Transcript The Artificial life approach - LIRIS

Artificial life
Based on Luc Steels (1995)
Subject
• Study :
 research and synthesis towards the artificial life domain
• Context :
 limits of system expert
 growth of computer power
 cognition approach
2
Start point
• Scientific article :
« The Homo Cyber Sapiens, the Robot Homonidus
Intelligens, and the ‘artificial life’ approach to artificial
intelligence »
Luc Steels (1995)
3
Luc Steels
• Specialized in the domain of artificial intelligence and
artificial life applied to robot architectures and to the
study of language
Fig 1. Luc Steels
4
Luc Steels’ background
• Studied computer science at MIT
(Massachusetts Institute of Technology – USA)
• Director of Sony Computer Science Laboratory in Paris
• Professor computer science at the University of Brussels
• Founded the VUB AI Laboratory (1983)
• Reviewer at CNRS
5
Once upon a time…
Bionic man
?
Intelligent
systems
Artificial life
evolution
Homo
Erectus
Homo
Sapiens
Us
After us ?
6
Axes of discussion
• Bionic man or Homo cyber sapiens
• Intelligent systems or Robot Homonidus Intelligens
• Artificial life
7
Artificial Life
Bionic man
or Homo Cyber Sapiens
Homo Cyber Sapiens
• Intelligence evolving towards greater :
 sophistication
 power
• Homo Cyber Sapiens
↔ technological extensions of the human brain.
9
Homo Cyber Sapiens
• Artificial brain extensions should mimic the operation of
human neurophysiology.
 Neural modeling is implemented in chips
• Artificial brain may be completely different from natural
brain.
 The build of bridges will establish data communication and
processing.
10
History
• Brief History of Homo Cyber Sapiens/Post Humans.
• Mary Shelley : Frankenstein (1831)
• K.Eric Drexler (1980-1990) : Nanotechnology
11
Evolution of Super Computer
•
Brain versus Super Computers
 Ian Pearson, Chris Winter & Peter Cochrane (1995)
Fig.1 Projection of
supercomputer speed
12
Use Case
• Two Examples :
13
Artificial Life
Intelligent Systems
or Robot Homonidus Intelligens
Intelligent systems
• Cybernetic and Artificial Intelligence :
already 50 years of experiment
• Many advantages for computer science
• A whole range of programs exhibit features of human
intelligence
• But …
15
Limits of Intelligent systems
• Steels : 3 strong limits of Intelligent systems
 a ‘frozen intelligence’ and not an intelligent behavior
 intelligence needs to be embodied
 consciousness
16
First limit : frozen intelligence
• Expensive cost of construction
• Ephemeral validity
• Outdated by changes
• Expensive and unrealistic maintenance
Something more than knowledge needed to be intelligent
17
Second limit : lack of embodiment
• Knowledge systems :
 disembodied intelligence
 no direct link to the real world
• Intelligent behavior emerges from interactions
• Difficulties :
 link between the real world and the system symbols
 adaptation to unforeseen actions
18
Third limit : consciousness
• An intelligent system needs a sense of self and a
conscience
 Possible ?
 Existence of a true autonomous agent ?
19
State of research in 1995
• No technological obstacle
• The real obstacle : the lack of a theory of intelligence
20
State of research in 2005 (1/2)
• Knowledge systems : example of ‘frozen intelligence’
• Case Based Reasoning use the last experience
• Multi-agent systems :
 agents
 environment
 interactions
Fig 1. A robot soccer team by
Nikos Vlassis (Amsterdam)
21
State of research in 2005 (2/2)
• McCarthy (1995-2002) :
 consciousness does not yet exist in intelligent system
Intelligent systems
emotions
sub consciousness
consciousness
introspection
22
Artificial Life
The Artificial life approach :
Theoretical approach
Historic (1/2)
2005
Christopher Langton 1987
Connectionism 1980
John Conway 1970
Alan Turing 1948
John Von Neumann 1940
first scientific conference devoted to A-life
parallel, distributed processing, neural networks
AI ↔ cognitive science
game of life : simple system → complex selforganized structures
“ ‘Intelligent machinery’ , It’s the birth of
the concept of intelligent machines.”
cellular automat
24
Historic (2/2)
• Game of life : illustration
Fig 1. Random start
Fig 2. Stable state
25
Definitions of A-life (1/2)
• Langton (1989) :
 Artificial life (A-life) : study of ‘natural’ life by attempting to
recreate biological phenomena from scratch within computers
and other ‘artificial’ media.
• Rennard (2002) :
 Life : state of what is not inert.
 Artificial life : field of research witch intend to specify the
preceding definition.
26
Definitions of A-life (2/2)
• Doyne Farmer and d'A.Belin (1992) : A-Life as field of alive
 An artificial life must :
 be initiated by man
 be autonomous
 be in interaction with its environment
 induce the emergence of behaviors
 Optional :
 capacity to reproduce
 capacities of adaptation
27
Steels’ vision of A-life
• Dynamic system theory applied to Artificial Intelligence
• A-life → Unified theory of cognition
• Unified theory : explain the details of all mechanisms of all
problems within some domain.
 unified theory of cognition domain’s ↔ all cognitive behavior of
humans.
 experimental psychology could support such theories.
(Newell 1990)
28
Steels’ research path
• Two kinds of behavior expected :
 differentiation : individual agent get specific task
 recognition : make the difference between the member of
the group and those which don’t.
recognition → emergence of language.
29
Axes of research (1/2)
• Emergence of language (Steels & Kaplan)
 Emergence of common sense
 Adaptation to other agents
30
Axes of research (2/2)
• Autonomous robotic (Floreano)
 Genetic algorithms with neural networks
 Co-evolution
• Animat Approach (Meyer)
 Synthesizing animal intelligence
 Situated and incarnate cognition
31
Artificial Life
The Artificial life approach :
Experimental approach
Steels’ experimentation – 1995 (1/4)
• A complete artificial ecosystem
• An environment with different pressures for the robots
• Robots are required to do some work which is paid in
energy
• Cooperation and competition with each other
• Behavior systems
33
Steels’ experimentation – 1995 (2/4)
Fig 1. The ecosystem with the charging
station, a robot vehicle, and a competitor
Fig 2. A robot vehicle
34
Steels’ experimentation – 1995 (3/4)
Behavior system
Environment Perception
- Visual Perception Modules
Charging station, Competitors,
Other robots
- Finding resources
- Exploring
- Obstacle avoidance
- Align on charging station
- Align on competitors
- Sensors
Light, Tactile
- Turn left/right, Forward,
Retract, Stop
- Motors
35
Steels’ experimentation – 1995 (4/4)
• Interesting results :
 Behavior diversification
 Hard working gourp
 Lazy group
 Steels : something could emerge from the lazy group
36
Steel’s experimentation – 2001 (1/3)
• One speaker (S), one hearer (H)
• H tries to guess what S is talking about
• H guess wrong : correction (feedback)
• No explicit object designation : simple region pointing
37
Steel’s experimentation – 2001 (2/3)
Fig 3. The talking heads experiment
38
Steel’s experimentation – 2001 (3/3)
• Interesting results :
 Emergence of a shared word
 Winner-take-all
 Shared word repertoires after experiment
39
Other kind of experimentation (1/2)
Floreano & al. (2004)
• Evolution of Spiking Neural Networks in robots
• Objective : Vision-based navigation and wall avoidance
Fig 4. A Khepera robot in a square
arena
Fig 5. A Khepera robot
40
Other kind of experimentation (2/2)
• Interesting results :




Avoiding walls following with security distance
Biologically plausible connection patterns
Forward progression
Self adaptable speed : body adaptation
41
Artificial Life
Conclusion
Conclusion (1/3)
• 3 approaches
 Bionic man : ethic problems
 Intelligent systems : limits
 Artificial life :
 Tremendous possibilities
 Involving many fields, biologically-inspired
 Now a days the biological approach stay in progress.
43
Conclusion (2/3)
• Lack of intelligence theory
• Problem of consciousness in robots
• Is language needed for intelligence ?
• Sufficient pressures for a new species ?
• Does performance gain means Intelligence gain ?
44
Conclusion (3/3)
“Intelligence is like life or cosmos; its such a deep
phenomenon that we will still be trying to understand it
many centuries from now.”
Luc Steels
45
46
Homo Cyber Sapiens
• The Anatomical changes are defined by :
Homo erectus
New sensory modalities.
Homo Sapiens “wise man"
• The Extreme ecological pressures are defined by:
Homo erectus
Homo Sapiens “wise man"
47
Homo Cyber Sapiens
• The human species is today under just as much stress
as it must have been in the past,
Still Human Intelligence haven’t evolved !
• How realistic is the development of a Homo Cyber
Sapiens ?
48