Consciousness Sli

7/28/2019 Consciousness Sli

1/34

WHAT CONSCIOUSNESS DOES A ROBOT N

John McCarthy

Computer Science Department

Stanford [email protected]

http://www-formal.stanford.edu/jmc/

July 12, 2002

Almost all of my papers are on the web page. Thi

is http://www-formal.stanford.edu/consciousness


2/34

APPROACHES TO ARTIFICIAL INTELLIGE

biologicalHumans are intelligent; imitate huma

observe and imitate at either the psychological or

physiological level

engineeringThe world presents problems to inte

Study information and action available in the wor

1. Write programs using non-logical representatio2. represent facts about the world in logic and

what to do by logical inference

We aim at human level AI, and the key phenom

the common sense informatic situation.


3/34


4/34

A LOGICAL ROAD TO HUMAN LEVEL A

Use Drosophilas that illustrate aspects of repr

tion and reasoning problems.

Concepts, context, circumscription, counterfa

consciousness, creativity, approximation

narrative, projection, planning

mental situation calculus

domain dependent control of reasoning


5/34

Logic in AI

Features of the logic approach to AI.

Represent information by sentences in a logical lan

e.g. first order logic, second order logic, modal lo

theory in logic.

Auxiliary information in tables, programs, states,

described by logical sentences.

Inference is logical inferencededuction suppleme

some form of nonmonotonic inference, e.g. circution.


6/34

Action takes place when the system infers that itdo the action.

Observation of the environment results in sentememory.

Situation calculus formalizes the relationsholds(p, s), occurs(e, s) and the function result(e, shas a new situation as its value.

Formalizing consciousness involves giving situationtal components.

Self-observation results in sentences about the s

state of mind.


7/34

What Introspection do Robots Need?

Whats this?: What ability to observe its own

tational state and computational processes does

need to do its tasks?

General Knowledge?: What general informationreasoning processes does it need to plan its ment

Design approach: Asking what consciousness is

gives different answers from those trying to defi

sciouness has given.


8/34

Recommendation for AI: Introspection is needed to

whether to think or look, to learn from near mis

use counterfactuals and keep pedigrees of beliefs.

Recommendation for psychologists and philosophe

this direct design stance approach to your metho


9/34

What is Consciousness? We consider several kin

knowledge.

There are many unconscious stimulus-respo

relations in animal and humans, and there can machines.

Unconscious knowledge can affect behavio

Conscious knowledge and other conscious infor

can be observed by the actor.

Self-conscious knowledge is conscious knowle

about conscious information.

Some aspects of behavior require decisions o

whole system. Which way to run is an example. decisions are made by a central mechanism


10/34

In logical robots, the consciousness is be a sub-

of memory containing facts and other mental en

Reasoning involves the entities in consciousne

leads to decisions when the reasoning leads t

statement that an action should be performe

The capacity of consciousness is limited, so

information displaces old, which may go to a histo


11/34

Taxonomy of Consciousness

The consciousness itself can be observed and thevations enter consciousness.

Robot consciousness can be given powers peopl

have.

complete memory of the past

larger immediate memory

avoiding wishful thinking

ability to self-simulate


12/34

greater ability than humans at organizing expe

Most required features of robot consciousnesscorrespond to features of human consciousne


13/34

FEATURES OF FORMALIZED CONTEXT

Ist(c, p), V alue(c, exp)

c : p

C(SherlockHolmes) : Detective(Holmes)

entering and leaving contexts

introspection by transcending outermost cont

Assuming(c, p)


14/34

C(I , N o w)


15/34


16/34

What Tasks Require Self-Consciousness?

Tasks NOT requiring consciousness

Reacting directly to the environment.

Learning direct reactions to the environment.

Tasks requiring consciousness

Anticipating the future.

Analyzing the past. Self-criticism.


17/34

Speech requires introspection. Would this phrase

this object if I were in his place?

Mechanisms of consciousness operate

unconsciously.


18/34

More Tasks Requiring Consciousness

Observe physical body.

. . . : c(Here, N ow, I) : hungry in(pen, hand)

Do I know that proposition?

c(Now,I) : know(sitting(Clinton))

Do I know what thing is? What is it?

c(N o w , I , ) : know-what

c(Now,I) : is(,jdoe)


19/34

c(S-Symp, I) : is(,jdoe)

Did I ever do action? When and precisely what?

What are my goals?

What is currently happening?

What is the state of the actions I am currently pe

ing?


20/34

What are my intentions?

c(Now,I) : intend()

What does my belief in p depend on?

What are my choices for action?

c(Now,I) : can(lecture) can(walk-out)

Can I achieve possible-goal?

Does my mental state up to now have property p

How can I plan my thinking on this problem?


21/34

Yet more Introspection

Since I do not intend to call him again, Ill forget h

phone numberor put it in low priority storage. Pa

a proposition with a reason.

I know how to do A and dont know how to do B

Renting a cellular telephone is a new idea for me.

I tried that, and it didnt work. This isnt just bac

ing.

What would I do if I were she?


22/34

Understanding

The meaning of understanding is context dep

To understand something is to have the fac

reasoning methods about it that are relevant

context.

People who understand cars know about crank

Fish do not understand swimming, e.g. the

ponder how to swim better.


23/34

Comenicis coach understood womens gymna

but not from having done it.

Understanding is an approximate concept.


24/34

Inferring Non-knowledge

Inferring non-knowledge requires special logic

treatment.

According to Godels theorem, the consistency of a

system cannot be a theorem of the system.

Inferring that any proposition is unknown implies t

tem is consistent, because if the system is inconsis

sentences are theorems.


25/34

Godels notion of relative consistency permits pr

non-knowledge. Assume that the theory is con

and express this as a second order formula assert

existence of functions and predicates with the pos

properties. To show non-knowledge of a propo

prove that if predicates and functions exist satisfy

original theory, show that they still exist when th

added to the theory.

Second order logic is the natural toolrememberi

the proof of consistency must be accomplished

robots normal reasoning apparatus.


26/34

Not knowing Clinton is sitting

Theory with predicates including sits

A(P1, . . . , P n, sits)

(P1, . . . , P n sits

)A(P1, . . . , P n, sits

)

expresses consistency of the theory, and

(P1, . . . , P n sits

)(A(P1, . . . , P n, sits

) sits(Clinton, s))

expresses the consistency of the theory with the

assertion that Clinton is not sitting in the situat


27/34

Then

(8) (9)

asserts relative consistency.

(P2P3)A(P1, P

2, sits

) sits(Clinton, s).

asserts it with P1 fixed. If sits doesnt appear elsethe simplest case, we get by with

sits = (x ss)((x = Clinton ss = s) sits(x,


28/34

Ad hoc context c(prob) for a problem prob

The c(prob) consists mainly of a theory includin

deemed relevant to prob.

c(prob) is initially empty.

c(prob is referred to from the context c0 in wh

problem is posed by lifting relations

If c(problem) is small enough, whether the pro

solvable in the context is definite and decidea


29/34

Second order logic instead of model theory ke

cisions about.whether there is enough informa

solve the problem within the logical language.


30/34

Relevant Work Some non-real time work is relevarobot examining its mental processes in real ti

Rationalize skillBratko, Michie, Muggleton et. a

Sternberg.

Inductive learning systematizes and generalizes fa

predicate logic.Muggleton


31/34

Chemistry and Logic

The interaction of chemistry and logic in humans

is something we dont need in robots.

Here are some aspects of it.

When a paranoid takes his medicine, he no longer b

the CIA is following him and influencing his thoug

radio waves. When he stops taking the medicine

lieves it again.

Both the medicine and the substance to which

antagonist are too simple to encode beliefs abo

CIA.


32/34

Hormones analogous to neurostransmitters open s

gates to admit whole classes of beliefs into conscio

They are analogs of similar substances and gates

mals.

It would seem that such mechanisms wont be u

robots.


33/34

Philosophical and Psychological Remarks

The strong design stance has advantages for philo

Gives adequacy criteria. Will the mechanism w

Forces a greater concreteness than is custom

Shows weaknesses of a priori reasoning.

Relative consistency evades mathematical diffic


34/34

Consciousness Sli

Documents

Transcript of Consciousness Sli