Click here and press the right key for the next slide (or swipe left)
also ...
Press the left key to go backwards (or swipe right)
Press n to toggle whether notes are shown (or add '?notes' to the url before the #)
Press m or double tap to slide thumbnails (menu)
Press ? at any time to show the keyboard shortcuts
How do rational agents decide which of several available actions to perform?
How do rational agents decide which of several available actions to perform?
How you act
is a function of two things:
your preferences concerning various outcomes
and your beliefs about how likely an available action is to satisfy each of your preferences.
‘we should think of meanings and beliefs as interrelated constructs of a single theory just as we already view subjective values and probabilities as interrelated constructs of decision theory’
Davidson, 1974 p. 146
The Roots, 2006
A game is ‘any interaction between agents that is governed by a set of rules specifying the possible moves for each participant and a set of outcomes for each possible combination of moves’ (Hargreaves and Varoufakis, 2004 p. 3)
Aim: describe rational behaviour in social interactions
‘we wish to find the mathematically complete principles which define “rational behavior” for the participants in a social economy, and to derive from them the general characteristics of that behavior’
von Neumann & Morgenstern, 1953 p. 31
me | |||
put £10 in box A | put £10 in box B | ||
you | open box A | £10 £0 | £0 £0 |
open box B | £0 £0 | £10 £0 |
me | |||
put £10 in box A | put £10 in box B | ||
you | open box A | £8 £2 | £0 £0 |
open box B | -£2 £2 | £10 £0 |
How you act
is a function of two things:
your preferences
and your beliefs about how others will act.
Prisoner X | |||
resist | confess | ||
Prisoner Y | resist | 3 3 | 0 4 |
confess | 4 0 | 1 1 |
How you act
is a function of two things:
your preferences
and your beliefs about how others will act.
The Roots, 2006
A game is ‘any interaction between agents that is governed by a set of rules specifying the possible moves for each participant and a set of outcomes for each possible combination of moves’ (Hargreaves and Varoufakis, 2004 p. 3)
Aim: describe rational behaviour in social interactions
‘we wish to find the mathematically complete principles which define “rational behavior” for the participants in a social economy, and to derive from them the general characteristics of that behavior’
von Neumann & Morgenstern, 1953 p. 31
A nash equilibrium for a game
is a profile of actions
from which no agent can unilaterally profitably deviate
see Osborne & Rubinstein, 1994 p. 14; Dixit et al, 2014 p. 95
Prisoner X | |||
resist | confess | ||
Prisoner Y | resist | 3 3 | 0 4 |
confess | 4 0 | 1 1 |
Gangster X | |||
back off | fight | ||
Gangster Y | back off | 3 3 | 1 4 |
fight | 4 1 | 0 0 |
Game Theory
Aim: describe rational behaviour in social interactions.
An action is rational
in a noncooperative game
if it is a member of a nash equilibrium?
How you act
is a function of two things:
your preferences
and your beliefs about how others will act.
Your beliefs about how others will act
are a function of your knowledge of two things:
your beliefs about their preferences
and your beliefs about how they believe others will act.
Your beliefs about how they believe others will act ...
decision theory
How do rational agents decide which of several available actions to perform?
How you act
is a function of two things:
your preferences concerning various outcomes
and your beliefs about how likely an available action is to satisfy each of your preferences.
game theory
When two or more agents interact,
so that which outcome one agent’s choice brings about depends on how another chooses,
how do their preferences guide their choices?
How you act
is a function of two things:
your preferences
and your beliefs about how others will act.
You might have been interpreting decision theory as an account of how people figure out what to do ...
... but decision theory is agnostic about processes.
On explanation: ‘Many events and outcomes prompt us to ask: Why did that happen? [...] For example, cutthroat competition in business is the result of the rivals being trapped in a prisoners’ dilemma’
Dixit et al, 2014 p. 36
Games with the Prisoner’s Dilemma structure arise in:
bower birds (maraud/guard nests)
business organisations (product pricing)
countries (international environmental policy)
individual adult humans (suspects under arrest)
Dixit et al, 2014 chapter 10
1. Applications range from microbial populations to countries.
2. The explanations are of the same type in every case.
3. The underlying processes probably differ.
4. Therefore, game theory is agnostic about processes.
Decision theory is agnostic about processes.
Compare explanation of patterns of behaviour in:
male side-blotched lizard morphology
vs
children playing rock-paper-scissors.
But psychological sciences are not agnostic about processes ...
What kinds of processes in
individual animals
guide actions?
habitual
Action occurs in the presence of Stimulus.
Agent is rewarded [/punished]
Stimulus-Action Link is strengthened [/weakened] due to reward [/punishment]
Given Stimulus, will Action occur? It depends on the strength of the Stimulus-Action Link.
habitual vs habitual
‘A habitual action, state, or way of behaving is one that someone usually does or has, especially one that is considered to be typical or characteristic of them.’
Habitual
Thorndyke’s Law of Effect:
‘The presentation of an effective [=rewarding] outcome following an action [...] reinforces
a connection between the stimuli present when the action is performed and the action itself
so that subsequent presentations of these stimuli elicit the [...] action as a response’
Dickinson, 1994 p. 48
habitual
Action occurs in the presence of Stimulus.
Agent is rewarded [/punished]
Stimulus-Action Link is strengthened [/weakened] due to reward [/punishment]
Given Stimulus, will Action occur? It depends on the strength of the Stimulus-Action Link.
instrumental
Action leads to Outcome.
Action-Outcome Link is strengthened.
Agent has strong [/weak] positive [/negative] Preference for Outcome
Will Action occur? It depends on the strength of Action-Outcome Link and Agent’s Preference.
Is this lever pressing habitual or instrumental aciton?
habitual
Stimulus is the layout of this room.
Rat (=Agent) is rewarded with food
Room-LeverPress (=Stimulus-Action) Link is strengthened due to reward
Thf LeverPress (=Action) will occur in this room (=Stimulus).
instrumental
Lever pressing (=Action) leads to food (=Outcome).
Thf LeverPress-Food (=Action-Outcome) Link is strong.
Rat (=Agent) has strong positive Preference for food.
Thf LeverPress (=Action) will occur.
Problem: different hypotheses, same prediction
What if we devalue the food?
habitual
Stimulus is the layout of this room.
Rat (=Agent) is rewarded with food
Room-LeverPress (=Stimulus-Action) Link is strengthened due to reward
Thf LeverPress (=Action) will occur in this room (=Stimulus).
instrumental
Lever pressing (=Action) leads to food (=Outcome).
Thf LeverPress-Food (=Action-Outcome) Link is strong.
Rat (=Agent) has strong positive Preference for food.
Thf LeverPress (=Action) will occur.
What if we devalue the food?
Instrumental : it will reduce lever pressing (to none)
Habitual : it will have no effect on lever pressing
Dickinson, 1985 figure 3
What if we devalue the food?
Instrumental : it will reduce lever pressing (to none)
Habitual : it will have no effect on lever pressing
‘the laboratory rat fits the teleological [instrumental] model; performance of this particular instrumental behaviour really does seem to be controlled byknowledge about the relation between the action and the goal’
Dickinson, 1985 p. 72
‘we did not conclude that all such responding was of this form.
Indeed, we observed some residual responding during the post-re-valuation test that appeared to be impervious to outcome devaluation and therefore autonomous of the current incentive value,
and we speculated that this responding was habitual
Dickinson, 2016 p. 179
Dickinson, 1985 figure 3
puzzle
If the action is habitual,
why is it modulated by devaulation?
If the action is instrumental,
why does it still occur (albeit less frequently) after devaluation?
Dual-Process Theory of Action
some actions are ‘controlled by two dissociable processes: a goal-directed [instrumental] and an habitual process’
Dickinson, 2016 p. 179
Is this lever pressing habitual or instrumental aciton?
Are the causes of this action habitual or instrumental? Both!
Neurophysiological Evidence
‘[instumental] and habitual control have been doubly dissociated in two brain regions.
In the PFC, lesions of the prelimbic and infralimbic areas disrupt goal-directed [instrumental] and habitual behavior, respectively ...
These dissociations suggest that different neural circuits mediate the two forms of control’
Dickinson, 2016 p. 184
‘instrumental behavior itself involves two systems, the goal-directed and the habitual’
Illustration: stress (Schwabe & Wolf, 2010)
Schwabe and Wolf, 2010 figure 1
Schwabe and Wolf, 2010 figure 6
When stressed,
your preferences matter less:
habits dominate.
Q1
Donald Davidson asks, ‘What is the mark that distinguishes ... actions?’ Are scientific discoveries relevant to answering this question?
Q2
What is the mark that distinguishes actions?
Q3
How are non-accidental matches between intentions and motor representations possible?
If events can be actions despite being dominated by habitual processes, why are several philosophers confident that all actions are intentional?
‘once we accept that there are complex and subtle non-intentional processes, such as those mediating basic goal-approach and the adjustment to changes in motivational state, that can mimic true intentional control in many situations, we can understand why the propensity to perceive actions as intentional may have developed. Given that
either there is nothing in the stimulus input per se to distinguish intentional from non-intentional behaviour
or that
such a discrimination yields little of consequence in most situations,
it may well pay the perceiver to treat both classes of behaviour as intentional in predicting the subsequent course of events’
Heyes & Dickinson, 1990 p. 102
tension
The range of applications of Decision Theory shows it is agnostic about processes.
Dickinson’s instrumental process is characterised by appeal to Decision Theory.
habitual
Action occurs in the presence of Stimulus.
Agent is rewarded [/punished]
Stimulus-Action Link is strengthened [/weakened] due to reward [/punishment]
Given Stimulus, will Action occur? It depends on the strength of the Stimulus-Action Link.
instrumental
Action leads to Outcome.
Action-Outcome Link is strengthened.
Agent has strong [/weak] positive [/negative] Preference for Outcome
Will Action occur? It depends on the strength of Action-Outcome Link and Agent’s Preference.
The range of applications of Decision Theory shows it is agnostic about processes.
Dickinson’s instrumental process is characterised by appeal to Decision Theory.
model vs construal
e.g. model of a house
another contrast:
model vs theory
‘Theories, as they are usually understood by philosophers, make claims about the world [...]
Models, in my sense, do not themselves say anything about the world.
Models are structures that can be used by scientists to say various different things about the world,
by means of commentaries that accompany models but are distinct from them’
Godfrey-Smith, 2005 p. 4
‘Two scientists can use the same model to help with the same target system while having quite different views of how the model might be representing the target system. I will describe this situation by saying that the two scientists have different construals of the model’
‘one scientist might [construe] some model simply as an input-output device, as a predictive tool.
Another might [construe] the same model as a faithful map of the inner workings of the target system’
Godfrey-Smith, 2005 p. 4
‘Basic facility with the folk-psychological model does not require using a particular construal of it. Many construals are possible.
And it is also possible to have facility with the model, and have a sense of which target systems are appropriate for it, while not having much of a construal at all’
Godfrey-Smith, 2005 p. 5
Decision Theory (like Game Theory) specifies a model.
Models can be construed in several different ways.
Decision Theory says nothing about how the model should be construed.
What construals of Decision Theory might be useful?
Decision Theory (like Game Theory) specifies a model.
Models can be construed in several different ways.
Decision Theory says nothing about how the model should be construed.
1. Applications range from microbial populations to countries.
2. The explanations are of the same type in every case.
3. The underlying processes probably differ.
4. Therefore, game theory is agnostic about processes.
Decision Theory (like Game Theory) specifies a model.
Models can be construed in several different ways.
Decision Theory says nothing about how the model should be construed.
tension
The range of applications of Decision Theory shows it is agnostic about processes.
Dickinson’s instrumental process is characterised by appeal to Decision Theory.
conclusion
We need a dual-process theory of (goal-selection for) action.
Decision Theory (like Game Theory) specifies a model.
Models can be construed in several different ways.
Decision Theory says nothing about how the model should be construed.
Decision Theory provides a model that characterises the instrumental goal-selection process
appendix
Klossek & Dickinson, 2012 figure 1a
Klossek & Dickinson, 2012 figure 2
Training Effects (Klossek, Yu & Dickinson, 2011)
Subjects: 3-4 year olds
Training:
Choice Group : perform Action1 to see Clip1 or Action2 to see Clip2
Single-Action Group : only one action is available at once
(Frequency of Action1 and Action2 is matched across groups!)
Devalue Clip1 (expose to satiety)
Test: both actions available. What do Ss select?
Results:
Choice group selects Action2
Single-Action Group selects Action1 and Action2 equally
Klossek et al, 2011 figure 1
Klossek et al, 2011 figure 2
Whether you learn about the effects of an action
can influence
whether that action becomes dominated by instrumental or habitual processes.