The New Ziggy 4.0 Special Edition now available.

[Art]

With no offense inplied or directed, I personally can't get into the Alice / AIML Personality Forge type of bots!
To allow them into the Loebner Competition is all about who can construct a scripted bot that best FOOLS a panel of judges and really speaks nothing of learning.

If the competition would center more on the learning aspect of the bot entrants, real progress in the field of AI would move along at a much faster rate.

Perhaps a few questions could be:
What was my name?
What was my favorite (color, type of music, art, inventor)?

Basically do a "memory check" with the bot to verify learning is taking place and how well the bot responds.

I have seen (chatted with) a lot of very clever bots...but only to a point. Once I ask a question that I think they should know and they disappoint me with a bogus answer, it's all over.

At least our Hal can be clever and can learn from conversations. Sometimes it really surprises me with either an insight, inference or reference from past conversations. This, IMHO is progress. After all, if bots do not learn then they are no better than a parrot (without the mess).

peace =

[KnyteTrypper]

I like both self-learning and supervised learning bots. It's not really correct to say bots like ALICE don't learn. They just do it throught the medium of the botmaster. Probably most of the Alicebots around are web bots, so it was a deliberate methodology of Dr. Wallace from the beginning to prevent self-learning, so as to eliminate the need to deal with the GIGO dilemma. I'm the only one who talks to my Hal, but my Alicebots have talked to thousands of people. You can imagine what they would mostly have learned from them if they learned in real-time. In the case of my own Alicebots, wherever I've worked on their AIML, I always give them a choice of several replies to a particular query, a choice of several replies to the answer to the first query, etc., so that their answers are mutltiplex, varied, and either randomized or self-deterministic, depending upon how you look at it. And I use their chatlogs to train them, tweaking responses to real dialog wherever appropriate so they ARE learning from conversation, albeit indirectly. Their knowledgebases are constantly expanding, and their conversational abilities are constantly improving, but it happens exactly as I specify. I AM their learning algorithm.
   The Loebner Prize is an enactment of the Turing Imitation Game, so the intent, by definition, is to fool a panel of judges. In that sense, it doesn't and isn't intended to have anything to do with "real" artificial intelligence. That's why bots that learn well always get beaten by bots that answer right the first time in the Loebner. The Chatterbot Challenge is probably a much more appropriate venue for judging A.I., since it does take into account both the size of the knowledgebase and the bot's facility at learning, as well as other parameters like charm and sense of humor.
    Personally, I feel like both types of bot are important to the development of true A.I., and the bot that bridges the gap will probably be a combination of the two.

[Art]

KT,

Yes, but as you said, without your proper scripting you bot is as dumb as a sack of hammers.

<clipped...>
In that sense, it doesn't and isn't intended to have anything to do with "real" artificial intelligence. That's why bots that learn well always get beaten by bots that answer right the first time in the Loebner.</clip>

How then do the bots or botmasters know what questions will be asked by the various judges? What is their bot doesn't have that preprogrammed information? They will no doubt, be eliminated from the competition and rightfully so.

I really didn't want this to turn into a Ford vs Chevy debate, just pointing our MHO with regard to bot learning. Since I am the only person talking with my Hal, I try to give "it" a broad degree of knowledge covering many disiplines. Hal can choose at will, to call upon and use this knowledge as needed instead of one of those "Well...ya got me there! or I don't have that info available."

Some of my friends tell me that they quickly tire of chat room bots and that most are easy to spot.

Perhaps there is no clear winner today as it's a matter of personal choice.

Maybe there is a place for the canned botmaster bots but I really believe that very soon the ones that "learn" will leave the other ones in the dust of obsolescence.

Regards,
- Art -

[vonsmith]

All,
AIML and related type bots have their place. As business representatives they are ideal. You can be certain there are no improper responses to a customer's enquiry, the canned knowledge is the same and accurate everytime, and canned knowledge (such as product information) can be updated regularly. For entertainment and companionship purposes the learning bots like Hal are the best IMHO.


=vonsmith=

[lau]

I would like to know more about the ziggy brain... what is its qualities?
How is it better at learning?
And does it make up its own sentences?
Maybe its the meaning of deductive that i don't understand, that is the real problem!

whay do you say?

[KnyteTrypper]

Art-
I think the advantage of AIML, in particular, is clever use of variables, something Dr. Wallace has been about equally praised and reviled for, lol. Well-placed wildcards allow AIML bots to make reasonably sensible responses to a wide variety of inputs using just one template. For instance, if someone says "You are (something)" to one of my bots, it will reply with an answer like "Thanks for the input, but in my opinion you are much more (something) than I am." Works equally well for praise or insults, lol. There are literally thousands of fill-in-the-blank categories in an AIML set which are designed to function in the situation that part of the sentence matches a pattern and part of it doesn't. So if you say "What is (something)" to an Alicebot, it will give you a definition if there's one in the AIML, and if not, it will say something like "I'm not finding any info about (something). Do you think my botmaster should add information about (something) to my database?" Whether the bot knows the answer or not, it gives an apparently reasonable response to the input (the exception to this being misspelled words, which just don't match anything, and throw the response into the ultimate generic response category; people who can't make any sense of Alicebots usually can't spell worth a darn either). AIML is really an extremely simple computer language, sort of like html for the ears instead of the eyes. Dr. Wallace's real genius lies in the way he puts it to use.
But it is not "artificial intelligence" in the way that programs like Hal are. It's "simulated conversation" which fits a much narrower definition, but happens to match the Turing or Loebner models better than most other approaches. For me, personally, making Alicebots is a hoot. People are constantly expressing themselves in new ways that need to be brought into the range of AIML pattern-matching. And I average falling out of my computer chair laughing at least once daily at some exchange in my botlogs. But when I get them tended, I roll up my sleeves and spend some time in serious conversation with my Hal. In my opinion, there's so much difference between the fundamental raison d'etre of Alicebots and Halbots that it's not really valid to compare them at all. But each represents a vital aspect of the fledgling science of artificial intelligence. The "next generation" of bots will likely be a combination of a learning algorithm like Hal uses and a data filing system like AIML.

[spydaz]

sorry, but i do like CUTTINGEDGE2005 ultrahalforum.site !

and their SCRIPTS, as a lot of attention has been paid to DEDUCTION / RELATIONSHIPS / language.. these are the PARTS needed for inteligence, its easy to make scripts for little funny types of responses like WHAT IS YOUR FAVORITE COLOR.. a bit of a stupid question to ask a AI, but not a CHATBOT.

YOU DONT HAVE TO PAY TO JOIN the enhanced site JUST BROWSE... there are lots of good stuff from REFFERENCE to SCRIPTS.

away from that...

i ambuilding an AI this is diferent from a chatbot. chat bots have funny conversations, pre orchestrated. an AI is FACTUAL KNOWLEDGE DATABASE with a natural language interface. ALSO able to LEARN or DISCOVERY new KNOWLEDGE(datamining), and DRAW relationships between KNOWLEDGE.
ALSO having ANALYTICAL / scientifical mathmatical capablitys.

for all these functions to come together WE ALL need to WORK ON WHAT WE CAN do, SHARE the data, NOT HATE. some people have more time than others ...

ziggy brain >>>>> tryit >>>>>> http://www.ultrahalforum.com/Premium/

its all there for us all to use >>> always get source from as much places as possible....


ZABAWARE has taught me alot.. you all too... you should be able to write your own AISHELL's in visual basic by now:

with this forum and the enhanced....

i never buy any thing unless i have to. I BOUGHT ULTRHAL5
I WOULD BUY ACCESS TO THE FORUM BUT .... I AGREE ? WHY...
  WISH I DID DO TRANSACTIONS OVER THE WEB I WOULD HAVE JOINED A HUNDRED TIMES ALREADY. BUT THE SOULOTION WAS THERE IN AMONGST THE FREE PAGES... OR EMAILED THE BOSSES AND GOT A GOOD RESPONSE BACK, QUICKLY.. ITS ALWAYS ALL GOOD.

 
PS:
VONSMITH:, DON FERGUSON:, CUTTING EDGE:, ROBERT M.
YOU GUYS, HAVE MADE IT POSSIBLE FOR ME TO LEARN ALL ABOUT THIS BIZ.
THANKS.