This document defines the primary use case for the Open SGI Minimum Viable Product (MVP). Its purpose is to:
- Describe the MVP in operational, human terms (not object definitions), and
- Provide sufficient detail to derive concrete requirements for:
- OAII Base Model instances (e.g., specific Worlds, Events, Policies), and
- Open SGI subclasses and profiles (MVP-specific specializations).
The MVP is explicitly non-diagnostic, edge-primary, and privacy-first.
1. MVP Goal
Enable an older adult to live independently at home with quiet, respectful, local intelligence that:
- notices meaningful changes in routine or activity,
- stays silent when nothing important is happening,
- escalates attention only when warranted,
- and does so without surveillance or cloud dependence.
The system supports residents, caregivers, and family members — without replacing human judgment.
2. Primary Actors
2.1 Resident
- Lives independently in a private home
- Does not interact with the system continuously
- Expects dignity, privacy, and minimal intrusion
2.2 Caregiver / Family Contact
- Receives notifications when warranted
- Requires understandable explanations
- Does not want false alarms or constant monitoring
2.3 Open SGI System
- Runs locally within the home
- Hosts sensors, recognition logic, and policies
- Communicates externally only per policy
3. Operating Environment (World Definition)
3.1 Primary World: Home World
Characteristics:
- Physical residence (apartment, house)
- Stable but evolving routines
- Private, consent-bound space
Contextual Axes (minimum MVP):
- Temporal: time of day, duration, routine windows
- Spatial: rooms, zones, entry/exit
- Activity: observable actions (movement, presence)
- Device State: sensor/device health
- Policy Context: quiet hours, escalation rules
This World defines what “normal,” “unusual,” and “meaningful” mean.
4. Sensors and Signals (MVP Scope)
4.1 MVP Sensor Set (Minimum)
- Motion sensors (PIR / IMU)
- Door / contact sensors
- Optional: location inference (room-level)
- Optional: episodic audio cues (non-continuous)
4.2 Signal Characteristics
Signals are:
- non-semantic
- time-referenced
- ephemeral by default
- summarized before retention
No continuous audio or video storage is assumed.
5. Sensor Knowledge (MVP Scope)
Each Sensor maintains local Sensor Knowledge, including:
- calibration parameters
- typical activity levels by time of day
- room transition norms
- acceptable inactivity durations
Sensor Knowledge is:
- learned incrementally
- World-scoped
- replaceable
6. Event Recognition (Core MVP Function)
6.1 MVP Event Types (Illustrative)
World-scoped examples:
- ROUTINE_ACTIVITY_PRESENT
- PROLONGED_INACTIVITY
- UNEXPECTED_ENTRY_EXIT
- NIGHTTIME_ACTIVITY_ANOMALY
- SENSOR_DEGRADED
These Events are:
- comparative (current vs baseline)
- bounded in time
- revisable
6.2 Event Lifecycle
- Signals accumulate
- Event enters CANDIDATE state
- Confidence increases or decays
- Event declared ACTIVE or COMPLETED
- Event may be REVISED or INVALIDATED
7. Knowledge Formation (MVP Scope)
Knowledge objects include:
- daily and weekly routine summaries
- typical inactivity thresholds
- historical event frequency summaries
Knowledge is used to:
- contextualize new Events
- reduce false positives
- explain why an Event matters
8. Policies (MVP Requirements)
8.1 Privacy Policies
- raw Signals not exported
- summaries preferred over detail
- retention limits enforced
8.2 Notification Policies
- quiet hours
- escalation tiers
- confirmation requirements
8.3 Safety Policies
- degrade safely on sensor failure
- suppress alerts under uncertainty
Policies are:
- explicit
- World-scoped
- user-adjustable within bounds
9. Agent Behavior (MVP Scope)
The MVP Agent:
- monitors Events
- consults Knowledge
- evaluates Policies
- selects responses
Example responses:
- do nothing
- log only
- local prompt to resident
- notify caregiver
The Agent does not diagnose or infer intent.
10. Interfaces (MVP Scope)
10.1 Resident-Facing
- subtle local prompts (audio/light/text)
- confirmation requests
10.2 Caregiver-Facing
- summarized notifications
- event explanations
Interfaces enforce:
- privacy
- policy constraints
- accessibility requirements
11. Logs and Accountability
Logs record:
- Event declarations
- Policy decisions
- Agent actions
- Knowledge revisions
Logs are:
- reference-based
- privacy-aware
- exportable by consent
12. MVP Requirements Trace (Derivation Aid)
This use case implies the following minimum OAII Base Model instantiations:
- 1 Home World instance
- 1–N Device instances
- N Sensor instances with Sensor Knowledge
- Signal streams per Sensor
- Defined Event subclasses (World-scoped)
- Knowledge subclasses (routine, baseline)
- Policy subclasses (privacy, notification)
- 1 Agent instance
- Resident and caregiver Interface instances
- Log streams
13. Explicit Non-Goals (MVP)
The MVP does not:
- diagnose medical conditions
- predict long-term health outcomes
- continuously record audio/video
- centralize personal data
- replace caregivers
14. Purpose of This Use Case
This document exists to:
- ground Open SGI development in human reality
- prevent scope creep into surveillance or diagnosis
- guide subclassing of OAII Base Model objects
- support transparent review by collaborators
This use case defines the reference problem for the Open SGI MVP and serves as the basis for deriving concrete requirements and profiles from the OAII Base Model.
Open Autonomous Intelligence Initiative 
Leave a comment