From Narrow AI to Artificial General Intelligence (AGI)

IntelliFest 2012
International Conference on Reasoning Technologies
INTELLIGENCE IN THE CLOUD
From Classical (Narrow) AI to
AGI
Helgi Helgason
Perseptio

Helgi Páll Helgason
helgi@perseptio.com
In collaboration with: Dr. Kristinn R. Thórisson & Eric Nivel
Intellifest 2012
AI researcher, Ph.D. candidate
Center for Analysis and Design
of Intelligent Agents,
Reykjavik University
Founder / CEO,
Perseptio

Narrow AI
AGI
Constructivist AI
HUMANOBS Project
AGI & Cloud Computing
Intellifest 2012

 Displaying human-like behavior?
 Solving computationally complex problems (in
some unspecified amount of time)?
 Performing isolated tasks that have
conventionally required humans?
 Adapting to a complex, dynamic environment
with insufficient knowledge and resources?
Intellifest 2012

Intellifest 2012
Nature’s way of dealing with complexity
under resource and time constraints
No real-world intelligence exists that does
not address the passage of time head on

Intellifest 2012
“Intelligence is the capacity of a system
to adapt to its environment while
operating with insufficient
knowledge and resources.”
- Pei Wang
(Rigid flexibility: The Logic of Intelligence. Springer 2006)

Intellifest 2012
“If either time or computational
resources are infinite, intelligence is
irrelevant .”
- Dr. Kristinn R. Thórisson

Intellifest 2012
Time constraints
Abundant information Limited resources
ATTENTION

Intellifest 2012
Time constraints
Abundant information Limited resources
INTELLIGENCE

“Narrow” (classical) AI:
• Systems explicitly designed to solve
specific, reasonably well-defined problems
 E.g. Deep Blue, Watson, etc.
Artificial General Intelligence:
• Systems designed to autonomously learn new
tasks and adapt to changing environments
Intellifest 2012

 Neural-network based virus detection system
 Personalized movie recommendation system
 Software for prediction, classification and
pattern recognition
 NLP-based personalized news delivery
software
 Development and implementation of
algorithmic trading strategies using AI
techniques
Intellifest 2012

 Constitutes vast majority of work in field of AI to-date
 While proven useful in industry beyond doubt, by
definition unlikely to lead to human-level intelligence
• Highly unlikely that isolated bits and pieces of the “intelligence
puzzle” can somehow be fused into holistic intelligence
 Difficult to generalize for new problems
 Task environments are typically limited or simplified
representations of the real world
 Substantial adaption to tasks not required
Intellifest 2012

 AGI (Artificial General Intelligence):
• Relatively small group of researchers (so far)
refocusing on an original idea behind AI:
human-level AI
• First AGI conference: 2008
• Systems explicitly designed to autonomously learn
novel tasks and adapt to changing environments
• Ultimately targets human-level intelligence (and
beyond) in real world environments
Intellifest 2012

AGI (Artificial General Intelligence):
• Long-term research effort
• Challenging funding situation
• Results are not guaranteed
Intellifest 2012

Realworld
environment
Intellifest 2012
Sensors
Actuators
Data
Processes

Intellifest 2012
Sensors
Actuators
Data
Processes

Realworld
environment
Intellifest 2012
Sensors
Actuators
Data
Processes
Sensors
Actuators
Data
Processes

 Benefits
• Versatile, highly adaptive and autonomous systems
• Reduced design/development cost due to high reusability
 Downsides
• Predictability and determinism are sacrificed to some
degree
• Require learning phase before coming practically useful
Intellifest 2012

Intellifest 2012
 Software systems approaching human-level
intelligence will be massive and complex
• Significantly greater complexity than exists in current
software systems
• Cognitive limitations of designers/programmers
 Realistic to believe we will build such systems
using current software methodologies?
• Manual construction
• Coarse grained modular systems
• Divide-and-conquer

Intellifest 2012
Con - struct - ionist A.I.: Manually-built artificial
intelligence systems; learning restricted to combining
predefined situations and tasks, from detailed
specifications provided by a human programmer.
“The programmer as construction worker”

Intellifest 2012
Underlying assumptions:
• Systems of reasonable intelligence can be built
with an architecture of a few thousand manually
constructed modules
• Such a system could automatically:
 Tune parameters as needed
 Route information and control among the modules

Intellifest 2012
Reality:
• Intelligent systems are (functionally)
 Heterogeneous
 Large
 Densely-coupled
 Self-reflective
• Intelligence is the product of the operation of a
system

Intellifest 2012
Reality:
• Massive, complex dependencies exclude:
 manual construction of modules
 modular construction
 piecewise composition where each piece built in isolation
 Exceedingly large functional state-space
 Subdivision hides important interconnections
• Divide-and-conquer fails

Intellifest 2012
Available evidence strongly indicates that the
power of general intelligence, arising from a
high degree of architectural plasticity, is of a
complexity well beyond the maximum reach
of traditional software methodologies.

Intellifest 2012
“We’ve still got a
couple of years to
go before we’re
ready for the
moon.”

Intellifest 2012
Con - struct - ivist A.I.: Self-constructive artificial
intelligence systems with general knowledge
acquisition skills; systems develop from a seed
specification; capable of learning to perceive and act
in a wide range of novel tasks, situations and
domains.
Thórisson, K. R. (2009). From Constructionist to Constructivist A.I. Keynote, AAAI Fall
Symposium Series: Biologically Inspired Cognitive Architectures, Washington D.C., Nov. 5-
7, 175-183. AAAI Tech Report FS-09-01, AAAI press, Menlo Park, CA.

Intellifest 2012
 Developmental approach
 Targets a ratio of hand-coded to auto-
generated programs of magnitude 1:1000000
and up
 Requires:
• Small building blocks (peewee-size)
 Facilitates transfer and re-use between tasks and
domains
 Fast, (temporally) predictable execution

Intellifest 2012
 Standard programming languages
• Designed for humans
• Complex operational semantics
• Not suited for automatic self-programming
• No (explicit) temporal grounding
 Constructivist AI needs a new paradigm:
• Transparent operational semantics
• Machine-understandable language
• Explicit temporal grounding
• Self-organizing management
• Uniform representation

Intellifest 2012
 Fundamental to AGI systems
• All cognitive actions take time, impacting the system’s
place in the context of the real world
 Resource management
• Constant operating scenario: abundant information,
limited resources
• Range of time-constraints
 Many posed by the environment
 Temporal dimension of knowledge is
important
• Past events
• (Expected) future events

Intellifest 2012
System must be able to
• Predict the effects and side-effects of its actions in
the world
• Predict the effects and side-effects of its own
internal operations
Requires uniform representation
• That includes time at atomic operation level

Intellifest 2012
Humanoids that Learn Socio-Comunnicative Skills Through Observation
Funded by European Union 7th Framework Programme
Coordinator / Principal Investigator: Dr. Kristinn R. Thórisson

Intellifest 2012
Target domain: TV-style interview
Two roles: Interviewee, Interviewer
Scenario:
• Two humanoid avatars and props in a virtual 3D
environment
• Multiple modalities involved in perception and
control of an avatar
 Speech, intonation, gaze, head movements, hand
movements

Intellifest 2012
Developmental operation:
• 1. Motor babbling phase
 System tries out its actuators and builds models of how
it can impact the environment
• 2. Observation phase:
 Humans control both avatars and perform interview
while system observes
• 3. Operation phase:
 System takes over one of the roles

Intellifest 2012
 New AGI architecture developed for this
project:
Autocatalytic
Endogenous
Reflective
Architecture

Intellifest 2012
 Broad-scope, general-purpose architecture
addressing:
• Perception, decision, motor control
• General-purpose action learning in dynamic worlds
 Tiny construction components
• Relative to size of architecture
 Support of transversal cognitive skills, at multiple
levels of granularity and abstraction
• System-wide learning
• Temporal grounding
• Observation and imitation of complex realtime events
• Inference, abstraction, prediction, simulation ... and more

Intellifest 2012
Recursive
• Meta-control

Intellifest 2012
Replicode: New programming language
• Developed specifically for HUMANOBS/AERA
 Open source
• Designed to build model-based, model-driven
production systems that can modify their own code
 Containing a (very) large number of
concurrent, interacting programs

Intellifest 2012
 New programming language
• Encodes short parallel programs and executable models
• Explicit temporal grounding
 Soft realtime
• Data-driven execution model
 Computation based on pattern matching
• No explicit conditional statements (if-then) or loops
• All executable code runs concurrently

Intellifest 2012
 New programming language
• Code can be active or inactive
• Code can be input for some other code
• Dynamic code production
• Execution feedback
• Supports distribution of computation and knowledge across
clusters of computing nodes

Intellifest 2012
[FACT@T0: BOX AT POS (0,0)]
[CMD@T0: MOVE BOX (1, 0)]
[PRED@T1: BOX AT POS (1,0)]
[GOAL@T1: BOX AT POS (0,0)]
[FACT@T0: BOX AT POS (1,0)]
[CMD@T1: MOVE BOX (-1,0)]
PREDICTION
Model
PRESCRIBE ACTION

Intellifest 2012
_start:(pgm
|[]
|[]
[]
(inj []
p:(pgm
|[]
|[]
[]
(inj []
(mk.val self position (vec3 1 2 3) 1)
[SYNC_FRONT ( (+ now 10000)) 1 forever root nil]
)
(mod [this.vw.act -1])
1
)
[SYNC_FRONT now 1 forever root nil]
)
(inj []
(ins p |[] RUN_ALWAYS 50000us NOTIFY)
[SYNC_FRONT now 1 forever root nil 1]
)
1
)
|[]
i_start:(ipgm _start |[] RUN_ONCE 90000us NOTIFY 1)
[]
[SYNC_FRONT now 1 1 root nil 1]

Intellifest 2012
PROTOTYPE
DEMO
Verbally directed object manipultaion

Intellifest 2012
In the domain of (generally) intelligent
systems, the management of system
resources is typically called “attention”
Critical (and neglected) issue for AGI
• Systems constantly working with limited resources
under time constraints in environments providing
abundant information

Intellifest 2012
 Design of AGI systems needs to address
practical limitations from the outset
 AGI systems will face time-constraints and
need to be reactive and interruptible, yet
capable of planning
 Retrofitting AGI systems with resource
management highly challenging
• Duration of atomic operations becomes important

Intellifest 2012
 Control mechanism responsible for prioritizing data
and processess
 Targets equally
• External information (from the environment)
• Internal information (from within the system)
 General, no assumptions about
• Tasks
• Environments
• Modalities / Embodiment
 Adaptive
• Learns to improve itself based on experience

Attentional
patterns
Matching
Data items
Processes
Top-down
Bottom-up
Contextualized
process
performance
history
Contextual process
evaluation
Experience-based
process activation
Sensory
devices
Environment
(Real world)
Actuation
devices
Commands
Sampled data
Data
biasing
Goals / Predictions
Derived
Bottom-up
attentional
processess
Evaluation
Process
biasing
Data -> Process
mapping

Intellifest 2012
 Potential:
• Knowledge sharing
 Systems learning not just from their own experience, but
from the experience of other identical (or similar) systems
 On-demand access to knowledge bases & services
• Distributed resources
 Systems using computational resources they do not
physically contain
• Remote on-demand sensing

Intellifest 2012
 Possible limitations:
• Communication latency
 Operations involving network communication may introduce
time delays, which may be significant in terms of operation
• Communication bandwidth
 Sensory information can be a significant amount of data (100
MB+/sec)
• Cloud server load
 Response times for cloud-based operations less predictable

Intellifest 2012
 Interesting directions:
• Augment system knowledge from cloud during idle
time
• Compress and/or pre-process sensory information and
run cognitive processes in the cloud
 Latency still an issue
• Cloud-based services used as specialized tools
 When allowed for by temporal constraints

Intellifest 2012
Practically difficult:
• Distributing cognitive processes between system’s
own hardware and the cloud
 AGI systems likely to have a very large number of
components with rich, complex interconnections and
interactions
 Communication latency becomes a major issue

Intellifest 2012
Onboard cognitive
resources
Cloud cognitive
resources
Network barrier

Intellifest 2012
Perception
• Computer vision
• Image processing
• Feature detection
• Speech recognition

Intellifest 2012
System design
• More general & reusable systems
• Temporal issues
• Resource management

Intellifest 2012
 HUMANOBS project
• http://www.humanobs.org
 Replicode
• http://wiki.humanobs.org/public:replicode:replicode-main
 Publications
• From Constructionist to Constructivist A.I.
 Kristinn R. Thórisson (2009)
• Cognitive Architecture and Autonomy: A Comparative Review
 Kristinn R. Thórisson, Helgi Páll Helgason (2011)
 AGI 2012
• The Fifth Conference on AGI, Oxford, UK, Dec 8-11 2012
• http://agi-conference.org/2012/

From Narrow AI to Artificial General Intelligence (AGI)

More Related Content

What's hot (20)

Viewers also liked (7)

Similar to From Narrow AI to Artificial General Intelligence (AGI) (20)

Recently uploaded (20)

From Narrow AI to Artificial General Intelligence (AGI)

Editor's Notes