What Do We Need for an Agentic Society? DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY Woodstock ’18: ACM Symposium on Neural Gaze Detection, June 03–05, 2018, Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06 CCS: Human-centered computing Ubiquitous and mobile computing CCS: Human-centered computing Ubiquitous and mobile computing theory, concepts and paradigms CCS: Computing methodologies Multi-agent systems
Kwon Ko, Hyoungwook Jin · 2026
Read What Do We Need for an Agentic Society? DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY Woodstock ’18: ACM Symposium on Neural Gaze Detection, June 03–05, 2018, Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06 CCS: Human-centered computing Ubiquitous and mobile computing CCS: Human-centered computing Ubiquitous and mobile computing theory, concepts and paradigms CCS: Computing methodologies Multi-agent systems by Kwon Ko, Hyoungwook Jin (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-03938.txt.
Introduction
Kwon Ko and Hyoungwook Jin asked what an agentic society of everyday objects would need to coordinate well. Thirty years after Wooldridge and Jennings defined agents by autonomy, reactivity, pro-activeness, and social ability, smart objects can now act as agents together. A scenario of a bullied teenager showed both promise and failure modes: false positives that break trust, deadlocks that block action, and adversarial corruption. The open questions split into sharing choices, judgment calls, and timing of action.
Catalog record
| CREATOR | Kwon Ko, Hyoungwook Jin |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 19612 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.03938
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
What Do We Need for an Agentic Society? DOI: XXXXXXX.XXXXXXX Conference: Make sure to enter the correct conference title from your rights confirmation emai; June 03–05, 2018; Woodstock, NY Woodstock ’18: ACM Symposium on Neural Gaze Detection, June 03–05, 2018, Woodstock, NY ISBN: 978-1-4503-XXXX-X/18/06 CCS: Human-centered computing Ubiquitous and mobile computing CCS: Human-centered computing Ubiquitous and mobile computing theory, concepts and paradigms CCS: Computing methodologies Multi-agent systems Kwon Ko, Hyoungwook Jin Abstract . Thirty years ago, Wooldridge and Jennings defined intelligent agents through four properties: autonomy, reactivity, pro-activeness, and social ability. Today, advances in AI can empower everyday objects to become such intelligent agents. We call such objects agentic objects and envision that they can form an agentic society : a collective agentic environment that perceives patterns, makes judgments, and takes actions that no single object could achieve alone. However, individual capability does not guarantee coordination. Through an illustrative scenario of a teenager experiencing bullying and depression, we demonstrate both the promise of coordination and its failure modes: false positives that destroy trust, deadlocks that prevent action, and adversarial corruption that poisons judgment. These failures reveal open questions spanning three phases: what to share, how to judge, and when to act. These questions chart a research agenda for building agentic societies. 1. Introduction Thirty years ago, Wooldridge and Jennings defined what makes a software system an intelligent agent: autonomy, reactivity, pro-activeness, and social ability ( 10 ) . Since then, agent research has progressed from modeling how individual agents reason internally (what they believe, what they want, and what they commit to doing), through coordinating multiple agents toward shared goals, to today’s LLM-powered agents that plan, use tools, and collaborate ( 9 ) . These capabilities are no longer confined to software. Technology is increasingly designed into everyday objects rather than housed in separate devices. Styluses that sense pressure and tilt, speakers that understand spoken language, and rings that track biometrics all blur the boundary between object and computer. As the field matures, a much wider range of physical objects can satisfy the same criteria. A bed can detect disrupted sleep patterns and judge whether they signal a health concern (autonomy). A lamp can sense when someone is tired and adjust its light (reactivity). A phone can predict emotional decline from usage patterns and suggest a break (pro-activeness). And these objects can work together: a bed might tell a phone to delay morning alarms after a restless night (social ability). In this work, we refer to such objects as agentic objects to distinguish them from software-based agents: they are embodied, spatially distributed across physical environments, and continuously co-present with users. Since modern life fragments a person across contexts (home, school, work, online), cross-context patterns are hard to recognize. Those most affected are the ones who cannot advocate for themselves: young people who lack the language or power to ask for help, elderly individuals living alone, and people with cognitive or communicative disabilities. A parent sees the child at home; a teacher sees the student at school. For example, a teenager experiencing bullying at school may appear merely tired at home, and a parent attributing withdrawal to “being a teenager” lacks the cross-context information to recognize a pattern. Agentic objects can collectively capture the distributed contexts. A school desk observes classroom engagement; a bedroom lamp observes evening behavior; a phone bridges everything. If coordinated, these objects could collectively perceive the pattern, judge its severity, and act before anyone asks. We call this an agentic society : distributed everyday objects coordinating for collective perception, judgment, and action. The foundations for this vision already exist, across three fronts: how devices relate to one another, how agents produce collective behavior, and how intelligence can be embedded in individual objects. On the first, Jung et al. mapped how a person’s artifacts form interconnected relationships, each shaping how others are used, though the coordination remains in the hands of the user ( 3 ) . On the second, Park et al.’s generative agents showed what happens when agents begin to coordinate on their own: they spontaneously organized activities, formed relationships, and coordinated a party without being explicitly programmed to do so ( 7 ) . On the third, work on thoughtful things has demonstrated that even lightweight, on-device language models can give a single object the capacity to interpret user goals and explain its own behavior ( 5 ) , while systems like Sasha ( 4 ) and SAGE ( 8 ) coordinate multiple devices through a central LLM within one environment. Each of these advances brings the vision of an agentic society closer to reality, but none of them addresses coordination across physically and contextually separate environments, where no single agent has access to the full picture. We claim that Wooldridge and Jennings’s four agent properties are necessary but not sufficient for collective functioning: objects can be autonomous, reactive, pro-active, and social yet still fail to function as a collective ( Figure 1 ). We propose three coordination phases: perception (what should be shared, and across what boundaries), judgment (how should observations combine, and conflicts resolve), and action (when should the collective intervene, and who is accountable). We develop the scenario of Peter, a fourteen-year-old being bullied who has become increasingly depressed, although he has told no one. Through this scenario, we surface three failure modes (false positives, deadlock, and adversarial corruption) and derive nine open questions for reaching an agentic society. desk tray lamp bed phone Agentic Objects autonomy · reactivity · pro-activeness · social ? Open Questions desk tray lamp bed phone Agentic Society perception · judgment · action Figure 1. The gap between agentic objects and agentic society. Individual objects may possess all four agent properties yet remain uncoordinated (left). Forming a society requires addressing coordination challenges across perception, judgment, and action (right). 2. Scenario: Peter We return to the teenager from the introduction. Peter’s story is not a prediction but a probe—a way to surface concrete failures in coordination: cases where agentic objects exist but fail to form a society. Peter is fourteen. Over the past month, he has grown quieter at school, sleeps poorly, and eats less, but he has not told anyone why. His parents notice the change yet chalk it up to adolescence. Five agentic objects span his environments: a desk and cafeteria tray at school, a bedroom lamp and bed at home, and a phone that bridges everything. Each possesses autonomy, reactivity, pro-activeness, and social ability. Each can observe, judge, and communicate. The question is how they can form a society—and what happens when they try. The failures that follow are not arguments against agentic societies but conditions any viable design must address. Success Case. It begins with the cafeteria tray. Three days running, Peter returns it nearly full—food weight barely changed from serving. The tray forms a hypothesis—something is wrong—but cannot determine what, so it queries the desk: “I am seeing missed meals. Are you noticing anything?” The desk reports that Peter’s typing has slowed, he no longer leans into group discussions, and his surface temperature runs cold—signs of withdrawn posture. Two objects, two contexts, one converging signal. They jointly query the phone, which reports outgoing