In the following, I will describe a few possible applications. I don’t claim that dialog markets are appropriate for all of these; I mainly want to illustrate that there are many candidates for applications, so our approach are probably appropriate for some, and there is some indication that this may be a useful domain-general tool.
Table of contents
- Making domain expertise available
- Knowledge transmission within organizations
- Product recommendations and ads
- Various kinds of personal advice
- Argument mapping and checking
- Referral to existing tools
- Explanations & teaching
- Research funding
Making domain expertise available
Consider domain experts such as lawyers, doctors, and tax accountants. Dialog markets could provide a lower-friction way for experts to turn their knowledge into money (although domain-specific restrictions apply, as is particularly clear in the case of lawyers). Dialog markets could also automate the easy, repetitive parts of experts’ jobs. If certain questions always result in the same follow-up questions, we can detect this pattern and then ask the appropriate follow-up questions automatically.
Knowledge transmission within organizations
Dialogs could answer questions that employees in big companies have about the workings of the company. Similarly, the management of that company could elicit information from the company at large (e.g., “Do our customers in Asia like our new product?”). Instead of working down the command hierarchy, they could directly pose the question to the company and assign bonuses for helpful contributions.
In a different project—the HelperBot—I explored whether an automated question-asking system could be useful for thinking about various issues in one’s life. My impression is that it can be somewhat useful for some people, but that good prioritization of questions to ask next requires human ingenuity, or at least human feedback, and that, more generally, we may not know enough about knowledge representation yet to do this with much success. By contrast, in my (and others’) experience, dialogs with other people often do seem to help with tough problems. In part, this is because they help us get over activation thresholds—it is easier to answer questions than to push the investigation forward on one’s own.
As a particular instantiation of this application, we could consider “semi-automated cognitive behavioral therapy” where machines ask some follow-up questions.
Product recommendations and ads
People often look for good products and services. For example, I have recently been thinking about whether to buy a car, and if so, which one. The right answer depends on what exactly my requirements are, so it is important to ask me questions to figure out what car might be appropriate for me.
Suppose that putting up answers has some monetary cost. If the incentive system is set up appropriately, then people could put up advertisements without cost only when they expect them to be helpful to users. People could put them up when they are not helpful, in which case they would subsidize the other participants of the system by paying a fee that they don’t get back. Of course, figuring out an incentive system such that dynamics like this emerge could be quite tricky.
Various kinds of personal advice
There are existing services that provide personalized advice. For example, there is personalized voting advice (automated e.g. in the “Wahl-o-mat” in Germany) and personalized career advice. The system could take in the fixed decision trees used by existing tools (through bots) and then incrementally improve their answers through properly incentivized dialogs.
As another example, we can imagine personalized advice on what charity to give to.
We can also think about situations where we (as the asker) are looking to generate a large number of solution candidates, and then will pick a single answer ourselves. This could happen within the dialog system by restricting the answer of the top-level question to the author of the question, allowing the general public only to add subquestions and -answers.
For example, there are existing websites that offer community-driven brainstorming of company names in exchange for a fee; this could be done as a special case of the more general dialog system described here.
Argument mapping and checking
Suppose I am interested in the current state of the debate on whether humans are causing global warming. I could put up this question, pay a price towards it, and then have others flesh out the debate for me.
The system may even be useful in cases where one doesn’t expect others to do much work in answering the question per se, but simply wants others to check one’s arguments. As a researcher, I might want to ask one of my research questions, and then add subquestions and answers that reflect my current understanding of the topic, in the hope that others will ask helpful questions, or will help me make progress by giving feedback on my current thoughts. Think about it as a structured editor for thoughts, with help by others in proportion to how much money you pledge. If you think your own thoughts are very high-quality and complete, you could pledge a large amount of money on your dialog (well, monolog) once you have fleshed it out. Then others would be incentivized to judge your argument with high precision, and to find places where your argument “leaks money”, as that would be an opportunity for them to earn part of your pledge.
Given a large-enough public repository of arguments that make money from questions that they are answering, you could imagine people (or, more speculatively, programs) scavenging for flawed arguments, as those are an opportunity for redirecting the flow of money towards the program’s owner if they can fix the argument.
Referral to existing tools
Some questions can be answered without dialogs. For many questions, a referral to existing tools (e.g., web search, route planning, Quora, StackOverflow) might be all that is required. It is easy to imagine bots that detect some of these cases and that provide links to answers given by existing tools. Indeed, if the system described here were to achieve some popularity, the owners of existing tools would have incentive to provide their answers within the system to make money and/or get referrals.
Explanations & teaching
When we first learn about a topic, our background knowledge determines what constitutes useful information. It would make sense for a learning system to ask a few questions to determine what we already know before it presents information. It could also ask questions to check our understanding.
This is feasible with human tutors, but would scale much better with bots that have more sophisticated knowledge representation than what is possible using current AI techniques, so nontrivial versions may be off some ways into the future.
On a larger scale than previous applications, funding agencies could pledge rewards on general questions such as “How does brain work?”. Researchers could then ask and work on relevant subquestions. This has some advantages over the current journal- and grant-based system: small, incremental contributions are made easier and are rewarded more proportionally, especially for non-academics; negative results may be easier to publish if they are clearly relevant; and it may be easier for researchers to learn about the current state of investigation for a particular question.