Why it matters
Without AI, manual records miss subtle welfare cues, monitoring gaps allow mis-timed procedures, and accreditation reviews struggle with inconsistent training documentation and repeatable welfare signals.
From design to safety
Design
Design AI-supported training workflow
Define data flows from training records, voluntary behaviours, blood draws, ultrasound practice, and reinforcement signals to establish decision rules for welfare checks and procedural readiness, ensuring complete traceability from concept to implementation. Extend planning to include data governance, metadata schemas, role-based access, and a scalable data lake strategy that anticipates growth in biometric streams and session volumes for diverse species groups.
- training records
- voluntary behaviours
- reinforcement
Build
Build data capture and integration
Install observation cameras, wearable or veterinarian-noted data entry points, and secure databases linking to LSS controls and Exhibit management systems, with standardized schemas and access controls. Integrate COTS sensors, edge analytics devices, and console dashboards that present real-time red/amber/green status for key welfare indicators and training readiness checks, with offline modes for remote sites.
- training records
- blood draws
- ultrasound training
Operation
Operate AI monitoring during sessions
AI continuously reads session logs, biometric readings, and behavioural cues to validate procedure timing and detect stress indicators, integrating with dive logs and welfare dashboards. It correlates trainer cue timings, reinforce schedules, and animal responses to surface proactive adjustments and generate routine welfare summaries for the care team and visiting observers.
- voluntary behaviours
- blood draws
- ultrasound training
Maintenance
Maintain data quality and calibration
Regularly calibrate sensors, verify data integrity, update reinforcement cues, and audit training records for completeness, logging versioned changes and audit trails. Schedule quarterly calibration against a standard protocol, track sensor drift, and maintain a change-log with reviewer initials and archival IDs for traceability.
- reinforcement
- training records
- blood draws
Safety
Safety and accreditation-ready oversight
AI issues flags for welfare concerns, procedural non-conformities, and ensures traceable records for audits and studbook reporting, with escalation to veterinary review and regulatory sign-off. Maintain a safety matrix mapping to accreditation criteria, and preserve a tamper-evident chain of custody for all welfare decisions and training adjustments.
- ultrasound training
- reinforcement
- training records
Part by part
- 1
training records
- Watch for
- incomplete entries, missing trainer initials, inconsistent timestamps across sessions
- AI action
- flag gaps, prompt for missing fields, trigger review workflow
- 2
voluntary behaviours
- Watch for
- decline in participation rate, abrupt behavioural changes, fatigue cues
- AI action
- score welfare state, recommend pacing adjustments or rest days
- 3
blood draws
- Watch for
- unusual vein sites, repeated sampling in short intervals, signs of distress
- AI action
- verify protocol compliance, suggest alternative sites, log observer notes
- 4
ultrasound training
- Watch for
- equipment idle time, suboptimal probe angles, poor image quality
- AI action
- check training cadence, reassign tech support, log quality metrics
- 5
reinforcement
- Watch for
- incongruent rewards, delayed reinforcement, trainer cues conflicts
- AI action
- align cueing with welfare rules, auto-generate reinforcement logs
- 6
trainer cue device
- Watch for
- device malfunction, battery wear, cue drift
- AI action
- monitor device health, schedule maintenance, adjust training plans
Worked example
Catching a dolphin peripheral vein draw before a medical check
Illustrative scenario with sample data.
Routine blood sampling requires careful timing and welfare monitoring during a voluntary approach behaviour.
Signal in
session_id=DPL-1024, date=2026-08-12, volunteer_id=VV-673, session_duration=28m, behaviours=approach:good, tail-slap:none, vitals=HR:60-76, SpO2:98, age:12, weight:183kg, trainer_id=T-19, location=Pool-A, ambientTemp=22.6C, waterTemp=22.0C
System out
AI flags potential distress: elevated HR during venipuncture; recommend pausing, re-evaluating sedation plan, and recording welfare notes; create a standardized blood draw procedure entry with risk factors and consent; auto-create a welfare review task and attach it to the study plan.
Human sign-off: Aquatic veterinarian reviews the flag, approves pause, and updates the training protocol and welfare notes in the records; senior curator signs off on revised procedures and notes; update study plan and notify training staff.
How to roll it out
Phase 1
Weeks 1 to 4
Baseline data capture from training records and voluntary behaviours across two exhibits.
Phase 2
Weeks 5 to 12
Integrate blood draws and ultrasound practice with reinforcement signals; begin real-time AI alerts.
Phase 3
Weeks 13 to 20
Full workflow operational; audits and accreditation-ready reporting enabled.
Where people stay in charge
- Decisions affecting animal welfare must be signed off by the aquatic veterinarian or senior curator.
- AI outputs indicating potential distress or adverse events require immediate human review before action.
- Records affecting accreditation and studbook compliance must be validated by a designated records officer.
Questions owners and crews ask
What data should we feed the AI for training logs and behaviour tracking?
Feed structured session data, trainer notes, welfare indicators, behaviour labels, and biometric readings; ensure timestamps and observer IDs are consistent for auditability, including location, equipment IDs, and session outcomes.
Who approves AI-driven welfare decisions?
Qualified personnel such as the aquatic veterinarian or senior curator must review and sign off on AI-flagged welfare concerns or procedural changes, documenting justification and sign-off in the records.
How does AI handle inconsistent trainer input in training records?
AI detects gaps, reconciles timestamps, prompts for verification, and triggers a manual review workflow to ensure data integrity, with audit-ready notes and corrective action tracking.
Can this system support accreditation inspections?
Yes, automating traceable training records, welfare signals, and procedural checks strengthens evidence for compliance and studbook reporting, with exportable reports aligned to inspection criteria.
Sources and further reading
- AMMPA: Standards and GuidelinesAccreditation criteria for marine mammal pools, training, and welfare records that facilities must meet.
- AMMPA: Aquatic Mammals JournalPeer-reviewed studies on marine mammal behaviour changes, participation, and fatigue indicators in training.
- GOV.UK: Standards of Modern Zoo PracticeStatutory zoo licensing standards setting expectations for welfare records and documentation.
- The Marine Mammal Center: RehabilitationMarine mammal rehabilitation pool practice, medical care, and patient progress monitoring for distress signs.
- WAZA: Approach to Animal WelfareGlobal welfare strategy describing five domains used to assess welfare across collections.
- International Association for Aquatic Animal MedicineProfessional network and conferences for aquatic animal veterinarians handling procedural welfare decisions.
- Species360: MissionZIMS data standards and global animal records structures for consistent collection data.
- HSE: Diving at WorkDiving at Work Regulations and approved codes covering dive team records and safety supervision.
- EAZA: Population ManagementStudbook management and population records guiding breeding programmes for collection animals.
- GOV.UK: Keeping Zoo AnimalsZoo licence requirements and inspection obligations for aquariums open to the public.