The previous post introduced the distinction between planning and design, and the formal apparatus that planning presupposes. In PDDL - the Planning Domain Definition Language used to specify problems for AI planners - the first step is to define a domain: the types of things that can exist, the predicates that describe their properties, and the actions that can change them. Only then can you specify a problem: the particular objects, the initial state, and the goal.
The word "objects" is doing significant work here. The planning formalism requires specifying what objects exist before anything else can follow. States are configurations of objects and their properties. Actions operate on objects. Goals are desired configurations of objects. Without objects, there is no state space to navigate and nothing for events to act upon.
This post asks what seems like a simple question: what counts as an "object" in the context of public services? The answer turns out to be less straightforward than it appears. Different philosophical and theoretical traditions define objects in substantially different ways, and those differences have consequences for how one constructs the kind of state-space representations this series is concerned with.
The Ordinary Sense, and Its Limits
In everyday language, "objects" are physical things - chairs, documents, patient records, buildings. ISO 704:2022, the international standard for terminology work, offers a broader definition: an object is "anything perceivable or conceivable". This encompasses material objects (an engine, a medical device), abstract objects (a diagnosis, a legal status, a classification code), and immaterial objects (a right, an obligation, a relationship).
For public services, the objects of interest span all three categories. A vocational rehabilitation service engages with bodies (material), diagnoses and work-capacity assessments (abstract), and entitlements, obligations, and coordination agreements (immaterial). The Pathway Generator that prompted this doctoral research treats each patient as an object defined by a vector of variables - depression scores, employment history, functional capacity, self-efficacy measures. The BIP assessment framework used by SCÖ treats work readiness as an object with five broad dimensions. The ICF treats functioning as a structured matrix of codes.
Each of these is a decision about what kind of object the person is, for the purposes of the system. And each makes different things computable at the cost of making other things invisible. This is the object question in its most consequential form: not "what objects exist?" in the abstract, but "what objects must a state-space representation contain, and what properties must they have, for the representation to be useful?"
Objects in Conceptual Modelling
The information systems tradition treats objects as formal entities that can be modelled, structured, and computed upon. Two frameworks are particularly relevant.
John Mylopoulos's foundational work on conceptual modelling (1992) distinguishes three types of objects in his Requirements Modelling Language: "entities, activities and assertions, all of which have attributes that relate them to other objects". The later goal-oriented requirements work with Chung and Yu extends this to include "goals, agents, alternatives, events, actions, existence modalities and agent dependencies". These are pragmatic categories - objects are whatever you need to represent to build a system. The emphasis is on making the implicit explicit and the tacit computable.
Object-Process Methodology (OPM), described in the Handbook of Conceptual Modeling (Embley and Thalheim, 2012), makes a more direct connection to the ontological distinction this series is concerned with. In OPM, "objects and processes, collectively referred to as OPM things, are the two types of OPM's universal building blocks". Objects are the static entities that persist; processes are the dynamic transformations that change them. This directly parallels the objects-and-events distinction that the next post will develop through cognitive science and life-course theory - but formalises it for system design.
There is a useful complication here. Johnson and Henderson (2011) observe that in conceptual models, "events (an object) presumably have attributes (e.g., Name, Description, Start Time, End Time, Date)". An appointment, a referral, a transition - these are events when they are happening, but objects when they are being tracked, recorded, and reasoned about. This duality matters for service systems, which need to represent both the dynamic unfolding of events and their administrative reification as records. A referral is an event (something that happens) and an object (something that is stored, queried, audited). The same entity shifts ontological category depending on whether you are experiencing it or managing it. I examined the implications of this for how different stakeholders model the same domain - and the conflicts that arise when their models differ - in the post on what concepts are.
Objects in Cognitive Science
Peter Gärdenfors, whose conceptual spaces theory provides a geometric foundation for meaning (and who will appear frequently in this series), argues that objects are cognitively primary. "The ontologically (and developmentally) primary references of nouns are object categories" (Gärdenfors, 2017). Infants perceive objects before they perceive events; nouns are typically acquired before verbs.
Gärdenfors and Zenker (2015) distinguish "proto-object from proto-action/event, according to the integral dimension of dynamics". A proto-object is a cognitive representation that lacks the dynamic dimension - it is a thing that exists at a point in time, with properties but without the temporal unfolding that characterises events. This distinction is not merely philosophical; it reflects how the brain processes experience. Objects and events are different kinds of cognitive structures.
In the conceptual spaces framework, objects are positions in multidimensional quality spaces. A chair occupies a region in a space whose dimensions include shape, material, size, and function. A patient in vocational rehabilitation occupies a region in a space whose dimensions might include health status, employment history, functional capacity, and psychological wellbeing - this is essentially what the Pathway Generator's patient vector encodes, although in some cases using discrete variables rather than continuous dimensions.
Gärdenfors further argues that "a natural property is a convex region of a domain in a conceptual space" - meaning that natural categories have coherent, continuous structure. If two things are both instances of "depression", then anything perceptually between them is also an instance. This matters for rehabilitation assessment, where the categories used to classify patients (mild/moderate/severe depression, low/medium/high work capacity) impose boundaries on what is, in cognitive reality, continuous variation. The next post but one on conceptual spaces develops this geometric framework more fully. The point here is that objects, in the cognitive science account, are not arbitrary collections of properties but structured regions in spaces defined by quality dimensions - some innate, some learned, some culturally dependent.
Object Identity and Persistence
What makes an object the "same" object across time? This is not an abstract philosophical concern - it is central to how services function. A patient referred to rehabilitation in January and assessed again in June must be recognisable as the same person for the service to work. But what has persisted? Their body has changed, their psychological state has changed, their employment situation may have changed entirely. What makes them "the same patient"?
Franssen, Lokhorst, and van de Poel (2013) argue that "the sortal persistence conditions of an individual object clearly depend upon its kind". Different types of object persist according to different criteria. A physical object persists through material continuity. A person persists through (roughly) bodily and psychological continuity. A legal entity persists through institutional recognition. A case record persists through administrative convention - a database entry with a unique identifier.
This matters for state-space representations because they necessarily presuppose a persistence criterion. The Pathway Generator tracks a patient across time through a vector of variables. But the identity of the tracked entity is not given by the variables - it is presupposed by the system. A unique patient identifier links records across time, but what that identifier refers to - a biological organism, a social role, a legal person, an administrative case - is a design decision, usually an implicit one. The ICF classification, the BIP assessment, and the JANUS variables each assume a particular account of what persists across the events that change a patient's situation, even if they do not make that assumption explicit.
There is a deeper issue here. Whitehead, whose process philosophy inverts the usual ontological priority, argues that "actual entities are not events" but that both are constituted through processes of becoming. Objects, in this view, are not static substrates that events act upon; they are themselves the outcomes of processes that have temporarily stabilised. A patient's "work capacity" is not a fixed property of the person but a temporarily stable configuration of biological, psychological, and social processes. This perspective is philosophically demanding but practically relevant: it helps explain why the same assessment, administered six months apart, can produce different results even when "nothing has changed" - because what we take to be a stable object is, at a finer resolution, a process.
Relational Accounts: Affordances and Networks
Not all traditions treat objects as self-contained entities with intrinsic properties. Several relational perspectives offer different accounts of what objects are by focusing on what they do and what they connect to.
James Gibson's ecological psychology defines objects through what they afford - the action possibilities they offer in relation to an organism or user. Klaus Krippendorff (2005), whose trajectory of artificiality I have been reflecting upon separately, summarises the position: "everyday objects are perceived and conceptualized in terms of affordances: cups by our ability to hold and drink from them, knobs by their graspability". Jenny Davis (2020) extends the analysis, emphasising that affordances are "action-based, dynamic, and necessarily relational" - they belong neither to the object alone nor to the user alone, but to the relationship between them.
For service design, this reframes what counts as an object's properties. An assessment instrument is not merely a form with fields; it is something that affords structured disclosure, that affords classification, that affords (or constrains) the kinds of conversation caseworker and patient can have. A digital platform affords certain workflows and resists others. The materiality of service objects - their affordances - shapes what events are possible. An object's affordances are not intrinsic but depend on the conditions under which it is encountered.
Actor-Network Theory (Latour, Callon, Law) goes further. Objects are not merely relational in their affordances; they are constituted through their participation in networks. A patient record is not a self-contained object but a node in a network that includes clinicians, administrators, software systems, regulatory frameworks, and the patient themselves. Its properties emerge from its relations. Remove the record from the network - from the governance framework that authorises it, the software that renders it, the clinical protocols that structure it - and it ceases to be what it was. Latour's "interobjectivity" treats objects as active participants in social networks, not passive substrates for human action.
This is useful for understanding how service objects function in practice, but it raises a question about formalisation. If objects are constituted by their relations, then any representation must capture relational structure, not just attribute values. Attribute-vector models - where an object is a tuple of properties - abstract away the relational context. But graph-based representations take relations as primary: a node is defined by its edges, and the structure of connections carries as much meaning as the properties of individual nodes. Russell and Norvig (2021) make this explicit: state spaces can be represented as graphs in which vertices are states and directed edges are actions. The question for this series is not whether relational ontology can be formalised, but which formalisms preserve relational structure and which flatten it - and what the consequences are for service design's representational tools. I explore this more fully in a later post on graphs and service representations.
Graham Harman's Object-Oriented Ontology (OOO) offers a direct critique of the fully relational position. Harman argues that objects always withdraw from their relations - there is always more to an object than what it reveals in any particular encounter. He defines an object as "any unified entity, whether it has reality in the world or only in the mind", and insists that "all objects are equally objects, in that no object can be reduced to anything else".
Harman distinguishes two kinds of reduction: "undermining" (explaining objects by dissolving them into smaller components) and "overmining" (explaining objects by dissolving them into their effects or social constructions). Both, he argues, fail to take objects seriously as objects. A hospital bed is not just its atoms, nor just its function in healthcare delivery, nor just its meaning to patients and staff. It has a reality that exceeds all these characterisations.
For the purposes of this series, OOO provides a useful corrective rather than a foundation. The insight that any formalisation - any patient vector, any classification scheme, any conceptual model - is necessarily partial is important precisely because the series is about constructing such formalisations. The partiality is not a failure to be overcome; it is a condition of the enterprise. Any state-space representation captures some of what an object is and leaves the rest in withdrawal. The question is whether what it captures is sufficient for the purpose at hand.
Agents as Objects: Promise Theory
Mark Burgess's Promise Theory introduces, as did the Linköping introduction to Artificial Intelligence course, a distinctive kind of object: the autonomous agent. In Promise Theory, "promise theory is about what can happen in a collection of autonomous agents that interact through the promises they make to one another" (Burgess, 2020). An agent is any entity capable of making and keeping (or failing to keep) promises - a person, an organisation, a software service, a device.
This reframes the object question for services. The fundamental objects are not passive entities with properties waiting to be changed by external events, but autonomous agents with intent. A caseworker is an agent who promises to assess, refer, and support. A digital platform is an agent that promises to store, retrieve, and display. A patient is an agent who promises to attend, participate, and report. The objects in a service's state space are not just things with properties; they are agents with commitments.
Burgess argues that "designing a system, which explicitly promises and documents intent, means one can gradually test hypotheses as a system evolves". This connects the object question to the design question at the heart of this series: constructing a state space involves not just specifying what exists and what properties it has, but specifying what promises are in play and what commitments have been made. Promise Theory is treated at length later in this series; the point here is that one productive answer to "what is an object?" in the service context is "an autonomous agent capable of making promises".
Entity-First Thinking in Design Practice
The theoretical traditions surveyed above might seem disparate and in some cases remote from design practice, but several design communities have arrived independently at object-first or entity-first approaches.
Object-Oriented UX (OOUX), developed by Sophia Prater, begins with a deceptively simple question: "what are the objects in this system?" Rather than starting with user tasks and flows (verbs), OOUX asks designers to identify the objects, their attributes, their relationships, and only then the actions that can be performed on them. The actions are understood as affordances of objects - what the objects invite users to do.
Daniel Jackson's The Essence of Software (2021) makes a related argument through what he calls "concept design". Software should be understood as a composition of concepts - each with a name, a purpose, a set of states, a set of actions, and an operational principle that explains how the concept fulfils its purpose. The operational principle is a narrative scenario showing how states and actions work together to deliver value. A "reservation" concept has states (held, confirmed, cancelled), actions (reserve, confirm, cancel), and an operational principle ("after reserving a table and later confirming, the user is guaranteed a place"). This is much closer to how service designers already think than raw statecharts, but it makes state explicit rather than implicit.
The Handbook of Conceptual Modeling formalises this convergence through OPM: objects and processes as the two universal building blocks of system models. Enterprise architecture has its own tradition, with entity lifecycles as a primary modelling perspective - tracking how entities change state as they move through organisational processes. The NATO Architecture Framework includes state transition models as a standard view.
The pattern across these traditions is that the noun-first question - what are the things? - turns out to be logically prior to the verb-first question - what do users do? This runs counter to a generation of service design practice organised around verbs and actions, from Lou Downe's "good services are verbs" to Jobs to Be Done. The theoretical point for this series is that a state-space representation must first identify what objects occupy the space before it can specify what events change their states.
Objects as the Complement to Events
If state spaces are the formal apparatus for representing dynamic social situations, then the ontological picture requires two complementary halves. Objects define what occupies the state space - the entities whose configurations constitute "states". Events define the transitions between states - the happenings that change object properties. Properties (quality dimensions, attributes, variables) define the axes of the space. And identity and persistence determine what it means for an object to be tracked across transitions.
The Pathway Generator's patient vector is one formalisation of this: the patient is the object, the variables are the properties, the recommended pathway is a sequence of events intended to move the object toward a goal state. The ICF is another: the person is the object, the classification domains are the properties, functioning is a configuration in that property space. Each makes different design decisions about what kind of object the patient is, what properties matter, and what it means for the patient to change.
Different traditions foreground different aspects of objecthood. Conceptual modelling offers the formal apparatus. Gärdenfors provides the cognitive grounding. Persistence conditions explain what it means for objects to endure across change. Affordances explain what objects make possible. Actor-Network Theory explains how objects function in practice. OOO reminds us that our formalisations are always partial. Promise Theory reconceives objects as agents with intent. None of these is sufficient alone. The state-space representations this series is working toward will need to draw on several of them, depending on what is being represented and for what purpose.
The next post turns to the other half of the ontological picture: the events that happen to objects - how they are cognitively structured, how they accumulate across life courses, and how they occur in nested ecological contexts.
References
Burgess, M. (2020). A Treatise on Systems Volume 2: Intentional Systems and Complexity Science. ChiTek-i.
Davis, J. (2020). How Artifacts Afford: The Power and Politics of Everyday Things. MIT Press.
Embley, D. and Thalheim, B. (eds.) (2012). Handbook of Conceptual Modeling: Theory, Practice, and Research Challenges. Springer.
Franssen, M., Lokhorst, G.-J. and van de Poel, I. (2013). Ontology and the Human-Made World. In Olsen, J.K.B., Pedersen, S.A. and Hendricks, V.F. (eds.) A Companion to the Philosophy of Technology. Blackwell.
Gärdenfors, P. (2000). Conceptual Spaces: The Geometry of Thought. MIT Press.
Gärdenfors, P. (2017). The Geometry of Meaning: Semantics Based on Conceptual Spaces. MIT Press.
Harman, G. (2018). Object-Oriented Ontology: A New Theory of Everything. Penguin.
International Organisation for Standardisation (2022). ISO 704:2022 - Terminology work - Principles and methods. ISO.
Jackson, D. (2021). The Essence of Software: Why Concepts Matter for Great Design. Princeton University Press.
Johnson, J. and Henderson, A. (2011). Conceptual Models: Core to Good Design. Morgan & Claypool.
Krippendorff, K. (2005). The Semantic Turn: A New Foundation for Design. CRC Press.
Mylopoulos, J. (1992). Conceptual Modelling and Telos. In Loucopoulos, P. and Zicari, R. (eds.) Conceptual Modelling, Databases, and CASE. Wiley.
Zenker, F. and Gärdenfors, P. (2015). Communication, Cognition, and Technology. In Zenker, F. and Gärdenfors, P. (eds.) Applications of Conceptual Spaces: The Case for Geometric Knowledge Representation. Springer.