Ai Dreams Forum
Artificial Intelligence => General AI Discussion => Topic started by: chattable on July 14, 2021, 12:03:06 pm
-
are their any chatbots that make up their own questions from their own database?
what i mean is they find out what they do not know from their database then ask a question about that.
-
I don't think any bot will decide what question to ask, without it being explicitly coded.
The closest I have to that is if Kuki is stuck for something to say, she will go through what she knows about the user and if she doesn't know something about them, she will ask questions like, "What is your job?", "Who is your best friend?", "Do you have any pets?" etc
-
the chatbot would have this in it's database
a cat is a animal.
a dog is a animal.
spaghetti is a food.
pudding is a food.
then the chatbot could make up the questions is a cat a food?
is a dog a food?
because food would be assigned as a category.
if it asked itself a question and it could answer it.
the chatbot would not ask that question.
because it would detect it's own default saying.
-
Acuitas does this: http://writerofminds.blogspot.com/2017/08/acuitas-diary-5-august-2017.html
-
does acuitas learn the answer to questions if you say yes or no to a question?
example
acuitas = is a dog a food
me = no
acuitas = i understand a dog is not food
acuitas = is a cat an animal
me = yes
acuitas = i undertand a cat is a animal now.
-
does acuitas learn the answer to questions if you say yes or no to a question?
example
acuitas = is a dog a food
me = no
acuitas = i understand a dog is not food
acuitas = is a cat an animal
me = yes
acuitas = i undertand a cat is a animal now.
Yes. He can detect (some) valid responses to his questions and learn the answer.
-
can he detect those?
-
Yeah. The yes/no responses are easy.
-
For a while I had my program ask straight questions about facts from its database that it was uncertain of, not unlike your description. This mainly led to random questions like "Can dogs breathe?". If one answered the question it would update its database and not ask a second time, but as they say in my country, one fool can ask more questions than a thousand scholars can answer. Its ignorance was limitless and it hindered conversation more than improved it, so I dropped this line of questioning. The questions would need much more direction to make them relevant.
-
the chatbots could try to guess the answer to the question it makes up sometimes.
then the user could say yes or no.
then the chatbot could learn like that.
that would be funny i think.
-
Oh chatbots can do that. A classic example I use is if someone asks, "Can you eat a tree?". I haven't hardcoded a response to that but Kuki can look up the properties of a tree, find it's made from wood, realise that you can't eat wood and reply, "no".
-
are you using concept net in kuki?
-
an easy way to utilize this is to assign the synonyms to its chat patterns, but u may as well just as tell them it all at the beginning, gets the job done easier. But I guess maybe theres something more interesting to do! dont know tho. :)
-
this is about getting a chatbot to think for itself.
once a chatbot learns more categories it could make up more questions.
having the chatbot ask questions about the categories as well. may make it more interesting.
-
this is about getting a chatbot to think for itself.
once a chatbot learns more categories it could make up more questions.
having the chatbot ask questions about the categories as well. may make it more interesting.
If it doesnt go into some kind of fixed framework, I dont think its possible. But it could still be fun.
-
example=
a cat is a animal
a couch is a type of furniture
a chatbot could identify
the index position and the strings in those sentences.
then put the last word into a list of categories if not in the list of categories.
then it could ask questions like this.
are animals furniture?
only if it asked that question to itself and it could not answer the question.
it seems possible to me.
-
Yeh but it has to say something educational about it at least... like when it is in a doungeon, it can know what dragons are, know they live in doungeons, but it has to make the dragon act like a dragon, putting all the knowledge together into a verbacular activity, that would be something. (thats actually close to what im doing.)
But I bet theres something more than that as well. Just doing action directly, is possibly not as clever as being able to "tell a story about the dragon" in a more indirect way, or talk about dragons, different "tenses??"
Its the pipeline from input to output, not just the data store and acquisition!
-
are you using concept net in kuki?
No, it's all handwritten AIML rules. I wrote a blog about it back in 2018 here:
https://medium.com/pandorabots-blog/can-you-eat-a-chair-aaa80d251f6b
Kuki used to be called Mitsuku.
-
what is your project magnuswootton?
have you played ai dungeon?
i made a adult chatbot on the personalityforge websight named lilaith x which is a succubus.
she eats,sleeps,changes her clothes.
those are the role play actions.
she can tell you the name of a thing in a particular category and answer a lot of questions.
you can ask her what she is wearing.
she is programmed to let you walk to different places in the house.
you can touch her in roleplay and tickle her.
-
I do a lot of 3d space locomotion stuff, but I would love to cross it with "text" in the form of symbolic representation of a visual scene, its an idea id like to see put together, but havent had the strength to get it done.
It would look further in front, because its all been simplified, I was think of doing the initial spacial geometry numerical stuff, then it slowly disappears off into symbols for a general description of whats there.
If you sequence activities of activities of activities, it seems to work nicely, and u can get alot more in front simulation distance, if u keep the amount of activities down to a repeating set, it can be brute forced, with not too much computation power.
And its like a chat bot, except its modifying the scene with its hands, instead of talking, but its text the same. :)
but it doesnt learn the sequences, i give it the map of its activities, and program how they "replug" the symbols.
-
okay.