Transcript ppt - CSE, IIT Bombay

Artificial Consciousness
Group 3
Mohit, Raghuvar, Avin, Aniruddha
What is Consciousness?

Consciousness denotes being awake and
responsive to one's environment; this
contrasts with being asleep or being in a
coma

Being self-aware

Derives from Latin conscienta which primarily
means moral conscience
Aspects of Consciousness

Awareness


Anticipation


Awareness of one's environment and one's own
existence
The ability to predict foreseeable events
Learning

The ability to learn from experience
Aspects of Consciousness (contd)

Sentience


Utilization of sensory organs
Sapience

Ability of an entity to act with judgment
Types of consciousness

Phenomenal



Experience (Qualia)
Hard part
Access




Verbal report
Reasoning
Control of behavior
Easy part
Hard Problem of Consciousness

Explain how flows and electrochemical
processes in the brain give rise to the inner
experience of subjective awareness.

Neural correlate of Consciousness
Artificial Consciousness (AC)

Consciousness in machine

Engineering artifacts which possess




Self awareness
Wisdom
Ability to learn
Ability to Sense
Why Artificial Consciousness?

Creating machines resembling human beings

Understanding the nature of consciousness

Implementing more efficient control systems.
Goals of Artificial Consciousness

Twofold target in Artificial Consciousness:



The nature of phenomenal consciousness
The active role of consciousness in controlling
and planning the behaviour of an agent
Important Question

Is Artificial Consciousness possible?
Outline

Raghuvar


Avin


Why AC may not be possible?
How to build a Conscious System?
Aniruddha

Applications and Future Work
Several key questions...




Can computers think or do they just
calculate?
Is consciousness a human prerogative?
Does consciousness depend on the material
that comprises the human brain, or can
computer hardware replicate consciousness?
Is simulation of consciousness the same as
consciousness?
Turing test




Test for machine’s ability to demonstrate
intelligence
Machine and human in separate room
converse with a human interrogator
Goal is to convince him that you are human
Annual competition : Loebner Prize
($100,000) to computer that passes : no
winners yet!
Turing Test :implications

Main argument:



If a computer can portray itself as a human
successfully, then essentially there is no
difference between a human and a computer
Ability to actively process linguistic syntax,
infer context, have the notion of ‘self’, deal
with a finite knowledge base and moral
questions
The computer is conscious!!....or is it?
Chinese Room
Chinese Room




Person inside the room is a non-speaker of
Chinese
He is given Chinese symbols corresponding
to questions
Instruction book in English telling him which
symbol he has to output corresponding to the
input
To the outside world, he’s an excellent
speaker of Chinese
Searle, 1984




Brains cause minds.
Syntax is not sufficient for semantics.
Computer programs are entirely defined by
their formal, or syntactical, structure.
Minds have mental contents; specifically, they
have semantic contents.
Searle’s argument
(1)
(2)
(3)
If Strong AI is true, then there is a program for Chinese
such that if any computing system runs that program,
that system thereby comes to understand Chinese
I could run a program for Chinese without thereby
coming to understand Chinese.
Therefore Strong AI is false.
Widely debated by philosophers and scientists.
No conclusion as to whether it is possible to
develop a truly conscious machine
Computational Barrier

Minimum level of computation necessary to achieve
human like consciousness: can be calculated using
Neural networks
 The human brain has about 1012 neurons, and each
neuron makes about 103 synaptic connections with
other neurons, on average, for a total of 1015
synapses.
 Neural Network : each synapse : 4bytes
 1015 synapses : 4million GB
 Add auxiliary variables : 5 million GB !
 Minimum prerequisite : Necessary, not sufficient
condition
Consciousness System - Imitation and
Self-others distinction




System for generating consciousness
Implementation of imitation behavior
important as the first stage of study of
consciousness
Distinguish between self and others
Consciousness generated by consistency of
cognition and behavior
Structure of a Consciousness System
Consciousness
System comprises of :

a)
b)
c)
d)
Cognition System
Behavior System
Primary
Representation
Symbolic
Representation
Cognition System




Information enters the consciousness system
at Input and reaches the Cognition System
Used for neural computation together with
the information from BL (internal
information)
Derived information transmitted to RL
(cognized information)
Language labels have different areas for
condition of self and others
Behavior System



Behavior information from BL passes to the
Behavior System
Information added to input and sensory
information by neural computation
Resultant information is transmitted to
Output to implement a behavior
Primary Representation




Common area for the cognition and behavior
system
Behavior learning during cognition and
cognition learning during behavior
Each piece of information is correctly related
to language labels of symbolic representation
The consciousness system brings a process of
artificial thoughts as information is
circulated
Example – Human Language





Conversation between self and other
Input: the speech of both self and the other
Somatic sensation that self is talking is fed
back as input
Behavior of a new language label (through
circulation) gives rise to thinking and
expectations
Possible to offer new topics for conversation.
Consciousness through Imitation

Mirror neurons





Brain of monkeys
Neuron fires when implementing a certain behavior by
itself or upon observing others with the same behavior
Primary representation equivalent to mirror neuron
Imitation occurs while information circulates
through primary and symbolic representation
Circulation of information through external models
and one’s own brain is necessary for imitation
Learning in a Consciousness System
(b)
c)
Neural Network (NN) of a conscious system
Features of NN in a consciousness system

Two structural features to implement consistency of
cognition and behavior




Recursiveness - Somatic sensation of behavior of self is fed
back to enhance learning efficiency
Presence of a common area for cognition and behavior
and data circulation
Imitation learning and cognition of behavior of self
and others possible
Back propagation (BP) method of supervised
learning used for NN learning through simulation
Ongoing Research

Hot research area in AI

Several serious attempts to make
conscious machines




Franklin’s Intelligent Distribution Agent
Ron Sun's cognitive architecture CLARION
Haikonen’s cognitive architecture
Cognitive Robots
Intelligent Distribution Agent





Negotiate new assignments for sailors in the
US Navy
Interacts with Navy databases and
communicates with the sailors via natural
language email dialog
Based on Global Workspace Theory - Baars
Relies heavily on codelets
Not attributed as conscious
CLARION





Fundamental structures of the human mind
Distinction between conscious and unconscious
mental processes
Bottom-up learning - learning that involves acquiring
first implicit knowledge and then acquiring explicit
knowledge on its basis)
Successful in accounting for a variety of psychological
data
Skill learning tasks : SRT, AGL, PC, CI
Haikonen’s cognitive architecture





Rule-based computing inadequate for
achieving AC
A special cognitive architecture
Artificial neuron
A low-complexity implementation of the
architecture proposed (2004)
Not capable of AC – but exhibit emotions as
expected
Conscious Robots
Evolution of Conscious Robots:
 Simple reflex
 Simple reflex with memory
 Perception with meaning and associative
memory
 Perception with associative memory and
report
 The robot perceives itself perceiving
Looking Ahead

Robots

References in Movies


The Matrix
I, Robot
Cognitive Robots

Sensors, path planning, and manipulator design and
control vs ability to reason, act and perceive

Able to “know what they are doing”

Consciousness per se does not make any sense
unless accompanied by end to end processes like
perception and behavior.

Future cognitive robots will be able to interact with
humans, acting and learning in unpredictable
environments.
Cognitive Robots Applications






Service robots, social robots, and personal
assistance robots (like Maggie)
Assistant robots for disabled people (like Asibot)
Human-like performing robots (like Manfred)
Autonomous space exploration robots and robotic
assistance for orbital assembly and repair (like
Robonaut)
Robocup players
Autonomous emergency and rescue robots (like
BEAR)
Kismet, the Sociable Robot
AIBO, Robot pet dog
Open questions




Someday, when machines claim they are
conscious, will we believe them?
Will we be able to “test” consciousness?
Could machines become “deluded” that they
are conscious?
Could the World Wide Web be a form of
Consciousness?
Summary




Artificial consciousness will lead to intelligent
and more useful agents
Difficult to understand and validate
Attempts to model consciousness through a
neural network
Field of intensive research and philosophical
debate in the years to come
References





Block, N. On a confusion about a function of consciousness.
Behavioral and Brain Sciences 18 (2): 227-287, 1995
Buttazzo, G. Artificial Consciousness: Utopia or Real
Possibility? Spectrum IEEE Computer 18: 24-30, 2001
Conscious Robot That Distinguishes between Self and
Others and Implements Imitation Behavior
18th International Conference on Industrial and Engineering
Applications of Artificial Intelligence and Expert Systems,2005
Franklin ,S. IDA: A Conscious Artifact? Journal of
Consciousness Studies 10:47-66,2003
Wikipedia