Innovation Through Artificial Intelligence

Creativity Machine Paradigm
hopping synaptic perturbations
Simple & Elegant
A noise-driven dialog between at least two artificial neural networks generates new ideas.
Makes Itself Arbitrarily Complex
Boldly Innovating Innovation via Creative Machine Intelligence

(A Feigned Retreat from Complexity)
Stephen L. Thaler, Ph.D. President & CEO, Imagination Engines, Inc. and Founder, In Its Image, Inc. Innovation Demands Boldness Scottsdale, AZ 30 September, 2010 __________________________________
A radically new form of totally self-organizing, autonomous, contemplative, and creative machine intelligence may possibly be the poster child for all disruptive technologies since it may, by definition, generate all subsequent technologies. The lessons cumulatively learned from the advancement of what might arguably be called the ultimate idea should be especially interesting to this workshop since the approaches used have boldly exploited the current retreat from complexity, bypassed conventional review methods, and harnessed the concepts inherent disruptiveness, both technologically and philosophically, as an extremely effective marketing vehicle.
Self-organization inherent to artificial neural networks makes this so.

Scalable Machine Intelligence & Consciousness
ideas
Neural brainstorming session is the basis of human cognition.
opinions
Is that all there is?
The Ultimate Idea

because it generates all others!
Requires utmost boldness to advance, realizing its philosophical implications.
2010, Imagination Engines, Inc.
THE PERCEPTRON raw environmental input patterns
associated output patterns (i.e., opinions)

THE IMAGITRON
1 0.9
memories + confabulations
optimal memory generation rate highest constraint satisfaction by confabulations
input
Pmem , probability of activating a memory
0.8
0.7
0.6
memories
output
0.5
0.4
i=1
0.3
(w w ) 2
i i
N wi = ith weight N = no. of weights
0.2
0.1
0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
root-mean-square transient synaptic fluctuation

DEVICE FOR THE AUTONOMOUS GENERATION OF USEFUL INFORMATION (DAGUI)
auxiliary sensory inputs
Imagitron quickly absorbs Zen of conceptual space, without human involvement. Perceptron quickly learns to form opinions about novel patterns emerging from imagitron, without human involvement. Perceptron manages weight fluctuations within imagitron until solution pattern is obtained. Can be used to invent significance to raw sensory input patterns (i.e., sense making). Compound DAGUIs emulate juxtapositional invention, logical deduction/induction, and create theories from analogy modules.
IMAGITRON weight fluctuations
potential ideas
PERCEPTRON
opinions
US Patent 5,659,666, DAGUI (Creativity Machine Paradigm)

SELF-TRAINING ARTIFICIAL NEURAL NETWORK OBJECT (STANNO)
data patterns
No explicit, human-conceived training algorithm. Trainer absorbed into trainee so as to create a monolithic neural net capable of autonomously learning. Class wrapper supplied to produce most efficient object-oriented neural net in world. 10 million attribute networks enabled for PCs and GPUs, completing training cycles on millisecond time scales. At core of extremely advanced automotive machine vision systems. Automates DAGUIs by allowing component nets to learn from successes and failures.
TRAINEE
weight updates
error patterns TRAINER weight update strategy
US Patent 5,845,271, Non-Algorithmically Implemented ANNs

DEVICE FOR THE AUTONOMOUS BOOTSTRAPPING OF USEFUL INFORMATION (DABUI)
auxiliary sensory inputs
Untrained STANNO-based imagitron generates potential idea. Untrained STANNO-based perceptron generates some figure of merit to it. If figure of merit (opinion) is sufficient, reinforcement takes place in all networks. If figure of merit (opinion) is insufficient, training is weakened in imagitron, with reinforcement learning occurring in perceptron. Bootstrapping is continued until ideas have matured. Thereafter, derivative ideas may be generated on demand.
weight fluctuations / learning
STANNO-BASED IMAGITRON potential ideas STANNO-BASED PERCEPTRON
opinions
US Patent 7,454,388, DABUI

SUPERNETS
Any number of sensors and actuators may be tied together via an adaptive/creative synthetic brain. Myriad STANNO modules may interconnect into supernets involving self-forming group membership filters (GMF) and Creativity Machines. Various neural correlates to the human brain automatically form. Synthetic intelligence shapes itself in response to environmental and corporeal demands. We are skipping simulation of human brain and striving for trans-human level intelligence.
US Patent 7,454,388, DABUI

CREATIVE ROBOT MASTERS SAND

Main problem is accumulating mound of sand in front of robot. Robot develops its own squat and leap strategy via underlying DABUI system. Side-to-side motion disperses sand accumulation. Behavior developed in less than two minutes, tabula rasa.
CREATIVE ROBOT MASTERS SOIL

Main problem again is accumulation of soil in front of robot. DABUI system develops its own strategy for crawling through potting soil, leveling material with front legs, while thrusting with rear legs. Behavior developed in less than a minute!
CREATIVE ROBOT LEARNS TO DREAD ROCKS

Main problem is nondeterministic behavior of rock beneath robot. DABUI system develops a strategy of stabilization within the rock through inplace rocking, so that legs penetrate as deeply as possible. Garden rock definitely offers a higher crawling impedance than sand or soil.
CONTEMPLATIVE TERRAIN-SENSING ROBOTS

Vision pathways, navigation field generators, creative motor control, and path optimization modules knit themselves together into a single Supernet. Robot contemplates terrain, decides upon path of least resistance, and selects appropriate gaits along way to reach target, avoiding rocky patch.
target high impedance rock patch
CONTEMPLATIVE TERRAIN-SENSING ROBOTS

When rocky patch is replaced by sand, DABUIbased control system picks a more direct route toward the target on left. Project was reborn as autonomous rendezvous and docking for NASA. Later, off-world robots prototyped for NASA, utilizing tabula rasa behavioral development.
target
low impedance sand patch
STIFF CHALLENGES / BOLD SOLUTIONS

Retreat from Complexity Emphasize simplicity and elegance. Disruptive Technology Identify proper point of entry into industry. Academic Adaptation Treat university researchers as esteemed knowledge workers. Timid Incrementalism Seek out the more visionary man in uniform. Poverty of Imagination Stimulate imagination via potential applications and the viability of machine consciousness.
cortical imagitron
thalamic perceptron
STIFF CHALLENGES / BOLD SOLUTIONS

let a thousand flowers bloom Allowed IEI in door. There is vast duplication of effort. Grand Challenge / X-Prizes Small companies with limited budgets and vanguard technologies cant afford to drop everything. Too many simplifications create overabundant competition. Become engineering and not scientific competitions. Greatest Curse This is an unanticipated technology that does everything. We are drowning in the possibilities. There is hope in collaboration.
grand challenges too simplified.
IEI DECADE (1997-2007) IN A NUTSHELL

Innovation Through Artificial Intelligence

Uploaded by

Copyright:

Available Formats

Innovation Through Artificial Intelligence

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Innovation Through Artificial Intelligence

Uploaded by

Copyright:

Available Formats

Creativity Machine Paradigm

hopping synaptic perturbations

Simple & Elegant

Boldly Innovating Innovation via Creative Machine Intelligence

Self-organization inherent to artificial neural networks makes this so.

Neural brainstorming session is the basis of human cognition.

Is that all there is?

The Ultimate Idea

Requires utmost boldness to advance, realizing its philosophical implications.

2010, Imagination Engines, Inc.

2010, Imagination Engines, Inc.

THE PERCEPTRON raw environmental input patterns

associated output patterns (i.e., opinions)

Pmem , probability of activating a memory

N wi = ith weight N = no. of weights

0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

root-mean-square transient synaptic fluctuation

DEVICE FOR THE AUTONOMOUS GENERATION OF USEFUL INFORMATION (DAGUI)

auxiliary sensory inputs

IMAGITRON weight fluctuations

US Patent 5,659,666, DAGUI (Creativity Machine Paradigm)

SELF-TRAINING ARTIFICIAL NEURAL NETWORK OBJECT (STANNO)

error patterns TRAINER weight update strategy

US Patent 5,845,271, Non-Algorithmically Implemented ANNs

DEVICE FOR THE AUTONOMOUS BOOTSTRAPPING OF USEFUL INFORMATION (DABUI)

auxiliary sensory inputs

weight fluctuations / learning

STANNO-BASED IMAGITRON potential ideas STANNO-BASED PERCEPTRON

US Patent 7,454,388, DABUI

US Patent 7,454,388, DABUI

CREATIVE ROBOT MASTERS SAND

2010, Imagination Engines, Inc.

CREATIVE ROBOT MASTERS SOIL

2010, Imagination Engines, Inc.

CREATIVE ROBOT LEARNS TO DREAD ROCKS

2010, Imagination Engines, Inc.

CONTEMPLATIVE TERRAIN-SENSING ROBOTS

target high impedance rock patch

2010, Imagination Engines, Inc.

CONTEMPLATIVE TERRAIN-SENSING ROBOTS

low impedance sand patch

2010, Imagination Engines, Inc.

STIFF CHALLENGES / BOLD SOLUTIONS

2010, Imagination Engines, Inc.

STIFF CHALLENGES / BOLD SOLUTIONS

grand challenges too simplified.

2010, Imagination Engines, Inc.

IEI DECADE (1997-2007) IN A NUTSHELL

2010, Imagination Engines, Inc.

You might also like