Eliza - der Computer als Gesprächspartner
Autor: Andre Adrian
Version: 22.Apr.2007
Der Eliza-Effekt
Das Programm Eliza wurde 1966 von Joseph Weizenbaum geschrieben. Eliza
analysiert die Texteingabe des Gesprächspartners nach einfachen
Regeln und produziert aus Teilen der Eingabe und eigenen Satzbausteinen
eine Antwort. Berühmt wurde Eliza mit einem Regelsatz welcher die
Gesprächpsychotherapie von Carl Rogers nachbildet.
Joseph Weizenbaum hat mit Eliza ein Künstliches Intelligenz
Programm geschaffen welches nach 40 Jahren immer noch beachtet wird.
Der Erfolg von Eliza war dem Autor wohl etwas unheimlich. Die Eliza
Manie zeigt doch nur, daß ein Mensch alles vermenschlicht. Blitz
und Donner wurden als Götter vermenschlicht. In Fabeln,
Märchen und Romanen wie z.B. "Watership Down" treten sprechende,
vermenschlichte Tiere auf.
Es scheint dem Menschen unmöglich zu sein den menschlichen
Blickwinkel abzulegen. Alles nicht-menschliche in der Umwelt wird als
mehr oder minder seltsamer Mitmensch angesehen. Vielleicht ist dies die
Funktionsweise
der Spiegelneuronen: Wir können uns in ein rollendes Rad welches
kurz vor dem Umkippen ist hineinversetzen. Wir können sinnvoll die
Zukunft "gleich kippt es um" vorhersehen. Und weil wir das nötige
"physikalische Gefühl" für das Rad haben, funktioniert die
menschliche Empathie (Einfühlungsvermögen). Die Gedankenkette
wird mit der Schlußfolgerung: "Wenn ich mich hineinversetzen
kann, dann muß es ein Mitmensch sein" abgeschlossen.
Alle Tamagotchis
dieser Welt und wohl ein Großteil der Dackel funktionieren
für ihre Besitzer nach diesem Muster: Das Verhalten
ist verständlich weil vorhersehbar, der Besitzer kann sich
hineinversetzen. Die Spiegelneuronen funktionieren wie wenn ein
Mitmensch die beobachtete Handlung ausführt. Und weil die
Spiegelneuronen arbeiten muß das beobachtete Objekt ein Mitmensch
sein. Vielleicht aus Plastik, mit drei Knöpfen und einem LCD
Bildschirm, aber trotzdem.
Eigentlich ist der Eliza-Effekt nichts Neues und vor allem nichts
Unheimliches.
Leider gab es im Bereich Künstliche Intelligenz die
Verwechslung zwischen "der Programm Autor ist intelligent" und "das
Programm ist intelligent". Das Eliza Programm zeigt keine Intelligenz
im Sinne von "lernfähiges Programm".
Klientenzentrierte Psychotherapie
Die Therapie von Carl Rogers ist heute fest etabliert. Einmal in der
Besserung oder Heilung von psychischen und psychosomatischen
Störungen. Viel wichtiger ist ihre Bedeutung für die
Kommunikation zwischen "Gesunden". Niccolò Machiavelli sagt "der
Mensch ist schlecht". Carl Rogers sagt "der Mensch ist gut". Beide
Autoren können sich auf empirische Beweise stützen.
Vielleicht zeigen Machiavelli und Rogers die beiden Extreme des
menschlichen Verhaltens.
Die klientenzentrierte Psychotherapie macht den Therapeuten zum
Geburtshelfer der Selbsthilfe. Eigentlich weiss der Klient ja selbst,
was am besten für ihn ist. Nur kommt dieses Wissen ohne Therapeut
nicht heraus. Für die Geburtshelfer Funktion des Therapeuten sind
drei Verhaltenselemente nötig:
- Bedingungslose positive Wertschätzung
- Einfühlungsvermögen (Empathie)
- Wahrhaftigkeit gegenüber den Klienten (Kongruenz)
Das Eliza Programm hat diese Therapeuten Anforderungen erfüllt,
oder besser gesagt: Eliza hat die Gesprächspartner nicht
enttäuscht. Wenn im Regelsatz von Eliza nur Aussagen mit positiver
Wertschätzung sind kann Eliza nur eine positive Wertschätzung
zeigen. Wenn die Sätze des Klienten als Fragen gespiegelt werden,
wird eine "Seelenverwandtschaft" zwischen Klienten und Programm
suggeriert. Sind die Antwortphrasen auf Schlüsselwörter des
Klienten gut gewählt, entsteht sogar der Eindruck von
"Eigeninitiative" des Programmes. Werden die Antwortphrasen passend zu
den Schlüsselwörtern gewählt, "lügt" Eliza nicht.
Joseph Weizenbaum schreibt hierzu: "It is important to note that this
assumption is one made by the speaker.
Whether it is realistic or not is an altogether separate question.
In any case, it has a crucial psychological utility in that
it serves the speaker to maintain his sense of being heard and
understood.
The speaker furher defends his impression
(which even in real life may be illusory)
by attributing to his conversational partner all sorts of background
knowledge,
insights and reasoning ability.
But again,
these are the
speaker's
contribution to the conversation."
Gespräch zwischen zwei KIs
Neben Eliza sind noch weitere ähnliche Künstliche Intelligenz
Programm geschrieben worden. RFC439
enthält das Gespräch zwischen Eliza und Parry vom 18
September 1972. Dabei ist Eliza der Doktor und Parry der Patient. Das
Programm Parry simuliert einen paranoiden Schizophrenen. Der Autor von
Parry ist Kenneth Colby. Ein Auszug (Parry beginnt):
*Do you know anything about bookies?
What makes you think I know anything about bookies?
*A bookie didn't pay me off once.
What does that suggest to you?
*I beat up the bookie who did not pay off, and I thought he might use
his friends in the underworld to get even with me.
Die Eliza Regeln
Joseph Weizenbaum hat die Eliza Regeln in dem Papier "ELIZA--A
Computer Program For the Study of Natural Language Communication
Between Man and Machine" beschrieben. Schlüsselwort und Satz
Transformation sind die beiden Kernelemente von Eliza
Eliza Schlüsselwörter
In der Eingabe des Gesprächspartners wird nach
Schlüsselworten gesucht. Die Schlüsselwörter in dem
kleine Eliza Program therapist von
Joe Strout in Python sind:
- Ich Satzanfänge
I need .., I think .., I feel .., I have .., I would .., I want .., I
can't .., I am .., I'm .., I don't .., Can I .., My ..
- Du Satzanfänge
Why don't you .., Are you .., Can you .., You are .., You're .., You ..
- Es Satzanfänge
Is it .., It is .., Is there ..
- Fragewort Satzanfänge
Why can't I .., Why .., What .., How ..
- Sonstige Satzanfänge
Because .., Hello .., Yes
- Hauptwörter
.. mother .., .. father .., .. child .., .. friend .., .. computer ..
- Sonstige Schlüsselwörter
.. sorry ..
Eliza Transformationen
Eine Eliza Sitzung
>>> import eliza
>>>
eliza.command_interface()
Therapist
---------
Talk to the program by typing in
plain English, using normal upper-
and lower-case letters and
punctuation. Enter "quit" when done.
========================================================================
Hello. How are you feeling
today?
>I am happy.
Did you come to me because you
are happy?
>Yes.
OK, but can you elaborate a bit?
>I think I like to share my
happyness.
Do you really think so?
>Why not?
Why don't you tell me the reason
why not?
>Because positive thinking
wins the world.
If positive thinking wins the
world, what else must be true?
>That all men are brothers.
How does that make you feel?
>Happy.
Please tell me more.
>You should try happyness
yourself.
Why do you say that about me?
>Because you are so negative.
Does that reason apply to
anything else?
>Computers are negative, too.
Why do you say that computers am
negative, too?
>Because computers can not
talk proper english.
What other reasons come to mind?
>That this chat is long enough.
Can you elaborate on that?
>quit
Thank you for talking with me.
>>>
Der Quelltext
Der Quelltext ist kompatibel zu Python
2.5.1.
#----------------------------------------------------------------------
# eliza.py
#
# a cheezy little Eliza
knock-off by Joe Strout <joe@strout.net>
# with some updates by Jeff
Epler <jepler@inetnebr.com>
# hacked into a module and
updated by Jez Higgins <jez@jezuk.co.uk>
# changed whrandom to
random by Andre Adrian
# last revised: 21 April
2007
#----------------------------------------------------------------------
import string
import re
import random
class eliza:
def __init__(self):
self.keys =
map(lambda x:re.compile(x[0], re.IGNORECASE),gPats)
self.values =
map(lambda x:x[1],gPats)
#----------------------------------------------------------------------
# translate: take a
string, replace any words found in dict.keys()
# with the
corresponding dict.values()
#----------------------------------------------------------------------
def
translate(self,str,dict):
words =
string.split(string.lower(str))
keys =
dict.keys();
for i in
range(0,len(words)):
if
words[i] in keys:
words[i] = dict[words[i]]
return
string.join(words)
#----------------------------------------------------------------------
# respond: take a
string, a set of regexps, and a corresponding
# set of
response lists; find a match, and return a randomly
# chosen
response from the corresponding list.
#----------------------------------------------------------------------
def respond(self,str):
# find a match
among keys
for i in
range(0,len(self.keys)):
match = self.keys[i].match(str)
if
match:
# found a match ... stuff with corresponding value
# chosen randomly from among the available options
resp = random.choice(self.values[i])
# we've got a response... stuff in reflected text where indicated
pos = string.find(resp,'%')
while pos > -1:
num = string.atoi(resp[pos+1:pos+2])
resp = resp[:pos] + \
self.translate(match.group(num),gReflections) + \
resp[pos+2:]
pos = string.find(resp,'%')
# fix munged punctuation at the end
if resp[-2:] == '?.': resp = resp[:-2] + '.'
if resp[-2:] == '??': resp = resp[:-2] + '?'
return resp
#----------------------------------------------------------------------
# gReflections, a translation
table used to convert things you say
# into things
the computer says back, e.g. "I am" --> "you are"
#----------------------------------------------------------------------
gReflections = {
"am" : "are",
"was" : "were",
"i" :
"you",
"i'd" : "you would",
"i've" : "you have",
"i'll" : "you will",
"my" : "your",
"are" : "am",
"you've": "I have",
"you'll": "I will",
"your" : "my",
"yours" : "mine",
"you" : "me",
"me" : "you"
}
#----------------------------------------------------------------------
# gPats, the main response
table. Each element of the list is a
# two-element list; the
first is a regexp, and the second is a
# list of possible
responses, with group-macros labelled as
# %1, %2, etc.
#----------------------------------------------------------------------
gPats = [
[r'I need (.*)',
[ "Why do you need
%1?",
"Would it
really help you to get %1?",
"Are you sure
you need %1?"]],
[r'Why don\'?t you
([^\?]*)\??',
[ "Do you really
think I don't %1?",
"Perhaps
eventually I will %1.",
"Do you really
want me to %1?"]],
[r'Why can\'?t I
([^\?]*)\??',
[ "Do you think you
should be able to %1?",
"If you could
%1, what would you do?",
"I don't know
-- why can't you %1?",
"Have you
really tried?"]],
[r'I can\'?t (.*)',
[ "How do you know
you can't %1?",
"Perhaps you
could %1 if you tried.",
"What would it
take for you to %1?"]],
[r'I am (.*)',
[ "Did you come to
me because you are %1?",
"How long have
you been %1?",
"How do you
feel about being %1?"]],
[r'I\'?m (.*)',
[ "How does being %1
make you feel?",
"Do you enjoy
being %1?",
"Why do you
tell me you're %1?",
"Why do you
think you're %1?"]],
[r'Are you ([^\?]*)\??',
[ "Why does it
matter whether I am %1?",
"Would you
prefer it if I were not %1?",
"Perhaps you
believe I am %1.",
"I may be %1
-- what do you think?"]],
[r'What (.*)',
[ "Why do you ask?",
"How would an
answer to that help you?",
"What do you
think?"]],
[r'How (.*)',
[ "How do you
suppose?",
"Perhaps you
can answer your own question.",
"What is it
you're really asking?"]],
[r'Because (.*)',
[ "Is that the real
reason?",
"What other
reasons come to mind?",
"Does that
reason apply to anything else?",
"If %1, what
else must be true?"]],
[r'(.*) sorry (.*)',
[ "There are many
times when no apology is needed.",
"What feelings
do you have when you apologize?"]],
[r'Hello(.*)',
[ "Hello... I'm glad
you could drop by today.",
"Hi there...
how are you today?",
"Hello, how
are you feeling today?"]],
[r'I think (.*)',
[ "Do you doubt %1?",
"Do you really
think so?",
"But you're
not sure %1?"]],
[r'(.*) friend (.*)',
[ "Tell me more
about your friends.",
"When you
think of a friend, what comes to mind?",
"Why don't you
tell me about a childhood friend?"]],
[r'Yes',
[ "You seem quite
sure.",
"OK, but can
you elaborate a bit?"]],
[r'(.*) computer(.*)',
[ "Are you really
talking about me?",
"Does it seem
strange to talk to a computer?",
"How do
computers make you feel?",
"Do you feel
threatened by computers?"]],
[r'Is it (.*)',
[ "Do you think it
is %1?",
"Perhaps it's
%1 -- what do you think?",
"If it were
%1, what would you do?",
"It could well
be that %1."]],
[r'It is (.*)',
[ "You seem very
certain.",
"If I told you
that it probably isn't %1, what would you feel?"]],
[r'Can you ([^\?]*)\??',
[ "What makes you
think I can't %1?",
"If I could
%1, then what?",
"Why do you
ask if I can %1?"]],
[r'Can I ([^\?]*)\??',
[ "Perhaps you don't
want to %1.",
"Do you want
to be able to %1?",
"If you could
%1, would you?"]],
[r'You are (.*)',
[ "Why do you think
I am %1?",
"Does it
please you to think that I'm %1?",
"Perhaps you
would like me to be %1.",
"Perhaps
you're really talking about yourself?"]],
[r'You\'?re (.*)',
[ "Why do you say I
am %1?",
"Why do you
think I am %1?",
"Are we
talking about you, or me?"]],
[r'I don\'?t (.*)',
[ "Don't you really
%1?",
"Why don't you
%1?",
"Do you want
to %1?"]],
[r'I feel (.*)',
[ "Good, tell me
more about these feelings.",
"Do you often
feel %1?",
"When do you
usually feel %1?",
"When you feel
%1, what do you do?"]],
[r'I have (.*)',
[ "Why do you tell
me that you've %1?",
"Have you
really %1?",
"Now that you
have %1, what will you do next?"]],
[r'I would (.*)',
[ "Could you explain
why you would %1?",
"Why would you
%1?",
"Who else
knows that you would %1?"]],
[r'Is there (.*)',
[ "Do you think
there is %1?",
"It's likely
that there is %1.",
"Would you
like there to be %1?"]],
[r'My (.*)',
[ "I see, your %1.",
"Why do you
say that your %1?",
"When your %1,
how do you feel?"]],
[r'You (.*)',
[ "We should be
discussing you, not me.",
"Why do you
say that about me?",
"Why do you
care whether I %1?"]],
[r'Why (.*)',
[ "Why don't you
tell me the reason why %1?",
"Why do you
think %1?" ]],
[r'I want (.*)',
[ "What would it
mean to you if you got %1?",
"Why do you
want %1?",
"What would
you do if you got %1?",
"If you got
%1, then what would you do?"]],
[r'(.*) mother(.*)',
[ "Tell me more
about your mother.",
"What was your
relationship with your mother like?",
"How do you
feel about your mother?",
"How does this
relate to your feelings today?",
"Good family
relations are important."]],
[r'(.*) father(.*)',
[ "Tell me more
about your father.",
"How did your
father make you feel?",
"How do you
feel about your father?",
"Does your
relationship with your father relate to your feelings today?",
"Do you have
trouble showing affection with your family?"]],
[r'(.*) child(.*)',
[ "Did you have
close friends as a child?",
"What is your
favorite childhood memory?",
"Do you
remember any dreams or nightmares from childhood?",
"Did the other
children sometimes tease you?",
"How do you
think your childhood experiences relate to your feelings today?"]],
[r'(.*)\?',
[ "Why do you ask
that?",
"Please
consider whether you can answer your own question.",
"Perhaps the
answer lies within yourself?",
"Why don't you
tell me?"]],
[r'quit',
[ "Thank you for
talking with me.",
"Good-bye.",
"Thank you,
that will be $150. Have a good day!"]],
[r'(.*)',
[ "Please tell me
more.",
"Let's change
focus a bit... Tell me about your family.",
"Can you
elaborate on that?",
"Why do you
say that %1?",
"I see.",
"Very
interesting.",
"%1.",
"I see.
And what does that tell you?",
"How does that
make you feel?",
"How do you
feel when you say that?"]]
]
#----------------------------------------------------------------------
# command_interface
#----------------------------------------------------------------------
def command_interface():
print
"Therapist\n---------"
print "Talk to the program
by typing in plain English, using normal upper-"
print 'and lower-case
letters and punctuation. Enter "quit" when done.'
print '='*72
print "Hello. How
are you feeling today?"
s = ""
therapist = eliza();
while s != "quit":
try: s =
raw_input(">")
except
EOFError:
s
= "quit"
print s
while s[-1] in
"!.": s = s[:-1]
print
therapist.respond(s)
if __name__ == "__main__":
command_interface()