What Is Artificial General Intelligence?

What Is Artificial General Intelligence?Clarifying The Goal For Engineering & Evaluation

Mark R. Waser

Wang’s 5 Definitions of AI

• Structure – neurons working in parallel (based on brain architecture)

• Behavior – acts like a human being (based on human psychology/Turing test)

• Capability – has the ability to solve problems (narrow AI)

• Function – has cognitive functions similar to that of humans (searching, reasoning, planning, etc.)

• Principle – operates according to a simple fundamental rational or optimal principle

Wang’s 5 Definitions (revised)

Architectures• Structure – brain architecture• Function – architecture of the mind• Principle – single rational problem solving

theory/architecture

Emergent Properties• Capability – what it can do• Behavior – what it actually does do

Principle

Structure

FunctionCapability Behavior

SoarLIDA

Novamente

AIXI

NARSchatbots

Narrow AI

ACT-RSAL

Neural NetworksHawkins/Blue Brain

Cyc CoSy

What do WE WANT?

What it IS (architecture)

What it CAN do (capability)

What it DOES do (behavior)

CAPABILITY and BEHAVIOR

What is OUR GOAL in CREATING AGI?

Intelligence = problem solving & goal achieving

Solve all of humanity’s problems

ORIs humanity

one of the problems to be solved?

FRIEND

ENEMY

ENTITYTOOL

SLAVE?

What is the difference betweenan intelligent tool and an entity?

To CREATE an ENTITY

with the ability and desire to

cooperateto solve problems, achieve goalsand improve life for everyone

What is OUR GOAL in CREATING AGI?

Lang

uage

/Rep

rese

ntat

ionC

ooperation/Ethics

Planning/Problem-Solving

Cho

msk

y/Pi

nkerR

awls/H

auser

Intelligence

How do we get there?

AutogenyAKA “Seed AI”

(Oblinger 2008)

OR

Wishful Thinking/Cargo Cult Engineering

Rationally Anticipated Emergent Properties

STAY AS CLOSE TO

EXISTING EXAMPLES

AS POSSIBLE

EncodePerceptual Learning

Procedural Learning

Cognitive Cycle

(Franklin 2007)

Sloman’s architecture

for ahuman-like agent

(Sloman 1999)

Baar’s Global Workspace Theory

• Most of cognition is implemented by a multitude of small, local, special purpose processes, that are almost always unconscious

• Coalitions of these processes compete, whenever necessary, for conscious attention (access to a limited capacity global workspace)

• Attention then serves as an INTEGRATION POINT that allows us to deal with novel or challenging situations that cannot be dealt with efficiently, or at all by local, routine, unconscious processes

(Also Perlis 2008)

Civilization advances by extending the number of important operations

which we can perform without thinking about them.

-- Alfred North Whitehead

ETHICALAUTOGENOUS

ATTENTIONAL

SUMMARY

EA3GI – The fastest, safest road to artificial general intelligence