Further Cognitive Systems

1
Further Cognitive Systems

• Learning
• Environmental interaction
• Artificial cognition?
• Current cognitive systems
• Science-fiction v fact
• Architectures
• Perception, Representation, Reasoning, Learning
  Action
• Learning Cognitive Systems
• Problems in LCS
• Advances in LCS

2
LCS for Cognitive Robotics
Latent learning in Tolmans T-maze Possible
through ACSII (LCS architecture), but slow 3
months for real experiments! Something
missing? What traits (shortcuts) through
generations? Cognitive and Conscious
inspiration sought
3
LCS for Cognitive Robotics
What went wrong? Large rule base is slow to
use No higher order patterns formed Serial,
linear, repetitive Stuck in unknown, but similar
situations Could sense, represent,
reason (including planning) learn act
but much missing
4
Inspiration for Cognitive Robotics

• Abstraction - higher order pattern learning
• Emotions - mimic capabilities
• Small-world theory and Synchronisation scaling
• Dissonance theory - Aversed to confusion
• Consciousness studies behaviours
• (qualia, explanatory gap, frame problem, )

5
Inspiration for Cognitive Robotics
Psychology
Anthropology
Philosophical
Emotions
Neuroscience Not as clear cut as it seems

• Cybernetics
• Control
• Communication

• Folksonomy
• What people create to be true

6
Emotions
Many interesting descriptions, functions and
notions, but not much agreement, e.g. emotions

• Give rise to affective experiences, such as
  feelings or arousal
• Generate cognitive processes
• Activate widespread physiological adjustments to
  the arousing conditions
• Lead to behaviour that is often, but not always,
  expressive, goal directed and adaptive.
  Michaud et al. 01

• Agreement on what emotions are not
• Simple descriptions after the event
• Production rules
• 'if state... then emotion...' symbolic
  statements

7
Emotions - Neuroscience
Rolls, defines emotions as the states elicited
by reward and punishers, including changes in
those rewards and punishments, by which a
reward is anything for which an animal will work,
i.e. a Positive Reinforcer and a punishment is
anything that an animal will work to escape or
avoid, i.e. a Negative Reinforcer. Damasio,
stimulus good/bad evaulation Emotions are
based on experiences and help guide future
actions.
8
Emotions in Production Rule systems
How to take benefits of production rule systems

• Reinforcement learning
• Global search
• Transparency

• And make compatible with emotional concepts for
  added benefits, e.g. Rolls 99
• changing autonomic systems, flexibility of
  behavioural response to reinforcing stimuli,
  triggering motivated behaviours, communication,
  social bonding, improving survival, affecting
  cognitive processing (mood congruence) and
  facilitating its continuity, facilitating memory
  storage, allowing persistence of motivated
  behaviours, facilitating memory recall

9
Needs (from Desires)
ssi , sr
a
Pleasure (reward) in satisfying a need (Q)
Desire used to switch needs (W)
Needs and desires
Environment
10
Emotions for Cognitive Robotics

• Emotions are not
• independent or
• linear
• Emotions are
• linked
• dynamic
• constrained
• attractor states.
• Temporality joy/sadness
• Identity acceptance/rejection
• Hierarchy anger/fear
• Territorality expectation/surprise
  Plutchik 91

Brain is a dynamic system with multiple causes
interacting over nested timescales
Smith 03
11
Other Emotional Architectures

• ASD (Maes 1990) Action selection dynamics
• ALEC (Gadanho 2002) Asynchronous Learning by
  Emotions and Cognition
• EBII (Gadanho 2003) Emotion-based architecture
• DARE (Marcia et al. 2001) Development of
  emotion-based robotic agents
• AD with ECS (Malfaz et al. 04) emotional control
  
• Motives (goals) interact with beliefs
  (predictions) to produce emotions.
• Cf. Sloman
• Feelings arise after enaction of emotions.

12
Augmented Reality
13
Nature
14
Emotional Architectures
15
Augmented Reality Normal Explore
16
Augmented Reality Tight Explore
17
Augmented Reality Emotional Explore
18
Explore task
Simple Environment Complex Environment
Emotional Control Benchmark
19
Explore effectiveness
Benchmark Emotional Control
20
Nurture
Now allow LCS to learn emotional rules (s-e as
before) If ((18ltHappinesslt88)
(6ltSadnesslt23) (12ltCuriositylt81)
(17ltAngerlt85) (17ltHopelt85) (8ltFearlt52))
Then Action 1 This corresponds to the
behaviour If agent is in the open space, then
forwards slowly
Simple Environment Run 4
Run 9
21
Discussions

• What makes emotions useful?
• Emerge rather than hard coded
• Generalise across known and unknown situations
• Episodic and temporal
• Fast response if necessary
• Non-linear, non-deterministic and stochastic
• Analogy
• Emotions act as a warehouse where items can be
  sorted, grouped and sent appropriately.
• Rather than every individual manufacturers
  delivering to each outlet shop.

22
Discussions

• What makes emotions useful?
• Practically
• Much less lines of code for better behaviour -
  40
• Easier to understand code for non-programmers
• Intuitive behaviours result
• Difficulties
• Emotions levels require tuning
• Diagnosis and prediction of behaviour difficult
• When is it justified to call a non-linear
  controller an emotion?

23
Concluding Remarks

• Experimental
• Robot-emotions useful for successfully completing
  real-world tasks.
• Emotions can set goals balance explore vs
  exploit
• Emotions can modify existing behaviours
• Emotions facilitate action in unknown domains
• Thought Experimental
• A need for predictive certainty coupled with
  emotional communication is postulated to lead to
  an agent approaching conscious behaviours when
  placing itself in an out-there world