Please discuss the role of knowledge bases for conversational AI. Possible topics include: ontologies, ontology design, works well / doesn't work, what are most pressing open questions, how to leverage the KB in the conversation?
What are the major challenges and steps in bridging dialog systems in research and in production?
3
Vote
Abhinav
Derek Schaper
Matias Mora
What are the viable approaches for multi-modal dialogue,e.g. web interfere,speaker video,robotic vision,etc.
Can be combined with "what are the challenges in using multi-modal datasets to build better dialo...more
by NA
2
Vote
EOK
Qing
How will I know how much your system knows about me? Is there a neat method of allowing the user to easily ask (or come to know) what information the system is using about them during the dialogue? How familiar and personalised are these systems going to become?
by Chris Watkins
Great way of formulating transparency on privacy concerns.
2
Vote
EOK
Matias Mora
What is the best way to start dialogue modelling on a completely new domain/application? How would you structure this application if you knew from the outset that one needed this application delivered in multiple languages.
2
Vote
Derek Schaper
Pablo Duboue
what are the challenges in using multi-modal datasets to build better dialogue agents ?
Can be combined with "What are the viable approaches for multi-modal dialogue,e.g. web interfere,...more
by NA
1
Vote
EOK
Considering that large scale production systems present vastly different challenges compared to experimental scenarios (e.g. data relevant to changing use cases, continuous performance monitoring, frequent retraining), new innovations may not necessarily be applicable in reality. How can we close...more
by Matias Mora
1
Vote
Nick
How would you apply limits to a neural conversational AI such that an agent could generate creative responses while respecting a generally prescribed policy? For example, an agent deployed in industry representing a business needs to ensure it doesn't say anything that contradicts business object...more
1
Vote
Liz
How to deal with sensitive topics which a user might not want to say out loud? E.g., a user in a public place needing a doctor's appointment.
by Martin Tapp
1
Vote
Pablo Duboue
Is it possible to have a conversational agent with also having a grounded understanding of language? In other words, if the agent participating in the conversation is not maintaining and evolving a world model, can it have a meaningful conversation?
1
Vote
vaibhava goel
How important is it to go beyond differentiable loss functions when designing open domain chat bots ? What lies beyond cross-entropy and neg lot likelihood ?
1
Vote
Abhinav
How many of our current challenges will go away if we had 100x-1000x more data?
by Gerry Tesauro
1
Vote
Pablo Duboue
When do you think we'll pass the Turing Test?
1
Vote
NA
What are some of the ethical and safety challenges you expect when productionizing data-driven dialog systems? Do you think these should be considered when doing purely research-based development?
1
Vote
Pablo Duboue
Conversation involves two (or more) parties. How can we build a solo Conversational AI?
by AR
We tend to model conversational AI systems as individual agents. But conversation is a multi-agen...more
by AR
0
Vote
We talked about a lot of structured dataset like Q&A dataset. What can we do with unstructured conversation without clear breaks
by Yang Liu
0
Vote
Augmented Conversational AI
by Alex
How can we build systems where there's permanently a human in the loop to catch mistakes and edge...more
0
Vote
What is the current state of affairs with generating longer answers, with complex internal narrative and grounding in external knowledge graphs?
by Michael
0
Vote
Any good metrics of quality of longer (e.g., > 1 page) generated answers?
by Michael
0
Vote
What do you think about the ability to compose(composition) word sequences by RNN structures is impeded by the word embeddings or the architecture of RNN?
0
Vote
what do you think is the right way forward in designing better metrics for dialogue evaluation ?
0
Vote
how do you think of the following statement about user simulators: "user behavior is more complicated than agent. If you can build a good user simulator for the RL algorithm to train, then you can might as well build a good agent simulator or agent policy using supervised learning. Then why shoul...more
0
Vote
Can the problem of conversational AI be re-formulated as a collaborative game between user and the bot given the success of Self-play RL similar to AlphaGo? End game situation is not one player winning but having a high conversation rating as a collaboration between two players.
0
Vote
How do you regulate bias in conversational agents?
by Sanjana
0
Vote
Given conversation is subjective, how can we evaluate the quality of Conversational AI?
by AR
0
Vote
What are the main challenges to move from Q&A agents to conversational agents?