Modeling Human Intelligence as a Slow Intelligence System

Modeling Human Intelligence as a Slow Intelligence System

Tiansi Dong

Department of Computer ScienceUniversity of Hagen

Germany

Outline

Slow Intelligence System (SIS) Properties of Human Intelligence The question Case study in Spatial Reasoning

– Within one snapshot view

– Between snapshot views Conclusion Outlooks

Slow Intelligence System

Solve problems by tryingContext-awareMay not perform well in a short runLearn to improve its performance

Slow Intelligence System

Problem enumerator adaptor concentratoreliminator Solution

propagator

timing controller

environment

environment

Human Intelligence is_a Slow Intelligence

Slow developmental

infants

pupils

students

doctors

professors

Properties of Human Intelligence

Babies cannot see constant objects

Properties of Human Intelligence

Now suppose it not about apple, rather

football moneybus

Spatial cognition is foundamental

Question Human intelligence is_a Slow Intelligence

System Spatial intelligence is foundamental to human

intelligence Slow Intelligence System has_a architechture Is it possible that spatial intelligence be

simulated within the SIS architechture?

A picture on a wall A lady in the picture The lady is back to

us A gentalman is near

the picture The gentalman is at

the left side of the picture

SIS for Spatial Knowledge within a Scene

Object categories

– A picture

– A lady

– A wall Spatial relations

– On, in

– Back, left

– Near

SIS for Spatial Knowledge within a Scene

Object categories

– A picture

– A lady

– A wall

Specific Question

Cross linguistic spatial relations

– in, on, near, front, left,...

– 上，左，前

?

Results in Psychology

Connection relation is primitive Orientation and distance relations are acquired

– Piaget (1954) The Construction of Reality in the Child. Routledge & Kegan Paul Ltd.

– Carey (2009) The Origin of Concepts. Oxford Press

Some existing work Neural Network

– Terry Regier (1996) “The Human Semantic Potential”, MIT Press.

• Spatial model is point-based

• 'connection' is not primitive Formal logic

– De Laguna (1922) Point, line and surface as sets of solids, The Journal of Philosophy

– T Dong (2008) Comment on RCC–From RCC to RCC++, Journal of Philosophical Logic

• Spatial model is region-based

• 'connection' is primitive

Object categories

Case study in Spatial Reasoning in SIS

+ Connection relation

near, in, on, left, right, ...

Context-aware

Problem-solving by trying

Spatial Reasoning for 'one foot away'

B

A

Trying all possible extension (problem solving by trying), and see whether one foot connects with the target object (context awareness).

∃foot [foot ∈ FOOT ⋀ C(A, foot) ⋀ C(foot, B)]

Spatial Reasoning for distance in SIS

In the UK “A is one foot away from B” means region B can be reached by a region of the same size as the British imperial foot from A.

China and Egypt Cun: the body segment between the wrist striation behind the thumb and

the pulsing point of the radial artery; Cubit: the segment between the bent elbow and the point of extended

middle finger. In modern physics: meter, light-year.

The meter is the distance traveled by light in vacuum during a time interval of 1/299 792 458 of a second

A B

X Y

Spatial Reasoning for distance comparison

“A is nearer to B than to C”: there is an X such that C(A, X) ⋀ C(X, B) and there is no X such that C(A, X) ⋀ C(X, B).

B

C

A

XX

Trying all possible extension (problem solving by trying), and see whether one x connects with B and non of x connects with C (context awareness).

x

Spatial Reasoning for orientation

“A is in front of B”: A is nearer to the front part of B than to its other parts.

BB

A

Spatial Reasoning for orientation

Orientation is determined by the shape of the reference object, and the method of distance comparison.

ReferenceObject

NW

SW

W

E

N NE

SES

Spatial Reasoning Performance

Performance in term of the accuracy increases, as the number of sides of the reference object increases.

Qualitative spatial orientation frameworks, e.g. Frank (1992), Freksa (1992), Hernández (1994), Freksa (1999)Renz and Mitra (2004), Dong and Guesgen (2007)

Quantitative spatial orientation frameworks,Euclidean geometry

Spatial Reasoning Performance

P: ae-iθ

Q: e-iθ

θ

The orientation of P can be defined as the point on the unit circle which is nearest to P.

O 1

W

SIS for Spatial Reasoning for one scene: Short Summary

Context-aware (Object, Connection)

Always trying (do spatial extension)

Continuously improve performance(do adaptation)

SIS for Spatial Reasoning between snapshots:

Object tracing

SIS for Spatial Reasoning between snapshots:

Object tracing

SIS for Spatial Reasoning between snapshots:

Object tracing

SIS for Spatial Reasoning between scene:Object tracing

Fast changining leads to an illusion

bird → rabbit, rabbit → bird

Otherwise, bird flies, rabbit moves

SIS for Spatial Reasoning between scene:Object tracing

A problem of object mapping between scenes Two object tracing results due to two different

priorities

– Priority on spatial changes (minimal spatial changes)

– Priority on object categories (objects are mapped within same categories)

SIS1 for Object tracing with priority on spatial changes

[permutation] list all possible mappings

[elimination+concentration] choose the mapping

with the minimal spatial changes

SIS2 for Object tracing with priority on object category

[permutation] list all possible mappings [elimination] remove mappings of different object

categories

[elimination+concentration] choose the mapping within minimal spatial changes

Why fast changing leads to illustiion?

Conjection: SIS2 takes more time than SIS1 in object mapping between scenes.

Conclusion

SIS shall be a Cognitive Architecture

– SIS for spatial knowldge acquisition within a scene

– SIS for spatial knowledge acquisition between scenes

– SIS for Spatial Cognition

– Spatial Cognition is foundamental to Human Intelligence

– SIS as a Cognitive Architecture for Human Intelligence

Outlooks

Relations between SIS and other Cognitive Architectures, e.g. ACT-R, CLARION, ...

Any difference to acquisit implicit knowledge and explicit knowledge