+ All Categories
Home > Documents > 1 Why BICA is Necessary for AGI Alexei Samsonovich (George Mason University) Biologically Inspired...

1 Why BICA is Necessary for AGI Alexei Samsonovich (George Mason University) Biologically Inspired...

Date post: 17-Dec-2015
Category:
Upload: randell-bailey
View: 215 times
Download: 0 times
Share this document with a friend
Popular Tags:
30
1 Why BICA is Necessary for AGI Alexei Samsonovich (George Mason University) Biologically Inspired Cognitive Architec
Transcript

1

Why BICA is Necessary for AGI

Alexei Samsonovich (George Mason University)

Biologically Inspired Cognitive Architecture

2

Because we need a human-like universal learner

One that describes human cognition and learning at a higher symbolic level

“Critical mass” includes human-like mental states that can act on each other

Questions Answers

Why BICA is necessary for achieving AGI?

What kind of a BICA?

What are the minimal starting requirements, i.e., the “critical mass”?

3

A few words about GMU-BICA

4

Mental states in GMU-BICA

A mental state in GMU-BICA includes:

Contents of awareness represented by schemas

A token representing an instance of the Self who is aware (labeled I-Now, I-Next, etc.)

Working memory: Active mental states of the Self

I-Imagine:

•Intermediate goal situation

I-Imagine:

•Intermediate goal situation

I-Goal:

•Stimulus satisfaction

I-Goal:

•Stimulus satisfaction

I-Next:

•Scheduled action

•Expectation

I-Next:

•Scheduled action

•Expectation

I-Previous:

•Ideas

•Visual input

I-Previous:

•Ideas

•Visual input

I-Meta:

•Scenario

•Analysis

I-Meta:

•Scenario

•Analysis

I-Past:

•Past experience

•Prospective memories

I-Past:

•Past experience

•Prospective memories

I-Now:

•Ideas

•Intent

I-Now:

•Ideas

•Intent

Episodic memory: Frozen mental states of the Self

I-Past-1I-Past-1

I-Past-2I-Past-2

I-Past-3I-Past-3

I-Past-4I-Past-4

5

Mental state dynamics in working memory of GMU-BICA: an example

meI-Now

Working memoryInput-output

Semantic memoryS

PQ

S

S

he

me'

he'

me

I-NextS

he

me

He-NowS

me'

me

He-Next

S R

6

Examples of types of mental states in GMU-BICA (a possible snapshot of working memory)

I-NowI-PreviousI-Next

I-Imagined-1

I-GoalI-Imagined-2

I-Meta-1

I-Past

I-Detail-2I-Feel

He-NowShe-Past He-Now-I-Now

I-Subgoal

I-False-Belief

I-Next-Next

I-Past-Revised

I-Meta-2

I-Detail-1

I-Imagined-3

She-Past-Prev

I-Alt-Goal

7

Models that we need to integrate

8

Self-regulated learning (SRL) model of problem solving

(based on Zimmerman & Kitsantas, 2006)

“…there is a need to build a unified model of meta-cognition and self-regulated learning that incorporates key aspects of existing models, assumptions, processes, mechanisms, and phases”

(Azevedo and Witherspoon, AAAI BICA-2008)

Object

Level

Meta-

Level

Ground

Level

Doing Reasoning Metareasoning

ActionSelection Control

Perception Monitoring

Model of meta-cognition(Cox & Raja, 2007)

9

Result: A Mental-state model of SRL

Forethought

Task analysis Self-beliefs

Performance

Self-control Self-observation

Reflection

Self-judgmentSelf-reactions

Task analysis

Identify goalSelect strategic steps (a plan)

Self-beliefs

Self-efficacyGoal-orientationIntrinsic interest

Self-observation

Self-recording using a worksheet

Self-control

Enact selected steps to solve the problem

Self-evaluation

Compare result to the standard (a template)

Homework task

Problem: ax+b = cGoal: Solve for x, i.e., have a formula x=…

Select strategic steps

Isolate x - use subtraction property - use division property

Enact strategic steps

ax+b = c | -bax = c-b | /ax = (c-b)/a

Result validation

x=(c-b)/a compare tox = …(no x in r.h.s.)There is a match.

Self-reaction

Met standardSkill masteredSelf-reward(Exit) -- OR –Did not meet standard Attribute failure to ineffective strategy selection(Loop reentry)

I-Now

I-Meta

I-Detail-

1

I-NextI-Next-Next

I-Detail-2

I-Meta-Next

I-Goal

HW Problem:Solve for x: ax+b=c

(Samsonovich, De Jong & Kitsantas, to appear in International Journal of Machine Consciousness, 1, June 2009)

10

Final take-home message

11

- How to build a universal learner?- Need to bootstrap from “critical mass” ( )- How to build a “critical mass” (suppose we know what)?

2. Brittle rapid prototype-demo

1. Incremental bottom-up engineering

3. SRL assistant (finessing lower levels by students!)

Thank you.

Without a good stimulus will take forever

Useless toy(BICA Phase I)

Feasible and practically useful stepping stone

There are at least three approaches to building a “critical mass”:

Watch for AAAI 2009 Fall Symposia (BICA, SRL-metacog)

12

End of Talk 1 / Beginning of Talk 2

13

A Cognitive Map of Natural Language

Alexei Samsonovich (George Mason University)

14

Theory

15

Introducing two notions of a semantic cognitive map (SCM): “Strong” SCM with a

dissimilarity metric

A is closer to B than C A is more similar to B than C

“Weak” SCM that captures both synonym and antonym relationsA CB

A and B are synonyms, A and C are antonyms. Don’t care about unrelated.

AB C

16

Background: Method of building an SCM

1. Represent symbols (words, documents, etc.) as vectors in Rn

2. Optimize vector coordinates to minimize H

3. Do truncated SVD of the resultant distribution

cSS

QQAS

QAS

QSA

yxyxHd

xxyxyxHc

xxyyxHb

xxyxyHa

22

2422

42

4

exp)(

)(

)(

)(

dot product

x, y Q – vectors in Rn

A – antonym pairsS – synonym pairs

(Samsonovich & Ascoli, Proceedings of AGI-2007)

17

Example: color map Sample N = 10,000 points on

a sphere (A) declare some pairs of points

‘synonyms’ (some of those that are close to each other)

declare some other pairs of points ‘antonyms’ (some of those that are separated far apart)

assign random coordinates to points in 10-dimensional space (B)

apply an optimization procedure to the set of 10,000 random vectors in order to minimize the following energy function:

The result is the reconstructed spatial distribution of colors (C)

A

B

C

xSxyAxy

xxyxyH 4

18

Geometric properties of the reconstructed color map are robust with respect to variation of model parameters

19

Results for Synonym-Antonym Dictionaries

20

Optimization results

21

Sorted words and antonym pairs (MS Word English)

22

Geometric characteristics of the SCM (MS Word English)

23

Semantic characteristics of the SCM

synonyms

antonyms

Synonym pairs and antonym pairs, if mixed together, can be separated with 99% accuracy based on the angle between vectors:

acute synonyms, obtuse antonyms

Semantics of the first 3 dimensions are more general than any words, yet clearly identifiable:

PC#1: success, positive, clear, makes good sense

PC#2: exciting, does not go easy

PC#3: beginning, source, origin, release, liberation, exposure

*

24

PC-by-PC correlation across languages and datasets

25

Clustering of words in the first SCM dimension: WordNet and ANEW vs. MS Word

26

Applications

27

Examples of “semantic twisting”

28

Sentiment analysis: 7 utterances automatically allocated on SCM

1. Please, chill out and be quiet. I am bored and want you to relax. Sit back and listen to me.

2. Excuse me, sorry, but I cannot follow you and am falling asleep. Can we pause? I've got tired and need a break.

3. I hate you, stupid idiot! You irritate me! Get disappeared, or I will hit you!

4. What you are telling me is terrible. I am very upset and curious: what's next?

5. Wow, this is really exciting! You are very smart and brilliant, aren't you?

6. I like very much every word that you say. Please, please, continue. I feel like I am falling in love with you.

7. We have finally found the solution. It looks easy after we found it. I feel completely satisfied and free to go home.

(Samsonovich & Ascoli, in Proc. of AAAI 2008 Workshop on Preference Handling)

29

Acquired 40+ reviews for each of three movies: Iron Man, Superhero and Prom Night, from the site www.mrqe.com

For each review, computed the average map coordinate of all identified indexed words and phrases.

RESULT: Statistics for PC#1 are consistent with grades given to the movies in the reviews.

Iron Man: (1.95, 0.52), Superhero: (1.49, 0.36), Prom Night: (1.17, 0.42)All differences are significant except PC#2 of Superhero vs. Prom Night

Sentiment analysis:Mapping movie reviews as ‘bags of words’

30

Weak SCM is low-dimensional, yet distinguishes almost all synonym-antonym pairs

SCM dimensions have clearly identifiable semantics that make sense virtually in all domains of knowledge

The map semantics and geometrical characteristics are consistent across corpora and across languages

Therefore, SCM can be used as a metric system for semantics (at least for the most general part of semantics)

SCM can be used to guide the process of thinking in symbolic cognitive architectures

Other potential applications include sentiment analysis, semantic twisting, document search, validation of translation

CONCLUSIONS

Thank you.Credits to Giorgio A. Ascoli, Rebecca F. Goldin, Thomas T. Sheehan


Recommended