Conflict Resolution

"Why accept the question that is asked?"— Paul Pangaro

Due to the limitations of AI today, voice assistants today serve users in a transactional manner. In this project, I reimagined how a conversational user interface (CUI) could help a user resolve conflict by shaping and challenging their way of thinking.

Organization

Designing for Conversation (CMU course)

Project Duration

6 weeks

Role

Conversation Designer

The Challenge

How can an agent challenge the user through conversation to create new value?

Conversation happens because at least one person has a goal that they can't accomplish themselves. In human-agent interaction, agents have the capacity to receive and respond to one goal at a time dictated by the user. Could the agent break out of users' expectations and insert their own goals to shape the user instead?

Furthermore, how can agents particularly add value to situations that ultimately necessitate person-to-person conversation—like conflict?

Opportunity

In speaking about resolving conflict, psychologist Susan Heitler refers to the "win-win waltz" as a series of steps during the treatment process. It includes expressing initial positions, exploring underlying concerns, and establishing a mutually agreed-upon plan. In the scope of a single user, a good opportunity area for the CUI to intervene is during the first two steps and prepping for the third.

Existing chatbots, like ELIZA, utilize "pattern matching" to give responses to make the user feel they were speaking with someone that understood their input. Reflective probing can help the user process the situation and surface hidden considerations.

Defining Scope

To narrow scope, we chose to focus on conflict in the workplace. Workplace conflicts happen frequently and ignoring them can be distressing and costly. Because they involve many factors, such as power dynamics, multiple parties, and palpable consequences, it can be hard for employees to feel like they have a safe outlet to manage conflicts.

Why Voice Agent?

Conflict is resolved through communication to work together rather than combat each other in order to overcome issues. Although resolve requires human-to-human interaction, voice agents have particular advantages over traditional human interaction and screen-based chatbots.

Voice Agents vs. Screens

• Emotion is not fully conveyed over screens
• Verbalizing out loud helps internalize thoughts

Voice Agents vs. Humans

• Using voice agents reduces time and human resources spent
• CUIs are unbiased and confidential

Modeling Conversation

We focused on the following scenario: our user has run into a specific problem in their workplace as their manager is taking credit for their work.

We first created rough scripts of conversations and annotated the range of turn sequences the CUI and user would make. Using the Cybernetic Modeling of conversations, I modeled goals and means pairings. For each paring, I would ask: might executing this <means> achieve this <goal> shown before it?

I explored different modes between the participants throughout the conversation, including controlling, guiding, delegating, and collaborating. Controlling and delegating modes are commonly employed when speaking to agents. (For example, the user may say "Order a milkshake for me from DoorDash" or "Could you get a milkshake for me?" respectively.) However, I postured the CUI to integrate guiding and collaborative turns into the conversation more. This way, the user is encouraged to verbalize their specific feelings and thoughts as opposed to being served generic strategies.

Building the Prototype

We mapped the conversation based on possible paths constructed in FigJam. The left section consists of exploring feelings about the conflict while the right utilizes the findings from the first to prepare the user for possible next steps, like confronting the other party in the conflict.

To create variations in branching, we thought of factors relating to the scenario, including:

These factors could result in infinite branches, but for the purpose of this project, each produced at least one split to create a variety of possibilities.

Next, we transposed the mapping into Voiceflow, coding in possible utterances to go down specific branches and creating slots to make the conversation personal to the user.

Key Turns

Below are highlighted turns that particularly show the potential of CUIs to challenge and reframe how users think.

1. The agent helps the user find clarity and reasoning behind their goals.

People might be unknowing of additional factors that shape the way they are thinking. The agent helps the user unpack their motivations to understand why they feel a certain way towards the other party.

Agent: How did you imagine the situation to go?

User: I expected my supervisor to redirect the compliments to me, but that never happened.

Agent: It sounds like there was a mismatch in expectation.

User: I don’t need thanks for everything I’ve done, but this presentation was a big one and it was my chance to be seen by leadership.

Agent: What do you mean by “chance to be seen by leadership”?

User: I want to accelerate in my career, and one way of ensuring that is proving your worth with the higher-ups.

Agent: Do you think that it’s possible that you feel like the situation is worse than it actually is because you want that promotion?

User: Possibly, is that a problem?

Agent: I don’t think it’s wrong to want a promotion, but sometimes higher motivations influence other feelings.

User: You make a fair point. I think my goal to get promoted makes interactions with him extra touchy.

2. The agent helps the user reconsider their original goals.

Users may come to the agent with clear intent. But is this the best intent? The agent challenges the user's expectations coming into the conversation and reshapes their goals.

Agent: I'm sorry to hear that. Can you tell me more about what happened?

User: He always does this in our weekly meetings.

Agent: Tell me about the instances of this happening.

User: Last month I conducted research and he shared out the results to the team. And...this past time, he took my research frameworks I wrote and implemented the process without giving me credit.

Agent: This doesn’t sound like “always” to me—rather two times.

User: Oh, I think it feels like always.

Agent: Either these two instances felt really impactful, or there are more reasons why you feel unsettled.

User: Now that you mention it, I think there is more. I’ve been asking for more opportunities because I’m growing fast. And, I think there’s tension there because I’m growing stronger in areas he can’t.

Agent: Maybe it might be a good time to have a larger conversation about your goals and personal growth at the company.

3. The agent gives advice on how to approach others.

In preparation for next steps, the agent can give feedback on what the user plans to say. AI is particularly useful here for utilizing sentiment analysis to assess the user's delivery.

Agent: What do you think you should do next?

User: Probably think about how to say something to them?

Agent: Yes! That sounds like a great plan. So how might you start?

User: I think I would say something along the lines of “Stop taking credit of my work! It is super discouraging when you do that and makes me feel like I am not being appreciated!” How does that sound?

Agent: Hmm...that’s a little direct which might throw them off. Do you think there are other ways you might be able to approach it?

User: Maybe I would say something along the lines of “I feel discouraged that I’m not mentioned in the discussion around the work I’ve done. I know it isn’t intentional but I would really appreciate being recognized for the work I've done.” How does that sound?

Agent: That’s a great start. Remember to be ready to hear your supervisor out and make sure the narratives in both of your heads are heard.

Reflection

Currently, digital systems accommodate and mold to the habits and preferences of users. Machines do not often diagnose what the user is doing poorly or wrong. They now are evaluated on how well they fulfill the user's needs: the user is rarely at fault for anything, and the machine should adjust. How interesting would it be for systems to challenge us and bring us to a future state we don't expect?

It boils down to what we want systems to become. Right now, the benefit from technologies ends at the human's ask, and rarely beyond. Services are transactional for now: if the user needs something, the machine fulfills it.

On our end as a society, I foresee us becoming more accepting of being challenged and brought to new norms through machines. Perfecting algorithms to provide what we want has brought upon new norms, but not in the way we imagined. It rather has fueled echo chambers, depression, and destruction. Having a more active posture with digital systems will allow us and them to grow and change together.

Team