AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

Turing Test requirements
 
Poll
What is the relative importance you place, in your project, on these components, (highest priority to lowest)?
1, 2, 3 2
1, 3, 2 3
2, 3, 1 0
2, 1, 3 0
3, 1, 2 2
3, 2, 1 0
I don’t care about those components, my system will break down the problem differently. 1
Total Votes: 8
You must be a logged-in member to vote
 

Having a software agent pass a Turing Test will require three major hurdles:

[1] Robust, flexible and very efficient algorithms to handle language.  So, the system’s ability to parse language, handle things like word sense disambiguation, anaphoric resolution, etc.

[2] An absolutely colossal amount of knowledge of the world (since one may ask anything question about any topic).  So, just the actual data that the system will need, right down to very basic things like ‘water is wet’.

[3] Handling human error.  Things that really have nothing to do with intelligence (in fact, more to do with stupidity), things like handling spelling errors, usage of incorrect terms, extra words in a sentence that do not need to be there, and missing words that did need to be there.  In the other hand, I guess one way to test intelligence is the ability to handle stupidity.

What is the relative importance you place, in your project, on these components, (highest priority to lowest)?

 

 
  [ # 1 ]

Very thoughtful questions Victor.

I put #2 last since knowledge acquisition is my primary goal and #1 first because I think it has to be solved before #3 can be tackled effectively. Also, in my system, the lines between #1 and #2 are a bit blurred because all my system’s knowledge about language is encoded as a very high level abstraction and counts as data, rather than being hard-coded in the software. The software takes that data and compiles it into executable code for efficiency, just as I expect to be able to do with other kinds of data as well. For what it’s worth, the Prolog programming language does that too.

 

 
  [ # 2 ]

Thanks, yes, you may be wondering why I even asked this question since in the other thread I stated I wasn’t interested in pursuing the TT.  But, doesn’t mean I’m not interested in others’ ideas, since I know many people on here ARE pursuing the TT.

Interesting, I also don’t treat grammar information (what I call “grammar production knowledge”, since when people hear “grammar rules” they think “hard coded”) as part of the program, but as information the program uses.  That is interesting that you compile the ‘production knowledge’ as well.  Right now in my system I have a perl script that generates low level C++, but that may change to compiling straight to machine language in the future.

 

 
  [ # 3 ]

Good poll. I pick the last option not really because I don’t care but because I think these 3 are somehow the same thing. Knowledge of the world is required in parsing a sentence/paragraph of a certain language, and human error has to be handled because it will happen. Actually some popular human error will be accepted as correct later if they’re popular enough. Grammer is very important because it controls the meaning of a sentence, eg “He is the manager” vs “He was the manager”.

TT is good tool to test AI but my bot is still far from making sense in chatting. I hope LP will continue and once my bot is ready I will for sure take the LP contest.

 

 
  [ # 4 ]

Great questions Victor!

If the focus is on just passing the Turing Test then my choice would be 1,3,2. With my current project (Marie), I would place the priorities more as, 2,1,3, because of the bot’s task oriented nature. Although I have been programming for over 25 years, this is my first AI Chatbot and it has been an eye opening learning experience. I also place a real importance in user engagement through an effective interface design since Marie is to be ultimately a commercial venture.

 

 
  [ # 5 ]

3, 1, 2

I tend to blend 3 and 1. Handling human error is a fact of life if you have an on-line bot. Even perfect English speakers make typos and use a kind of shorthand when chatting on-line. They often use the wrong word in a phrase. I don’t handle much in the way of things like word sense disambiguation, anaphoric resolution, etc, but I put a high value on robust, flexible and very efficient algorithms to handle language.

I now look at the AI input recognition task as a continuum.

BAG OF WORDS <——> EXACT STATEMENT

naive Bayes probability <——> Exact pattern match

As you use more advanced features you move more to the middle.

Generation of the output is another task all together. Actually fooling someone into thinking a computer is another human means build a personality which is a whole different discussion topic.

 

 
  [ # 6 ]
John Li - Jan 23, 2012:

Actually some popular human error will be accepted as correct later if they’re popular enough.

Hum… *accepted* perhaps, *accepted as correct*, depends who you ask.  If that person is me, then answer is no smile  The slag “ain’t” (I suppose meaning “isn’t”)  has been around for, I’m guessing, 100 years or more perhaps, and even the people that use it, know it is *not correct*.  So, yes, we may accept it, but we’ll always know it is incorrect.  But yes, I see your point, and I do intend on having my own bot accept some very common incorrect usages of languages.  A really great example of the ILLOGIC of the human mind is the slang, “Don’t do nothing”.  This, of course, logically means DO SOMETHING.  Yet, to many people, this somehow has the meaning “Don’t do ANYTHING”.  Utterly illogical, but that is just the “agreed upon” meaning of “don’t do nothing”.  It is the height of human stupidity… sad but true.

Laura Patterson - Jan 24, 2012:

Great questions Victor!

Thanks, and all the answers and comments are very interesting to read!

Laura Patterson - Jan 24, 2012:

Although I have been programming for over 25 years, this is my first AI Chatbot and it has been an eye opening learning experience.

Agreed.  It was+is an extemely interesting journey so far for me as well.

Merlin - Jan 24, 2012:

I now look at the AI input recognition task as a continuum.

BAG OF WORDS <——> EXACT STATEMENT
naive Bayes probability <——> Exact pattern match

Very interesting.  Recently I have updated my calculation of the crediblity of a parse, to not simply be a single scalar value, but a matrix.

 

 

 
  [ # 7 ]

That is something that I had thought about early on in development. The proper use of words is not always guaranteed, so this is where word association (reverse) is useful. Not only the use of American and British slang but also profanity (God forbid!) can be reversed associated and dealt with accordingly.

Half of the task of building an effective (intelligent) Chatbot is understanding human nature and that may be the biggest mystery of them all wink

 

 
  [ # 8 ]

Hum… *accepted* perhaps, *accepted as correct*, depends who you ask.  If that person is me, then answer is no smile The slag “ain’t” (I suppose meaning “isn’t”)  has been around for, I’m guessing, 100 years or more perhaps, and even the people that use it, know it is *not correct*.

So, does this mean you still speak like” though art not…’ or ‘you canst…)?

I find it somewhat a difficult question, kind of like what came first: the chicken or the egg?
Perhaps error detection can come last, since you need an expert system for that (how many kids are able to find grammar errors?).
You also need some routines to build up the data, so I guess, 1, 2,3

 

 
  [ # 9 ]
Victor Shulist - Jan 24, 2012:

Hum… *accepted* perhaps, *accepted as correct*, depends who you ask.  If that person is me, then answer is no smile  The slag “ain’t” (I suppose meaning “isn’t”)  has been around for, I’m guessing, 100 years or more perhaps, and even the people that use it, know it is *not correct*.  So, yes, we may accept it, but we’ll always know it is incorrect.  But yes, I see your point, and I do intend on having my own bot accept some very common incorrect usages of languages.  A really great example of the ILLOGIC of the human mind is the slang, “Don’t do nothing”.  This, of course, logically means DO SOMETHING.  Yet, to many people, this somehow has the meaning “Don’t do ANYTHING”.  Utterly illogical, but that is just the “agreed upon” meaning of “don’t do nothing”.  It is the height of human stupidity… sad but true.

Quite true, Victor, you gave a perfect example. Language changes, so I personally think chatbot should be able to recognize which grammer “set” should be applied to the inputted text, eg if the bot reads “Thou shall not ....” then it should expect “thy,  thee” rather than “you, your”; And if something like “I didn’t do nothing” is found then it should “know” that the input is probably not a formal document like legal agreement or business report.

But, when there’s a quotation, it’s a different story, eg. “You read the paragraph starting from “Thou shall not…” and then you are done.”

Maybe add another point for “style”? This is quite true for Chinese as well, ancient writings are quite different from the current, and knowing the style is the key to parsing.

 

 
  [ # 10 ]
Jan Bogaerts - Jan 24, 2012:

Hum… *accepted* perhaps, *accepted as correct*, depends who you ask.  If that person is me, then answer is no smile The slag “ain’t” (I suppose meaning “isn’t”)  has been around for, I’m guessing, 100 years or more perhaps, and even the people that use it, know it is *not correct*.

So, does this mean you still speak like” though art not…’ or ‘you canst…)?

Does that mean ifyou , in error, were to write ‘2002’ instead of ‘2012’...oopsy!  I guess it is the year 2002 instead of 2012?  If ,by some weird coincidence,  “many” (whatever number you deem is required) accidently wrote 2002, would everybody consider the current year to be 2002?

It is an interesting question.  I think I will study this more… how and why did English change from Old English.  I mean, was it REALLY just people making mistakes in language?

If so, why do we still consider “ain’t” to be incorrect? (everyone I know does anyway), but yet official Moden English words are not incorrect?

I think if a language is changed, it should be documented.  Simply saying that, ok, 51 % of the people say it is 2002, then, ok, guess what, we accept the year as being 2012.

Oh, enough people made the error that 1+1=3, oopsy, i guess 1+1=3.

Call me old fashioned, but I think, if a change is made, it should be centrally documented, officially agreed upon, and then taught in school.  Yes, you can use slang, but we still know it is slang, and incorrect….. or perhaps i’m just a ‘control freak’ lol smile

 

 
  [ # 11 ]

I mean, was it REALLY just people making mistakes in language?

some other reasons I can come up with:
-fashion (French dropping the rolling r)
-influences from foreign languages
-convention

Does that mean ifyou , in error, were to write ‘2002’ instead of ‘2012’...oopsy!  I guess it is the year 2002 instead of 2012?  If ,by some weird coincidence,  “many” (whatever number you deem is required) accidently wrote 2002, would everybody consider the current year to be 2002?

Ask a 2 year old.

I think if a language is changed, it should be documented.  Simply saying that, ok, 51 % of the people say it is 2002, then, ok, guess what, we accept the year as being 2012.

A standard language is just that: a documented convention. Take Dutch for instance: the official version used to be spoken by nobody, it was made up from a collection of all the dialects and then forced upon the population through the school system (much how they got everyone to speak Frencin France round the same time). There is a commission which updates the specifications every couple of years. People who learned to write 30 years ago, had to relearn (or can’t spell correctly anymore, which is the majority).

 

 
  [ # 12 ]
Jan Bogaerts - Jan 24, 2012:

Ask a 2 year old.

Hum… I thought I was smile

No, I know that is a silly example, but, just an extreme case to illustrate the point, that you just don’t simply say, a certain percentage of people are making an error, thus we’ll accept it as correct.

When you asked me if I only spoke Old English, you should have re-read my post.  I said I would accept , but not accept AS BEING CORRECT.    My bot will, for example, accept “aint”, and I even say it , just for fun sometimes, but I do NOT accept it as being correct.

 

 
  [ # 13 ]
John Li - Jan 24, 2012:

Quite true, Victor, you gave a perfect example. Language changes, so I personally think chatbot should be able to recognize .......

.......But, when there’s a quotation, it’s a different story, eg. “You read the paragraph starting from “Thou shall not…” and then you are done.”

Maybe add another point for “style”? This is quite true for Chinese as well, ancient writings are quite different from the current, and knowing the style is the key to parsing.

Yes, and, like I say, I’ll have my bot *accept*  (again, but not as correct), many common slang.

Example, teenagers have a language all their own.  They are currently using the word “sick” to mean something, I **think** to mean “exciting” or “exceptional” ??!!?!

Laura Patterson - Jan 24, 2012:

That is something that I had thought about early on in development. The proper use of words is not always guaranteed, so this is where word association (reverse) is useful. Not only the use of American and British slang but also profanity (God forbid!) can be reversed associated and dealt with accordingly.

Half of the task of building an effective (intelligent) Chatbot is understanding human nature and that may be the biggest mystery of them all wink

Interesting

 

 
  [ # 14 ]

When you asked me if I only spoke Old English, you should have re-read my post.  I said I would accept , but not accept AS BEING CORRECT.  My bot will, for example, accept “aint”, and I even say it , just for fun sometimes, but I do NOT accept it as being correct.

I was just yanking your chain. wink

 

 
  [ # 15 ]
Jan Bogaerts - Jan 24, 2012:

When you asked me if I only spoke Old English, you should have re-read my post.  I said I would accept , but not accept AS BEING CORRECT.  My bot will, for example, accept “aint”, and I even say it , just for fun sometimes, but I do NOT accept it as being correct.

I was just yanking your chain. wink

LOL… I know… same here with the 2002/2012 silly thing.

 

 1 2 > 
1 of 2
 
  login or register to react