What would you like to see happen in this conversation?
I've implemented active learning for the end design, but it's not great and I'd like to discuss it with you.
What kind of "active learning" is this "active learning"?
A method of machine learning in which the system asks humans questions about areas in which it is unsure, and humans answer them to efficiently collect teacher data.
What kind of "method" is this "method"?
It is a method to make AI efficient and smart.
Is there anything else you can tell us about the "method"?
Systematically, the question and answer cycle is starting to turn, but the current AI is too stupid to be very efficient.
What kind of "AI" is this "AI"?
What we're trying to do here is look at the flow of the dialogue and ask at the appropriate point, "Is this a good place to end?" and ask the question "Is this a good place to end?
A method to make AI more efficient and smarter
- Where is this "method" located?
I reused the one I implemented during "Motion Extraction."
A method to make AI more efficient and smarter
- Where is this "method"?
🙁
A method to make AI more efficient and smarter
- What does this "method" look like?
It's like a child asking me a question about something I'm interested in.
What kind of "interest" is that "interest"?
Identify how borderline cases should be identified when you want to identify Yes or No
What kind of "on the borderline" is that "on the borderline"?
Those with a probability of being Yes close to 50%.
What kind of "Yes" is that "Yes"?
The decision that "this is where we should ask the end of the line question".
What kind of "question" is that "question"?
Is this a good place to end up?"
What is it that you value?
When you ask that question at the right time and end with a "yes" or "let's end it then", it's more pleasant than if the dialogue continues to linger on and on.
Is there anything else about that "question"?
I haven't yet been able to verbalize very well what kind of timing I should do, either, so I tried to do it with machine learning.
What kind of "machine learning" is this "machine learning"?
A system in which a machine finds conditions by collecting data when a human cannot write conditional expressions.
What kind of "human being" is this "human being"?
I am.
What kind of "I" is that "I"?
I'm trying to implement this system, but I haven't yet been able to clearly verbalize when the end question should be posed.
Where is the "question"?
It should generally be around the time when dialogue has developed to some extent and settled down.
I'm trying to implement this system, but I haven't yet been able to clearly verbalize when to ask the end question.
- Where is this "question"?
After the user has gained something by answering a question, when there seems to be little to gain by continuing to ask more questions
How do you know what the "question" is?
When I coach people face to face, I judge by facial expressions, body language, and tension, but this system is a chat system, so it's hard to get information.
What kind of "system" is this "system"?
A system that encourages users to verbalize their inner world through dialogue
Is there anything else you can tell us about that "system"?
We encourage verbalization, so once something is verbalized, we have done our job.
What kind of "verbalization" is this "verbalization"?
If the user writes about a certain topic before using the system and after using the system, and if there is a difference between the two, then the difference is the effect of the linguistic support provided by the system.
What does that "system" look like?
This is the type of coaching that focuses on listening.
What is that "feeling" like?
The system does not give the answer, but expects crystallization to occur in the user by giving questions, which is the same thing.
Where does this "question" come from?
I took it from a technique called clean language or symbolic modeling.
How do you know what the "method" is?
Sure, maybe there's some criteria for asking the end question in the book you referenced or something.
What would you like to see happen regarding that "question"?
I'd like to see more language on what the determinants of when to ask a question are.
What needs to happen to the "question" for that to happen?
I should know more about when I myself should be asking this question.
As for the "question," can you do it?
Could be done by me reading the reference book back to back.
I should know more about when I myself should ask this question.
- What does this "question" seem to be?
(Here's where I ask, "Is this a good place to end?" I'd like to end with a resounding "yes."
https://keicho.netlify.app/#talk=SJSLzd0PCLcJ3Nzlfdc4
This page is auto-translated from /nishio/能動学習のうまくいかない話をしてたら本を読めという結論になった using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.