Three Fictional Perspectives
Before I go into the proposal per se, I will describe it from the perspective of three hypothetical users. This is intended as an introduction of the proposal that is easier to read at the cost of leaving out critical parts that are less visible to users.
Table of contents
A person with questions
“So I have a question I’m wondering about, really any kind of question. Recently, I’ve used it to better understand what’s wrong with my stomach before going to the doctor, and to figure out where we should go for our vacation.
I only use their app on my phone. I think they have a web page as well. I type in my question. Sometimes it immediately asks follow-up questions, sometimes it directly proposes an answer, but most of the time, it takes a while. If I care a lot about the answer, I can pledge money and I get better answers and follow-up questions.
Every now and then, I get a push notification on my phone—there are new comments on my question, or new follow-up questions for me. Most follow-up questions are simple multiple-choice questions, so it’s easy to answer them directly on my phone.
Over time, this builds up a dialog related to my initial question and I get more specific questions and proposals. Actually, I only participate in a small part of the process, the part that requires info about my life, but there are many subquestions that get resolved without me.
I don’t know where exactly the questions and answers are coming from—I think some are generated by computers, and some by humans? I just know that the dialogs are pretty helpful, like talking to a knowledgeable friend, at least for questions that a lot of other people have, too.
For more unusual questions, the dialogs take more time; in general, it’s not a great tool when I am in a hurry. But if I have time, it’s neat—I can ask a question, pay $10, get asked a bunch of easy-to-answer multiple-choice questions, and then get back a well-developed analysis of my question specific to my situation. Then, if I want to see more thought put into any subpart of the answer, I can just pay money towards the relevant subquestion and it will be used to reward mental work that helps solve it.”
A domain expert with answers
“I’m a med student by day. I don’t have much free time, but recently I’ve enjoyed participating in dialogs on that website.
I tend to search for medicine-related questions that I might be able to answer, but sometimes I also give other advice. If I find a question, I check the current state of the dialog. Sometimes, I can directly suggest possible answers based on the information that is there. Mostly, I ask a few clarifying questions and then get back later. I also like to check in on dialogs I have participated in earlier and ask further follow-up questions or give answers, but I don’t have to—if I don’t continue, others will usually take up the slack.
It’s enjoyable because it’s good practice for me, I know that I’m helping others, and I get paid if I make helpful contributions—even if I can’t directly answer someone’s question, which is the case most of the time.
I don’t repeat myself as often as you’d think; I think there is some sort of automation that reuses my previous answers in future dialogs.
Actually, most of what I do now is judging the quality of others’ questions and responses. This is even easier than writing responses, and I can do it from my phone. It gives me a question, some context, and a followup question or answer, and I give a rating 1-10 whether I think that it’s a good answer, and what I think other people will think about it.”
A programmer working on automation
“You know that site where you can get paid for contributing to dialogs? You don’t have to do it in person—you can also write programs to do this. If your programs produce good contributions to dialogs, you collect the same reward you would as a human contributor. They make the data for basically all dialogs available, in anonymized form. I’ve been writing programs that use data on past dialogs to automate parts of new dialogs.
Initially, I wrote programs that simply tried to find very similar existing dialogs and then asked follow-up questions that were asked and judged useful in earlier dialogs. This worked well for a while, but there is a lot of competition in that space now. Other people are writing programs that are way better at deciding whether two question strings have the same meaning than the simple bots I wrote.
Now I’m focusing on writing bots for specific domains. I’ve had good success with my med bot. I looked at a lot of medical dialogs, checked with text books, and my bot now successfully narrows down diagnoses for some symptoms. Unfortunately, a lot of people are working in this space nowadays, too, and they have better access to actual doctors than I do; I heard that there will soon be an entire company focusing on medical dialogs.”