Transcript PowerPoint - Sergio Pissanetzky

Emergent Inference,
or
How can a program become
a self-programming AGI system?
Sergio Pissanetzky
Self-programming Workshop
AGI-11
Overview of Emergent Inference
► 1850
● Helmholtz discovered Unconscious Inference.
► 2005
● Motivation: Interest in refactoring.
● Scope: Refactoring is universal.
● Approach: Computational experiments.
► Discoveries:
● Partially ordered sets.
● Emergent inference.
● Emergence in complex dynamical systems.
● EI happens in the brain.
Brain experiments
senses and
afferent nerves
knowledge
partially
ordered set
brain
feedback
emergent
inference
feedback
natural
structures
compare
predicted
structures
The first experiment
PROGRAM
(SCRAMBLED)
a=α*χ
b=δ*μ
c=α*ψ
d=δ*ω
e=β*φ
f= β*χ
g=β*ψ
h=α*φ
i= δ*ν
j= ν+e
k=h+i
l =a+b
m=μ+f
n=c +d
p=θ+n
q=ω+g
r=π+l
s=ρ+k
CANONICAL MATRIX
a
b
c
d
e
f
g
h
i
A
j
A A
k
A A
l
A
m
A A
n
A p
A
q
A
A
r
s
The result from the first experiment
REFACTORED
PROGRAM
d=δ*ω
c=α*ψ
n=c +d
p=θ+n
f= β*χ
m=μ+f
b=δ*μ
a=α*χ
l= a+b
r=π+l
e=β*φ
j= ν+e
i= δ*ν
h=α*φ
k=h+i
s=ρ+k
g=β*ψ
q=ω+g
CANONICAL MATRIX (STRUCTURED)
d
c
A A n
A p
f
A m
b
a
A A
l
A r
e
A
j
i
h
A A k
A s
g
A q
This process is emergent inference
The importance of this discovery:
● it is a rigorous mathematical solution obtained
from first principles, not a phenomenological
guess or an engineering compromise.
● it is universal
● it is a side effect of an unrelated process
● it requires no domain-specific knowledge
● it is not man-made
● it is ready for use in computers
Claim
Any system has a natural hierarchical structure
that can be found by emergent inference.
Conjectures
Emergent Inference explains emergence and
self-organization in complex dynamical systems.
Emergent inference in the brain gives rise to
intelligence.
Representing systems as partially ordered sets
Any system of Mathematics, Physics, Chemistry, Biology, Engineering, CS,...
• Model, theory, equations.
• z = f(x, y)
Set = {x, y, z}
Partial order = {x < z, y < z}
A computer program.
• Parser.
CFS Brain model = neural network + reduce energy consumption.
• C = interconnections  memory
• F = fire
 behavior
• S = shrink
 intelligence, emotions, creativity.
• Clustering takes place. Iteration forms clusters of clusters.
• Clusters are neural cliques, cortical columns, cortical modules.
Traditional software development cycle
BRAIN (human analyst)
knowledge
emergent
inference
emergent
inference
emergent
inference
structure
structure
•••
stream of
experience
data
structure
PROGRAM
•••
Source of
Intelligence
Autonomy
Appeal
Knowledge
base
frozen
human
some
practical
conventional
Narrow
AI
N
frozen
human
some
futuristic
conventional
Hybrid
frozen human
Y
+
+
adaptive machine
some
AGI
(future)
Y adaptive machine
full
Emergent
Inference
Regular
N
program
Category
Structures
or rules
Software Categorization
conventional
revolutionary
+
posets
singularity
partially
ordered sets
Traditional AI and AGI
car position
sensors
car driving
program
car
controls
stage sensors,
actors
stage control
program
stage
controls
chess
sensors
chess playing
program
chess
controls
There is no integration, and no refactoring
The brain
car
stage
chess
drive car
senses
human
brain
manage stage
play chess
The brain integrates and refactors naturally
Emergent inference
problem
of Physics
law of Physics
raw image
image
recognition
emergent
inference
token ring
(in C)
OO program
classes, objects
interdependent
tasks
parallel program
EI integrates and refactors naturally
We need a principle for intelligence
Aeronautical Engineering.
- 1800: lift force identified as the principle of flight.
Software engineering.
- 1980’s: the automation of objects.
- 1990’s: the automation of refactoring.
Artificial intelligence.
- 2000’s: the automation of integration.
- 2010’s: the automation of self-programming.
Neuroscience.
- “the exact way in which the brain enables thought is one
of the great mysteries of science.” (Russell-Norvig).
- “we are still a long way from understanding how
cognitive processes actually work.” (Russell-Norvig) .
Emergent inference is the principle for intelligence
Conclusions
● Self-programming can not be achieved without AGI.
● AGI can not be achieved by writing programs.
● EI is the principle for intelligence, AGI and selfprogramming.