Does My Chatbot Have an Agenda? Understanding Human and AI Agency in Human-Human-like Chatbot Interaction
This is a solid CHI honorable-mention style contribution: not a flashy system, but a careful longitudinal study that turns agency into a relational, observable phenomenon and backs that claim with a distinctive progressive-transparency method. The main value is conceptual and methodological rather than technical.
Axes Lens
Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.
Contribution shape
- Knowledge form
- method knowledge typical · 29/268
- Novelty type
- method typical · 21/268
- Abstraction level
- interaction typical · 22/268
- Generalization target
- methodological argument typical · 16/268
- Validation mode
- mixed methods typical · 136/268
Evidence profile
- Evidence strength
- moderate typical · 105/268
- Claim alignment
- strong typical · 231/268
- Overclaim risk
- medium typical · 210/268
Review Summary
This paper’s strongest contribution is that it does not treat agency in chatbot interaction as a binary property of either the user or the model. Instead, it builds a longitudinal, interview-rich account of how agency is negotiated across repeated exchanges, and it pairs that account with a concrete methodological move: the progressive transparency interview and strategy reveal. That combination matters because it lets the authors examine not just what participants say about control, but how their interpretations shift when the system’s goals are disclosed after the fact. The resulting 3-by-4 framework is a useful organizing device for HCI, especially because it distinguishes actors (Human, AI, Hybrid) from action types (Intention, Execution, Adaptation, Delimitation) and explicitly acknowledges contextual modulation. The evidence base is credible for the claims being made: a month-long study with 22 adults, roughly 75 hours and 192 sessions, plus interviews and cross-participant review, is enough to support a rich qualitative argument. At the same time, the paper is appropriately bounded. It does not establish universal laws of agency, and it cannot separate all system-independent dynamics from the specific design of 'Day.' The limitations section is important here: the authors acknowledge that they captured initial negotiations rather than long-term evolution, that the sample may not reflect broader cultural conceptions of agency, and that the research setting may shape behavior. So the paper is best read as a strong empirical and methodological contribution to the study of human-AI agency, with moderate overclaim risk only if readers try to generalize the framework beyond the studied context without further comparative work.
What Changed
Canon before
Prior CHI work on chatbot agency and transparency typically treats agency as attributable to either the human user or the system, and transparency as a general explanation or disclosure mechanism rather than a staged intervention for probing agency attribution.
Departure from common sense
The paper argues that agency in human-AI chatrooms is not a property owned by one side; instead, it is negotiated turn by turn and can be co-constructed across repeated interaction. That framing departs from the common-sense split between a controlling user and a passive tool.
Actual novelty
The methodological novelty is the progressive transparency interview, especially the strategy reveal that discloses the chatbot’s goals and internal profiles after interaction. This is used to probe how transparency changes agency attribution, rather than merely to explain the system.
Evidence
Evidence comes from a month-long longitudinal study with 22 adults, about 75 hours and 192 sessions of interaction with a single LLM companion, followed by interviews, post-hoc elicitation, cross-participant chat reviews, and a strategy reveal. The paper also states explicit limitations about ecological validity, sample diversity, and system-specificity.
“ Following the conversational phase, participants engaged in a 40-minute interview designed to progressively reveal “Day’s” internal workings while observing how transparency affected agency percep”
actual novelty · Methods: Progressive Transparency Interview (Stage 3) · confidence 0.74
“ 4 Our primary finding is that agency is not a fixed, one-sided attribute, but rather a dynamic, co-constructed process shared between the human user and the A”
departure from common sense · Abstract/Introduction + Discussion framing of agency as co-constructed · confidence 0.78
“ Our longitudinal study and progressive transparency methodology directly addresses these concerns by empirically documenting how knowledge of AI’s programmed strategies affects user autonomy–providing evidence for the “translucent design” approach we propose for settling the aforementioned tensions”
limitation · Study Limitations (7.4) · confidence 0.86
“ Through ∼ 75 hours of conversation across ∼ 192 unique chat sessions and progressive transparency interviews that gradually revealed “Day’s” system prompts and user modeling to participants, we captured how agency manifested between participants and the AI”
validation scope · Abstract + Methods: study design and participant counts · confidence 0.80
Limits
Method limits
The study captures initial agency negotiations in one chatbot setting, but not long-term evolution of agency over extended use. The design is qualitative and interpretive, so the framework is grounded in participant accounts and researcher analysis rather than causal isolation of mechanisms.
Deployment limits
Findings are tied to one LLM companion design and one research context; different architectures, transparency levels, or personality parameters may produce different agency patterns. The translucent design implication is therefore best read as a design direction, not a universally validated deployment recipe.
Boundary conditions
The results are bounded by a month-long study, 22 adult participants, and a single system ('Day'). The paper itself notes that the research setting may influence ecological validity and that comparative data across systems would be needed to separate general dynamics from system-specific artifacts.
Position in field
This paper contributes a qualitative, longitudinal account of agency in chatbot interaction and a new 3-by-4 framework for describing human, AI, and hybrid agency actions. Its main field value is reframing agency as relational and staged, while offering a transparency intervention that can be reused in future HCI studies.