5 0.5 0.5
ST
OP
c
b
Cancel time = -17Cancel time = -20Cancel time = -22
-15
-16
-18
-19
-21
-22
-23-24
Time from target (ms)0 200 400
SSRTStop SignalSSRT Stop Signal
4002000
100
Time from target (ms)
Act
ivat
ion
(Spi
kes/
sec)
Fixation cell activity from FEF & SC
Hanes, D.P., W.F. Patterson, J.D. Schall (1998) The role of frontal eye field in countermanding saccades: Visual, movement and fixation activity. Journal of Neurophysiology 79:817-834.
Pare M, Hanes DP (2003) Controlled movement processing: superior colliculus activity associated with countermanded saccades. Journal of Neuroscience 23:6480-6489.