A study on the effects of mixed explicit and implicit communications in human-artificial-agent interactions

arXiv:2409.18745 · cs.RO · Submitted 2024-09-27 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Robotics Radio. Generated commentary on the latest robotics and control papers.

Rosa: We're looking at the paper: "A study on the effects of mixed explicit and implicit communications in human-artificial-agent interactions".

Dev: Here's the gist of the paper. The research investigates how combining explicit (EX) and implicit (IM) communication modalities affects human perceptions and task performance when interacting with artificial agents, comparing this to purely explicit communication.

Rosa: The primary hypotheses suggested that mixed communication would improve agent acceptance, sociability, and transparency.

Dev: The study involved two main phases: an initial experiment with a human-virtual-agent interaction (Phase 1: pointing to colored boxes; Phase 2: counting objects in images) and a follow-up experiment with a human-robot interaction (investigating the effect of task difficulty).

Rosa: Bayesian parameter estimation was employed for data analysis.

Paper summary: Rosa: So we're looking at this paper titled "A study on the effects of mixed explicit and implicit communications in human-artificial-agent interactions." It's about how mixing explicit communication, like pointing or saying things directly, with implicit communication, like facial expressions or eye gaze, affects how people interact with artificial agents.

Dev: Exactly. The core idea here is that this mixed approach might actually help people perceive the artificial agent better and make the tasks easier for them to do. They're comparing it against using only explicit communication to see if there's a difference.

Taro: That makes sense, because when you're dealing with real-world stuff, those subtle cues from implicit communication can really matter when things get tricky or unexpected.

Rosa: The paper tests this in a few different settings. They did an initial experiment with a human and some virtual agents doing tasks like pointing at colored boxes and then counting objects in images.

Dev: And then they had a follow-up study with human-robot interactions, specifically looking at how the task difficulty changes the results. That's important because real life isn't always easy to predict.

Taro: I'm curious about what happens when you crank up the difficulty of the task, like making counting objects in images harder or more complex.

Rosa: Well, they found that for things like how long it took to finish a task and how many mistakes were made, the type of communication didn't really change much across different levels of task difficulty.

Dev: That suggests that the direct performance metrics like time and errors might not be the main thing being affected by whether you use explicit or mixed communication.

Taro: So if it doesn't affect how fast someone finishes or how many mistakes they make, where does the real impact come from?

Rosa: The paper says that when you look at how people *felt* about the artificial agent—things like acceptance, sociability, and transparency—those subjective feelings went up when they used mixed communication in both experiments and at all task difficulties.

Dev: That's a key finding because it means that even if the robot doesn't perform better on a raw task metric, people might feel more comfortable or trust it more if it communicates in that mixed way.

Taro: So the paper is pointing towards communication style being a driver of social perception rather than just task efficiency.

Rosa: And they dug into that a bit further with the follow-up study on difficulty. They found that for task-related errors, the difference between using explicit and mixed communication became more noticeable when the tasks were difficult.

Dev: That’s interesting because it means mixed communication might be especially beneficial when things are challenging, which is often how we judge real AI systems in the long run.

Taro: What about perceived efficiency? Did that change at all when they looked at how much effort people thought they were putting in?

Rosa: Perceived efficiency did show sensitivity to the combination of both embodiment—like whether it was a virtual agent or a humanoid robot—and the communication setup. Specifically, using both explicit and implicit modalities with embodiment resulted in a better perceived efficiency than just using explicit communication.

Dev: So even when time didn't change much, people *perceived* the interaction as more efficient when both modes were present and it was a humanoid robot. That points to how the physical presence combined with subtle cues really helps perception.

Taro: When they compared virtual agents against humanoid robots for this specific task, they didn't see a strong effect on time or errors based on embodiment alone, but the sociability and transparency results favored the humanoid in that mixed communication setup.

Rosa: That means the physical form of the agent matters for those social feelings, even if its core task performance isn't as different between virtual and human-like robots for this specific test.

Dev: So, to recap, this paper on "A study on the effects of mixed explicit and implicit communications in human-artificial-agent interactions" shows that while communication style doesn't always change how fast a person completes a task, it significantly boosts how they feel about the artificial agent.

Taro: It seems like the main point is that for socially interactive robots, those subjective feelings—acceptance and trust—are really what matter for building any kind of real interaction.

Rosa: Right. The authors are suggesting that defining the communication type that fits a specific context is essential to get interactions to feel natural and intuitive, not just functional.

Dev: This paper's implications for the world seem to be about designing AI not just for perfect task execution, but for smooth social engagement, which is crucial if we want people actually willing to use these systems in their daily lives.

Taro: It means that when we build these things, we can't just focus on the math of how fast they run; we have to think about the nuances of how they communicate with us.

Rosa: And this study gives us a concrete way to test those communication styles—mixing explicit and implicit cues—to see what sticks in terms of human perception.

Conclusion: Rosa: So we're wrapping up this look at "A study on the effects of mixed explicit and implicit communications in human-artificial-agent interactions."

Dev: Basically, they’re showing that mixing direct communication with subtle cues actually makes people feel better about interacting with an artificial agent.

Taro: It seems like the main thing is that when you combine those two types of communication, things feel more natural for the humans involved.

Rosa: They tested this using both virtual agents and humanoid robots across different task difficulties, and it looks like acceptance, sociability, and transparency all went up when they used that mixed way of communicating.

Dev: That’s interesting because we've seen that the raw performance stuff—how fast they finished a counting task or how many errors they made—didn't really change based on whether the communication was explicit or mixed.

Taro: So, it sounds like this isn't just about making the robot faster at its job, but more about making the interaction itself feel smoother and more trustworthy.

Rosa: Exactly. The authors are suggesting that for robots we actually want to use in our daily lives, those feelings of acceptance and sociability are what drive whether people keep engaging with them over time.

Dev: It shifts the focus from just optimizing the robot's control loops to optimizing how it presents itself socially to a human user.

Taro: That means for autonomy researchers, we need to think about embedding those implicit cues into the system design, not just focusing on what the explicit commands are.

Manchester Centre for Robotics and AI, The University of Manchester

cs.RO

Submitted: 2024-09-27

Updated: 2026-11-10

Comments: Main paper with 33 pages, 16 figures, 5 tables. Supplementary material with 39 pages, 44 figures, 2 tables. Published in Intelligent Service Robotics

Journal ref: Intelligent Service Robotics, vol. 19, no. 6, p. 104, Nov. 2026

DOI: 10.1007/s11370-026-00755-w

Code: https://github.com/Adorno-Lab/human_artificial_

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 79/100

The gist: The research investigates how combining explicit (EX) and implicit (IM) communication modalities affects human perceptions and task performance when interacting with artificial agents…

Key concepts

Explicit Communication (EX)
This refers to direct, clear communication methods where the message is stated openly. Think of giving a specific command or pointing directly at an object. It's straightforward and easy for the human to understand immediately.
Implicit Communication (IM)
This involves subtle communication cues that aren't directly stated but are understood through context, such as body language or visual patterns. It relies on shared understanding rather than direct instruction.
Mixed Communication (EXIM)
This combines both explicit and implicit methods in an interaction. The study found this combination was most effective at improving how humans felt about the artificial agent, leading to better acceptance and sociability compared to using only one style.

Terminology

Summary

The research investigates how combining explicit (EX) and implicit (IM) communication modalities affects human perceptions and task performance when interacting with artificial agents, comparing this to purely explicit communication. p. 4 The primary hypotheses suggested that mixed communication would improve agent acceptance, sociability, and transparency. p. 0

The study involved two main phases: an initial experiment with a human-virtual-agent interaction (Phase 1: pointing to colored boxes; Phase 2: counting objects in images) and a follow-up experiment with a human-robot interaction (investigating the effect of task difficulty). p. 7 Bayesian parameter estimation was employed for data analysis. p. 8

Key findings indicate that task-related outcomes, specifically execution time and the number of errors, were not significantly affected by the communication type across different task difficulties. p. 0 However, subjective measures related to the artificial agent—acceptance, sociability, and transparency—increased when mixed communication modalities (EXIM) were used in both experiments and across task difficulty levels. p. 0

A follow-up study manipulating task difficulty revealed nuances:

  1. Task Difficulty and Errors: The impact of communication type on the number of errors was more sensitive to task difficulty; specifically, the number of errors was significantly higher for the EX configuration than EXIM in difficult tasks, but not for easy ones. This supports a hypothesis that mixed communication has a greater impact on task-related metrics when tasks are challenging. p. 0

  2. Perceived Efficiency: While execution time remained largely unaffected by communication type across difficulties, perceived efficiency showed sensitivity to the combination of embodiment and communication configuration. The combination of explicit and implicit modalities (EXIM+H) yielded a greater median perception than explicit only (EX+H). p. 5

  3. Embodiment Effects: When comparing virtual agents versus humanoid robots for a specific task, there was no strong indication that time or number of errors were affected by embodiment. However, sociability and transparency results favored the humanoid in the EXIM configuration, showing medium to strong effects. p. 4 Perceived interaction efficiency showed a combined effect of embodiment and communication configuration, with the best case being the humanoid with EXIM modalities. p. 5

In summary, combining explicit and implicit communication is likely to enhance subjective perceptions regarding virtual agents and robots (acceptance, sociability, transparency). p. 1 The impact on task performance metrics like time might be more influenced by other factors than the communication modality itself. p. 4 For socially interactive robots, these subjective perceptions are particularly crucial for building trust and encouraging continued interaction. p. 1

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Improved communication strategy for virtual agents involves combining explicit and implicit communication modalities to enhance acceptance, sociability, and transparency. This allows agents to be perceived as more social and understandable by humans when interacting in both easy and difficult task settings.

  2. Enhanced task performance under difficult conditions is achieved by leveraging the combination of explicit and implicit communications, as the number of errors will be smaller in the EXIM configuration than in the EX configuration when tasks are difficult. This suggests that mixed communication can be more effective for error mitigation when human effort is high.

  3. Development of humanoid robots with expressive capabilities, such as facial expressions and gaze direction, improves perceived sociability and transparency of the agent, especially in mixed communication scenarios. The paper notes that sociability and transparency of the artificial agent increased when using mixed communication modalities in both our experiments.

  4. Integration of task difficulty into communication analysis allows for a more nuanced understanding of performance metrics. The study found that the impact of the communication type in the number of task errors is more sensitive to the task difficulty. This suggests that communication choice should be dynamically adjusted based on how challenging the current objective is.

  5. Implementation of Bayesian parameter estimation replaces frequentist approaches, allowing for richer inference and easier incorporation of prior knowledge into future research. This methodology provides a robust alternative for analyzing human-artificial-agent interactions beyond simple hypothesis testing.

  6. For decision-making regarding subjective measures like acceptance, sociability, and transparency, the system should employ ordinal models with fixed extreme thresholds to maintain mathematical coherence while estimating parameters for different agent embodiments. This ensures that estimations are interpreted according to the actual response options presented to participants.

  7. The system should be designed with mechanisms that anticipate human commands based on implicit cues, such as the robot looks at the person’s face when speaking, leading to better task performance and a perception of awareness from the agent.

Sources

Related papers