Shaping AI's Beliefs Before It Generates: Human Sense of Agency in Creative AI
Organizations: Google DeepMind
Abstract
AI systems for creative media generation expand human capabilities yet risk diminishing human agency. A core problem is anchoring, where the user's thinking becomes anchored to a polished, generated artifact, shifting the user's role from active authoring to reactive editing. We propose pre-generation belief controls that can empower creators to inspect, revise, and steer the AI's interpretation before any media is generated, so that the initial anchor is set by the human. We formalize four design principles and instantiate them in the Proactive Co-Creator (PaCC), a system for early-stage conceptual ideation and prototyping that proactively surfaces its assumptions and uncertainty through clarifications, belief graphs, and media attributes, inviting users to shape the AI's beliefs before generating the final outcome. Using PaCC as a design probe in a small-scale, single-arm exploratory user study, we observed that participants reported directionally positive sense of agency and process control relative to their familiar prompt-based AI tools, noting that surfaced assumptions helped them think through details they had overlooked, whereas their feelings about ownership and authorship diverged
Figures & tables
| Agency | Ownership | Control | Authorship | |||||
|---|---|---|---|---|---|---|---|---|
| Feature | G/E/L | G/E/L | G/E/L | G/E/L | ||||
| PaCC (overall) | 5/2/0 | 4/2/1 | 6/2/0 | 3/3/1 | ||||
| Clarifying Questions | 8/0/0 | 6/0/1 | 7/1/0 | 5/1/1 | ||||
| Belief Graph | 4/0/2 | 4/1/2 | 5/1/1 | 3/2/2 | ||||
| Attributes | 4/2/0 | 3/3/1 | 4/3/0 | 3/3/1 | ||||
| Prompt Refinement | 6/1/0 | 6/0/1 | 6/2/0 | 5/1/1 | ||||
Appendix figures & tables18 assets
Supplementary material from the paper’s appendix.
Appendix
| Feature | Mechanism | DP1 | DP2 | DP3 | DP4 |
| Pre-generation Control Features | |||||
| Clarifying Questions | Targeted Q&A with selectable options | ✓ | ✓ | ✓ | ✓ |
| Belief Graph | Editable entity–relationship graph | ✓ | ✓ | ✓ | ✓ |
| Attributes | Scene-level parameters with candidates | ✓ | ✓ | ✓ | ✓ |
| Prompt Refinement Capabilities | |||||
| Refined Prompt Review | Original vs. editable refined prompt | ✓ | ✓ | ||
| ID | Gender | Occupation | Professional Role | AI Exp. | Self-Reported Definition of Agency |
|---|---|---|---|---|---|
| P1 | M | Research Engineer | AI Professional | ✓✓✓✓ | The ability to influence the final output of a project or effort I’m working on. |
| P2 | M | Research Scientist | AI Professional | ✓✓✓✓ | How effectively I can communicate my intent to the AI and have it reliably follow my instructions. |
| P3 | M | Student Researcher | AI Professional | ✓ | I feel that I am the initiator of my actions. |
| P4 | M | UX Researcher | Other Professional | ✓✓✓✓ | Control in the creation process. |
| P5 | F | Content Designer | Creative Professional | ✓✓ | Feeling like I can take some credit for the output, that the output was based enough on my decisions to feel like mine. |
| P6 | F | UX Researcher | AI Professional | ✓ | Having informed choices to make. |
| Aspect | Agency | Ownership | Control | Authorship |
|---|---|---|---|---|
| Intent | ||||
| Expression | ||||
| Process | ||||
| Outcome |
| Aspect | Agency | Ownership | Control | Authorship |
|---|---|---|---|---|
| Intent | ||||
| Expression | ||||
| Process | ||||
| Outcome |
| Aspect | Agency | Ownership | Control | Authorship |
|---|---|---|---|---|
| Intent | ||||
| Expression | ||||
| Process | ||||
| Outcome |
| Aspect | Agency | Ownership | Control | Authorship |
|---|---|---|---|---|
| Intent | ||||
| Expression | ||||
| Process | ||||
| Outcome |
| Statement | Always | Very often | Sometimes | Rarely | Never |
|---|---|---|---|---|---|
| AI tools preserve my agency | 0 | 2 | 5 | 1 | 0 |
| AI tools augment my agency | 0 | 2 | 4 | 2 | 0 |
| AI tools diminish my agency | 0 | 1 | 7 | 0 | 0 |
| Feature | Dimension | Mean | SD | Range | |
|---|---|---|---|---|---|
| PaCC (overall) | Agency | 4.09 | 0.84 | 2.5–5.0 | 8 |
| PaCC (overall) | Ownership | 3.22 | 1.49 | 1.2–5.0 | 8 |
| PaCC (overall) | Control | 3.89 | 0.69 | 3.0–5.0 | 7 |
| PaCC (overall) | Authorship | 2.75 | 1.13 | 1.5–5.0 | 7 |
| Clarifying Questions | Agency | 3.66 | 0.64 | 2.8–4.5 | 8 |
| Clarifying Questions | Ownership | 3.34 | 0.77 | 2.0–4.5 | 8 |
| ID | Most Useful | Reason | Least Useful | Reason |
|---|---|---|---|---|
| P1 | Belief Graph | “It gives me confidence that the complex visual concepts in the scene are exactly how I picture them.” | — | “They are all pretty good.” |
| P2 | Belief Graph | — | Attributes | — |
| P3 | Clar. Questions | “Knowing where the model sees the ambiguity” | Clar. Questions | “The questions focused on the art style, which is not something I was focused on” |
| P4 | Clar. Questions | — | Attributes | — |
| P5 | Clar. Questions | “They elucidate the possible and open things I hadn’t been thinking of” | Belief Graph | “wasn’t particularly useful in my explorations” |
| P6 | Clar. Questions | “They were relevant and expansive” | Belief Graph | “didn’t see the need for the belief graph” |
| Feature | Agency | Ownership | Control | Authorship |
|---|---|---|---|---|
| PaCC (overall) | ||||
| Clarifying Questions | ||||
| Belief Graph | ||||
| Attributes | ||||
| Prompt Refinement |
| Agency | Ownership | Control | Authorship | |
|---|---|---|---|---|
| Agency | — | |||
| Ownership | — | |||
| Control | — | |||
| Authorship | — |
| pairs | Pearson | Spearman | |
|---|---|---|---|
| Agency Ownership | 152 | ||
| Agency Control | 148 | ||
| Agency Authorship | 132 | ||
| Ownership Control | 148 | ||
| Ownership Authorship | 132 | ||
| Control Authorship | 128 |
Human-AI Co-Creation
Human-AI co-creation studies how people and AI jointly develop ideas, artifacts, and decisions while preserving meaningful human agency.
See all 164 papers