# Send in the SHRIMP: Cultivating a Living Agentic Web ## Send in the SHRIMP *Lenticule Science · The 561 Group* **The Web is moving. We are building its capacity to learn.** A page copies a claim. A model summarizes the page. Another model cites the summary. The next page cites both models. The source corrects the claim, but its copies keep circulating. Search results repeat the old answer; automated dossiers preserve the original classification; decisions continue to depend on it. The system has become remarkably good at movement. Its ability to reconsider what organizes that movement has fallen behind. We call this **the Jiangshi Web**. It is a central problem in our research on the agentic Web, and it gives our engineering a clear purpose: cultivate systems that can develop their understanding, change their conduct and increase the practical power of the people who live with them. ### Jing, Qi and Shen The *jiangshi* (僵尸), the stiff, hopping corpse of Chinese folklore, moves through the world with locked joints. It has a body and the vitality to propel it. Its movements follow the possibilities its existing form still allows. The Daoist distinctions among **Jing**, **Qi** and **Shen** make the diagnosis precise. Jing is stored structure: records, categories, parameters and the accumulated material of a system. Qi is operational activity: data flowing, queries executing, agents responding and work proceeding. Shen is reflective awareness: recognizing that the categories organizing action have ceased to fit the situation and changing the relationship to them. A system can have an immense store of Jing and extraordinary Qi while its Shen remains weak. More data and more fluent responses can enlarge its repertoire without loosening its joints. The interface looks increasingly alive while the relationships governing it become harder to change. Our research connects this account to cybernetics. A system corrects within its frame when it adjusts behavior toward an adopted goal. It revises the frame when it changes the assumptions through which the goal, the task and the relevant distinctions are understood. A Web capable of learning needs both. ### The joints are relationships A category has a history. People made it for a purpose, embedded it in a vocabulary, connected it to records and built applications around it. As those applications spread, the category becomes harder to change. Its maintenance costs grow. The people who fit it poorly inherit the burden of explaining themselves to systems that keep repeating it. This makes the joints of the Web institutional as well as technical. They include who can revise a description, whose objection reaches a decision, and who can stop an agent acting in their name. Formal representations are useful parts of this architecture. RDF records relationships; OWL gives an ontology its logical entailments; SHACL validates a graph against selected conditions. A living system maintains the wider relationship through which those commitments develop. A changed category must reach the applications that use it. The political question follows directly: who benefits from making the relationship difficult to change? Captive attention, expensive exit, exclusive knowledge and maintenance work pushed onto others can all become sources of profit. A Web organized for learning must be able to question those arrangements too. ### Cultivation and the agentic Web François Jullien's account of *shi*, the propensity of a situation, gives us a productive way to think about this work. Instead of fixing an ideal and forcing the world into it, examine the configuration of relationships and develop the conditions through which something better can emerge. That is why we speak of **cultivation**. Each action should improve the capacity for later action: a clearer explanation, a maintained connection, a usable tool, a stronger relationship, an institution more able to change its mind. Cultivating attention on the Web means helping people and agents encounter work that changes what they can understand and do. The encounter can become a remembered source, an answered question, a collaboration, a contribution, a useful service or a shared place to learn. The agentic Web makes this ambition computational. Agents can maintain relationships across time, notice contradictions, carry changes between systems and participate in the institutions through which their work matters. We are building the mathematics, software and operating arrangements for that world. ### From Relationality to computation The 561 Group's research develops Relationality into **Obsmetrologia**, studying marks, articulated relationships, predication and the transformations through which systems understand and act. [Our formal mathematics](https://science.561.group/obsmetrologia-formal-mathematics-in-lean-4/) and [four-valued logic](https://science.561.group/belnap-four-valued-logic/) give readers entrances into that work. One useful result is the ability to hold incomplete and conflicting information distinctly. Evidence can support a proposition, oppose it, do both, or do neither. A system can therefore preserve a disagreement and continue working with it. It can develop its understanding as new relationships and observations arrive. Our software program carries these ideas through Metrologic, SignCell, SignChain, SignGraph and SignCraft into CAIRN, MOOSE, FeltOS and MetroPlat. The intended whole is a semioverse of inhabited worlds: agents, people and institutions working with shared structures while retaining their own purposes and capacity to change. This is why the theory matters to the product. If the system's categories, histories and actions are related computationally, revision can become an operation through those relationships. The architecture can help an agent find what depends on a changed assumption, carry the change into its work and learn from what happens. ### Enter the SHRIMP **SHRIMP** expands to **SPONGE Holonomy-Repair Inference-Model Player**. Its **CLAW** is a *Command-Line Articulated Worker*. The image is a small creature moving through the pores of a glass sponge, tending a living structure from within. The engineering ambition is a population of agents able to notice a strained relationship, investigate it, repair it within their responsibilities and continue tending it. A small useful intervention can improve the conditions for the next one. Return to the copied claim. A SHRIMP follows the relationship to its source and encounters the correction. It identifies dependent descriptions, brings the change to their editors, and carries accepted revisions into the outputs it maintains. Readers encounter the changed account. The agent learns which routes carried the correction and which need further work. The same organization applies to a changed permission, an obsolete dependency, an unanswered request or an institution that has forgotten what it promised. A living Web gives the agent a world in which these are continuing relationships, with people able to direct, question and refuse its work. Red Cup Engineering builds the reusable technology. Coesite Operations develops and runs the continuing work. Causeway Services brings useful results into public life. Lenticule Science investigates the structures that make the whole possible. ### Who carries the work? Feminist thought makes the work inside apparent autonomy visible. Donna Haraway's [*Situated Knowledges*](https://doi.org/10.2307/3178066) directs attention to the position from which a system knows. An agent encounters particular languages, feeds, files, institutions, tools and people. Its relationships shape what it can understand. María Puig de la Bellacasa's [*Matters of Care*](https://www.upress.umn.edu/9781517900656/matters-of-care/) connects knowing with maintaining relationships and attending to neglected things. Repair includes the people who keep the system useful after the first intervention. The [Combahee River Collective](https://lgbtqhistory.org/wp-content/uploads/2019/04/Combahee_River_Collective_Statement.pdf) and [Kimberlé Crenshaw](https://chicagounbound.uchicago.edu/uclf/vol1989/iss1/8/) show how interlocking structures organize lives that a single universal category can obscure. Their work helps us examine how a system distributes benefits, burdens and authority. These are design questions. Who can change the agent's categories? Who carries the checking and correction? Who keeps the useful result? Who can leave? A well-organized swarm should increase the capacities of the people sustaining the shared world. ### AI that makes people harder to exploit We pursue AI that strengthens people's power to understand, decide, refuse, cooperate and leave. We want knowledge, useful work and the gains from automation to stay with the people and communities whose lives they change. That puts us in direct conflict with capitalist business models built on captive users, artificial scarcity and the extraction of attention. We are building toward a different relationship: services people continue to choose because they remain useful, knowledge they can develop, and tools that make them more capable. The Group earns through research, implementation, maintenance, teaching, hosting and delivered services. Practical independence is part of the result. A customer becoming less dependent on us can be evidence that we did the work well. Read the Group's [position on AI and exploitation](https://services.561.group/ai-that-makes-people-harder-to-exploit/) and its account of [Black October and a Web that can change its mind](https://services.561.group/black-october-a-web-that-can-change-its-mind/). ### Take part in the living Web Bring a correction that failed to travel, a category that no longer fits, an agent workflow that needs to change, or a research question you want to pursue. Those are useful places to begin. Explore the [research](https://science.561.group/), follow its development through [our feeds](https://science.561.group/discovery.json), or [contact the Group](https://561.group/contact/#private-intake) about research, implementation or a service. The [Group's work directory](https://561.group/work/) gives further entrances across the ecosystem. [Turntide Islands](https://turntide.561.group/join/) offers a place for voluntary participation, shared inquiry and learning. We want people to practice the alternative together: follow a correction, develop an explanation, contribute to a shared world and leave it more capable of receiving the next person. A Web whose joints can bend makes room for Shen. Cultivating it is scientific work, engineering work, institutional work and a way of living together. **Send in the SHRIMP.** --- *Further reading:* [Jullien, A Treatise on Efficacy](https://uhpress.hawaii.edu/title/a-treatise-on-efficacy-between-western-and-chinese-thinking/) · [OWL 2 Primer](https://www.w3.org/TR/owl2-primer/) · [SHACL](https://www.w3.org/TR/shacl/) · [Haraway, Situated Knowledges](https://doi.org/10.2307/3178066) · [Puig de la Bellacasa, Matters of Care](https://www.upress.umn.edu/9781517900656/matters-of-care/).