Expert: Chuck Cosby - 8/21/2007

Question
QUESTION: Dear Chuck Cosby,

Can you tell if a program is intelligent or not?
How to tell that a program is intelligent or not?

Thank you,

Frans

ANSWER: The method I would use would only work with a program that handles natural speach and conversation.  You would talk to the computer.  If it has intellegent responses then that might be a good indicator.  You would want more than that, however.  Some programs mimic intelligence by making passable small talk.  That doesn't cut it.  Becuase it is a computer you expect it to be able to perform functions, like making travel arrangments, or manipulating a spreadsheet; all with conversational natural language interaction.  until a computer program can those things and do them well, then it is not even close to being intelligent.  PS - no program can do this yet.

---------- FOLLOW-UP ----------

QUESTION: Thank you for the answer.

How far AI has advanced? I mean, what is the most advanced AI scientist can do now?

Do you think we can build a program that close enough to human? Maybe not that close. At least we can talk to that program and that program can learn from conversations or learn from texts.

Thank you,

Frans.
ANSWER: AI is progressing extremely slowly.  We can't even talk to computers in a natural fashion about very limited subjects - like purchasing airline tickets.  there are speech recognition systems, but they are crude and un-responsive.  They ask simple questions - you must answer the question exactly as posed, you can't add any extra information in your answer. There are also 'dictation' systems - that let you talk and transcribe your speech to words on the screen - see Dragon Naturally Speaking.  These are not really 'AI' at all.  Basically we are no where near what you call 'AI'.

---------- FOLLOW-UP ----------

QUESTION: Hello,

Thank you for the answer. Mmmm... so there are great challenges before we get the real AI? Do you think we can have such technology?

Thank you,

Frans
ANSWER: It would have to be some kind of major breajthrough - or just decades and centuries of slow progress.  I have a suspestion that the human mind is far more complex and mysterious than anyone realizes.  The problem with a 'Real AI' is it has to mimic a complete human mind.  Lets' give 100 years and see where we are.

---------- FOLLOW-UP ----------

QUESTION: Hi,

Thank you so much for the reply. I have been studying AI autodidactly for 2 years. There are questions. I am sure it will be basic question to you. Do you mind if I ask you such questions?

Thank you.
ANSWER: sure - go ahead

---------- FOLLOW-UP ----------

QUESTION: Hi,

Thank you for the chance to ask you questions.
I have 4 questions. It is fine if you prefer answering one question at a time, at
least those question can give you a brief view of my knowledge.

1. Is First Order Logic a way to express thought? Like UML or flowchart is a way
  to express software structure? A midway before write codes?
2. Practice makes perfect. I have been doing the Wumpus problem to practice A*
  search algorithm, environment analysis and making decision. Do you know other
  problems I can use for practice?
3. I am planning to specialise in Natural Language Processing. Can you give me
  the road map to Natural Language Processing? What subjects I should learn? Can
  I just 'jump' into natural language books?
4. Do you know the character 'Data' in Star Trek? Is Data a fiction sample of
  'real AI'?

Thanks again for the chance to share knowledge.
ANSWER: #1 - I don't know anything about 'First Order Logic'  #2 I never heard of 'Wumpus' #3 I sugest focusing on concept matrix based systems.  These are systems that map words and phrases to a concept matrix to extract meaning.  You should avoid systems that are based on grammer.  Grammer is a dead end in the NL field - but is it strongly promoted in acedemia.   grammers are not language independant - concepts are.  #4 Yes 'Data' is an example of what I call 'Extreme AI' - in others words - highly unlikly techology in our lifetimes.

---------- FOLLOW-UP ----------

QUESTION: Hi,

Thank you for the replay.

You don't know 'First Order Logic'? So it is not a mandatory subject like you have to understand how to multiply in order to understand calculus. I only have one book 'Artificial Intelligence : Modern Approach' and First Order Logic is introduced in the book's first part.

What books you have read (from the first time you know AI until now)?

The matrix system.
Is it a system you developed or is it commonly used in NL?
Do you use neural network in NL? I was thinking/imaging that neural network can be added in NL so the system can learn new phrase, word, accumulate knowledge.

Thank you,

Frans

Answer
I don't read many books from others.  I have been developing my own technology for 32 years.  I have skimmed books and looked at some papers, and read a lot of patents.  (I have a patent myself) I also had a PH.D. computer scientist working for me for a while who was 'schooled' in all the approaches. Each group that works on NL will have their own terminology.  I don't care to become an expert on other peoples systems.  As far as the 'matrix' system - some others have used that term before.  My group has a very specific meaning for it, of course.  I do not use the neural network method.  it is a very primitive technology.  Nobody knows how to make it work.