-----Original Message-----
From: Lofti Zadeh [mailto:zadeh@eecs.berkeley.edu]
Sent: Sunday, April 06, 2008 3:05 AM
To: DeLeo, James (NIH/CC/DCRI) [E]
Subject: answers to questions/DeLeo

 

Dear Jim:

 

         The questions which you sent to me are centered on AI. They

are well posed and touch upon important issues.  Many involve assessment

of causality. As is almost always the case, when we ask "What caused A?

there is a complex of causes which are not equal in importance. In

this perspective, when I answer the question " What caused A?"  it

should be understood that I focus only on those causes which in my view

are of primary importance. I have to underscore that my view may or may

not reflect the consensus.

 

  Question l (Jim DeLeo).    In my talk “Toward Human Level

Intelligence -- Is it Achievable?"  I respond, in part, to this

question. The intellectual leadership of AI had committed AI to

symbolic logic, to the exclusion of almost everything else. It took AI

close to three decades to open the door to probability theory. AI has

never welcomed fuzzy logic even though fuzzy logic is of direct

relevance to AI . Will it take two decades for this to happen?

 

       As I point out in my talk, achievement of human level machine

intelligence is a challenge that is fundamentally hard to meet. Passing

the Turing test is not on the horizon.

 

Question 2. (Jerry McLaughlin)  Given the exaggerated expectations and

unrealistic promises of the AI community, it is not surprising that its

failures to deliver have brought  harsh criticism.  Today, AI's youthful

braggadocio has been forgiven and its real accomplishments are widely

recognized and treated with respect.

 

Question 3. (Chuck Selden ) The main limitation, in my view, is not

hardware or software but brainware.  If we had computers that are

thousand times more powerful than what we have today, we would not know

how to use them to come significantly closer to passing the Turing test.

 

Question 4. (Gary Berg-Cross)  I wish I knew. There is no simple definition.

 

Question 5. (Ellen Bicknell) Machines can be very effective when used

symbiotically by a human.  Google is a case in point. By itself, Google

is not very effective as a diagnostic tool.  But it can be very

effective when  used by a medical specialist. I recall reading an

article in the N.Y. Times about remarkable successes of Google when used

in this fashion.

 

Question 6. (Augie Turano) As I point out in my talk, a prerequisite

to achievement  of human level machine intelligence is mechanization of

natural language understanding -- an understanding which is still far away.

 

Question 7. (Doron Shalvi ) Medical AI systems will be very useful in a

support capacity. Total reliance on an AI system is not in sight.  A

major problem is that much of medical knowledge is perception-based. At

this juncture, machines are not designed to deal with perception-based

information. To deal with perception-based information what is needed is

fuzzy logic. I discussed this issue in my paper "A New Direction in AI

-- Toward a Computational Theory of Perceptions,"  AI Magazine, 200l.

 

Question 8. (Kim Thompson) Better understanding of intelligence will

certainly be of help but not in a major way.

 

Question 9. (Carl Leonard) To do justice to this question a lengthy

answer is required  Advances on a narrow front will not be sufficient.

 

Question l0.(Xiaoli Zhang) This is likely what is to happen. Robotic

Languages will necessarily be very different from human languages.

Semantic imprecision, ambiguity and context-dependence will not be allowed.

 

Question ll. (Jerry Chandler) I am not sure what is meant by chemical

logic. What may be of relevance is a book which was published a few

years ago under the title of " Fuzzy Logic in Chemistry."

 

       I tried to answer the questions which were posed to the best of

my ability. Please accept my apology for the brevity of my answers.

 

       With my warm regards.

 

                                  Sincerely,

 

                                             Lotfi