Transcript Emotionality

CE 00875 - 3
Character AI
Diane Bishton – K229
([email protected])
these slides at http://blackboard.staffs.ac.uk
Dimensions of Emotion
1
Introduction
• We will begin with a look at why studying
‘Emotion’ is becoming more important &
how ‘Emotion’ in humans may be defined
• We will also see that although emotions are
so varied:
there are commonalities across cultures
we can induce emotion by thought or
simply by changing our facial expression
2
Why is Study of Emotion so ‘hot’ ?
• Experiments in the early 1990’s showed people with
particular brain lesions who couldn’t experience
emotions had difficulty making rational decisions
Emotionality - a prerequisite of Rational Behaviour
(see rationality etc. in Free Will, (earlier lecture))
• Humans treat computers like people ‘Anthropomorphism’
People make Emotional connection with Computers
• The generation of human-like characteristics in film &
in games e.g. Beowulf (2007), Sims etc.
Synthetic characters develop their own behaviour, &
humans become co-stars
3
Key things about Emotions
 They can, & often do, begin quickly
 It’s difficult for us to control both the triggers of
emotion & the extent to which we react
 Emotion is evident on our face & in our voice
 Many of the expressions are subtle, some are clear
(you’ll have chance to do an experiment soon)
 Each emotion generates unique sensations
 We appear to behave so as to minimise negative
emotions & maximise positive ones
4
Emotion & Feedback
‘Behaviourism’
later)
Received + Omitted _
Stopped !
Emotional States
appear to be
tied to Reward
(R you want) &
Punishment (P
you want to
avoid) (refer to
From Rolls E T
in Trappl et al
(2002)
Emotions can be
elicited by R &
P : received, left
out, or stopped
5
Instrumental Reinforcers (IR)
One definition :
Emotions are ‘States’ that result from ‘Instrumental Reinforcers’
IR “are stimuli that, if their occurrence, termination, or omission is
made contingent upon the making of a response, alter the
probability of the future emission of that response”
(Rolls E T, in Trappl et al (2002) p13)
Reinforcers can be :
Primary (unlearned i.e. innate)
Secondary (learned by Association with Primary Reinforcers,
See Behaviourism, & separate ‘Conditioning’ handout)
Emotions can be produced by Recall of reinforcing events
(see Memory etc) “…emotion is produced by a stimulus… emotions
have an object in the real world… emotional states are
intentional…they are about something.”(Rolls E T, Trappl et al (2002) p15)
Behaviourism
 Developmental stages
 Schedules of reinforcement (an ‘exogenous’, outside influence
on cognitive growth)
 Task analysis - to devise ‘behavioural objectives’, but, what is
mastery learning, & what components ? - expert disagreement
e.g. Octal arithmetic
 Emphasis on Rules, instructional targets
 Covert (hidden) problem-solving steps ignored as irrelevant
 Reinforcement through stimulus-response
 Trained, measurable behaviours, Conditioning (Pavlov)
 ‘Higher’ animals - chimps etc, showed that we can’t generalise
‘Laws’ from a specific Context, nor even for same Subject 7
Functions of Emotion
 Elicitation of Autonomic Response & Endocrine Response i.e. Fight or
Flight
 Flexibility of Behaviour to Reinforcing Stimuli i.e. behaviour goals are
specified by R & P evaluation - different actions can achieve an R or avoid a P
 Motivation & Direction of personal behaviour
 Communication (of one’s emotion to another) to influence their behaviour
 Social Bonding
 Cognitive Evaluation of Events or Memories (via current mood state
(learning))
 Memory storage - especially of episodic memory, may contribute to memory
‘strength’
 Memory Recall - as just one cue or route to ‘a’ memory
(summarised from (Rolls E T, in Trappl et al (2002) pp16-19)
8
Motivations for Giving ‘Machines’
Emotional Ability
• Building robots & synthetic characters that can emulate
living things (see Personification, below)
• Building ‘intelligent’ machines (decision-making)
• To better understand human emotion
• Making interaction less frustrating
via Machines that can identify situations that lead to user
frustration (situated cognition & affective computing)
(Being able to recognise (particularly) user frustration could be a
key step in the development of Adaptive Systems )
• In ‘Personification’ of the interface - all sorts of
applications of Interface Agents , up to and including
Artificial Life (aLife) see the work of Maes, P. etc. 9
Anthropomorphism
• We already endow artefacts & natural
phenomena with human qualities
• Do we want emotional technology ?
• What advantages & disadvantages could
this bring ?
• How far could rationality go without an
emotional component?
• Pentagon ‘ethical’ AI soldiers?
10
Emotional computers
• Designers can engender the appearance of emotion
in interfaces e.g. Microsoft Office assistant can
appear to “Sulk”
• If emotion is about empathy & survival, at what
point do we suspend disbelief & ‘connect’ with the
character ? e.g. Wallace & Gromit ?
• Work has been carried out on a human head avatar
which reacts with similar facial expressions in order
to offer a more realistic interface see e.g. BT Talking
Head
11
Identifying Emotions
• Ekman and Friesen developed FACS
- Facial Action Coding System
• Still used today – uses facial muscles to
quantify emotions
• Plutchik in 1980 argued there were 8 pairs
of basic or primary emotions i.e.
12
13
14
Universal Emotions
Ekman, Friesen and Ellsworth (1972) identified
Six basic human emotions i.e.





Fear
Surprise
Disgust
Anger
Happiness
 Sadness
and Ekman (2003) later added a seventh: Contempt
Not surprisingly, these map on to the 6 innate facial
expressions identified throughout human cultures.
15
Cannon-Bard Theory
• Cannon and Bard in the 1920’s noted that the
same visceral (instinctive/primeval) changes occur
in different emotions
• Thalamus plays a central role
• This sends signals to the Autonomic Nervous
System (ANS) – responsible for unconscious
functions like heart rate and secretion of
hormones such as adrenaline
• and to the Cortex which interprets the situation
i.e.
16
The Cannon–Bard theory.
17
Emotional signs
Noticeable to outside world
• Facial Expression
• Voice Intonation
• Gesture, Movement
• Posture
• Pupil Dilation
Internal effects
• Respiration
• Heart Rate
• Temperature
• Electrodermal Response,
Perspiration
• Muscle Action Potentials
• Blood Pressure
18
Emotion Detection
Attributing a change in facial expression,
vocal intonation and (ideally) a person’s
actions to a specific emotion is difficult at
best and is responsible for many
misunderstandings (take the face test)
Sufferers of Autism have this kind of problem
• What capabilities would a computer require
to recognise emotional states?
19
Currently
• Technology exists to read physiological
changes such as heart rate, skin conductivity
• However the sensors are too intrusive for
everyday use and require expert analysis
• Also, individual physiological signs tend to
indicate a general increase in arousal – GAS
(General Adaptive Syndrome) - rather than
a specific emotion
20
Mirror Neurons – Empathy cells
• What about us ?
• In the early 1990’s Rizzolatti G. & Craighero L.
(2004) identified ‘mirror neurons’ or ‘copycat
cells’ during experiments involving monkeys.
• Mirror neurons show activation of the same neural
pathway that would operate to execute a particular
action whenever an action is observed. Basis for
learning by imitation. Very common in humans.
• They may allow us to read others’ emotions
through the pathways that generate our facial
21
expressions when we express emotion
Summary
• We’ve seen a couple of models of ‘Emotion’ and
some reasons for endowing machines with, at
least, the ability to recognise & respond to
Emotions
• We’ve also seen how intimately emotions are
tied to human existence, and connections have
been made to earlier lectures on NVC & Free
Will.
• Now take the Faces Test
22