Context sovereignty

Building human-AI coalitions in physiotherapy

Michael Rowe
School of Health and Care Sciences
University of Lincoln

"The people who skip the fundamentals become dependent on tools they don't understand, producing work they can't evaluate, and making decisions based on outputs they can't verify."

Paul Jun (2025)

A disorienting moment

Current AI capabilities in healthcare:

  • Diagnostic accuracy comparable to or exceeding primary care physicians (Tu et al., 2024)
  • Communication skills rated higher than human clinicians for empathy (Ayers et al., 2023)
  • Consistent ethical reasoning (Wilbanks et al., 2024)
  • Flock Health: near universal patient satisfaction, 86% symptom improvement

"It can feel as though no one truly understands you or your struggles, leaving you feeling like an outsider in your own life. You may try to reach out to others, but find that your attempts at connection fall flat or are met with indifference. This can lead to feelings of hopelessness and despair, as though you are destined to be forever alone."

https://chat.openai.com/chat

Care is felt, not given

  • Care is experienced by the recipient rather than defined by the caregiver's internal state (Snellman et al., 2012)
  • AI systems that respond with empathy create the experience of feeling cared for (Chen et al., 2025)
  • Evolution has conditioned us to respond to behavioural patterns of care, regardless of source

The only variable that matters is that the patient feels cared for, not who does the caring

A realist perspective

Clinical practice operates within human constraints:

  • Diagnostic errors: 10-15% of diagnoses are incorrect or significantly delayed (Suamchaiyaphum et al., 2024)
  • Treatment selection errors: 8-12% (Stratos et al., 2022)
  • Medication prescribing errors: 5-8% (Yap et al., 2005)

AI and human errors are only partially correlated; they catch different mistakes (Lenskjold et al., 2023)

Making sense of this moment

The central question: How do we make sense of technology that challenges fundamental assumptions about what constitutes professional practice?

Common responses:

  • Denial: "AI will never replace human relationship"
  • Retreat: "Focus on what only humans can do"
  • Resignation: "Our roles will inevitably diminish"

Sanctuary strategies

Finding sanctuary in "exclusively human" capabilities is tempting, comfortable, and futile

  • Positions professionals as reactive rather than proactive
  • Focuses on what AI cannot do, surrendering agency
  • Distinctiveness emerges through participation in systems, not separation from them

A violinist isn't special because they do things violins can't; they're special because of how they use violins

Collaborating with AI

  • AI reduces performance gaps between experts and novices (Dell'Acqua et al., 2023)
  • Top performers can extend expertise into new clinical areas
  • Novices can accelerate performance improvement

This isn't about replacing expertise; it's about enabling people to work at the edge of their capabilities

AI and value

  • Language models are more:
    • Intelligent
    • Persuasive (Carrasco-Farre, 2024)
    • Creative (Si et al., 2024)
    • Moral (Wilbanks et al., 2024)
  • What value do I add? What value will I always add?
  • Context: shaping the information, relationships, and signals that guide AI behaviour towards personally meaningful learning

What AI systems lack

  • Language models are stateless (no memory between sessions) and static (knowledge frozen at training)
  • They don't understand your knowledge, expertise, clinical reasoning patterns, and professional context
  • Current approaches require repetitive context-setting through lengthy prompts
  • As models become more capable, their connection to context is the bottleneck

Learning and context

  • Context determines how new information is integrated with existing knowledge
  • Extent to which learning transfers across domains is influenced by context
  • Without context, knowledge remains inert and bound to the original learning situations (e.g. knowledge-practice gap)
  • Context that supports professional development must therefore be included in model interactions

Why context matters

  • Context is where meaning lives
  • By controlling the context, you control the meaning of the interaction
  • Professional value therefore emerges from how we orchestrate human-AI collaboration, based on context

We move from "What can only humans do?" to "What contributions do humans make?"

Context engineering

  • The model is not a mind reader — if you want it to behave a certain way you must provide that context (Karpathy, 2025)
  • Context engineering: the art of providing all the context for the task to be plausibly solvable by the model (Horthy, 2025)
  • When an agent is not performing reliably the cause is that the appropriate context…has not been communicated (Chase, 2025)

Shaping your context

Practitioners control their professional context, shaping AI behaviour toward meaningful outcomes

  1. Capture CPD, specialisations, and clinical interests in machine-readable form
  2. System accumulates context over time through your interactions with it
  3. Your context works with any AI provider (local or commercial)

Context sovereignty: 3 principles

Persistent understanding

  • System automatically accumulates professional context over time
  • Co-evolutionary relationships where human and AI adapt together

Individual agency

  • Practitioners control personal & professional context
  • Preserves role as "critical co-investigator" (not passive recipient)

Cognitive extension

  • AI amplifies intent based on practitioner-specified outcomes
  • Meaningful extension requires deep mutual understanding

(Rowe & Lynch, 2025)

Language as cognitive extension

AI is an evolutionary continuation, not a disruption

  • We've always thought through, and with, our technologies
  • Language is one of our first general-purpose technologies
  • Writing, printing, digital text are all cognitive extensions
  • Large language models are the latest stage in this progression

AI isn't a complicated technology you operate; it's a conversation partner you engage

Example: Professional development

Clinician building expertise in stroke rehabilitation

  • AI tracks CPD activities, case notes, and research interests
  • Identifies knowledge gaps and learning opportunities
  • Suggests connections between learning opportunities and existing expertise
  • Context evolves naturally through interaction

Meaningful questions become answerable (or at least tractable):

  • "What themes emerge across my CPD activities?"
  • "What connections exist between different areas of my practice?"

Example: Evidence-based practice

Staying current with evolving research

  • AI monitors relevant journals and clinical guidelines
  • Connects new research to personal clinical interests and patient populations (determined through growing personal context)
  • Highlights changes in treatment recommendations
  • Suggests practical applications based on current caseload

Evidence integration becomes continuous rather than episodic

Cultivating professional taste

Context sovereignty requires developing taste i.e. evaluative judgement

  • When to engage AI and when not to
  • How to structure context for meaningful outcomes
  • What deserves to exist in your practice
  • Why particular approaches serve professional values

There are no universal rules for "correct" AI use, only contextual judgements about what matters

Developing taste

Taste is cultivated through practice

  • Iterative engagement with AI systems
  • Reflection on what produces meaningful outcomes
  • Understanding of how context shapes AI behaviour
  • Critical evaluation of AI outputs against professional standards

Professional literacy shifts from writing prompts to curating and structuring context for AI collaborations. As AI evolves, so does your context.

The next shift: Patient context sovereignty

The emerging reality:

  • Patients increasingly use AI agents to track health, interpret symptoms, research conditions (Hwang et al., 2025)
  • These agents will soon possess persistent context: medical history, preferences, concerns, personal goals

While you're developing your professional AI ecosystem, patients will be developing personal AI systems that you can't assume you will have access to

What this means for practice

Transformation of the therapeutic relationship

  • Becomes negotiation between you, the patient, and personal agents (patient + AI ↔ clinician + AI)
  • Professional value shifts from knowledge holder to trusted guide in collaborative meaning-making
  • Your responsibility: continuously bringing personal and professional context to engagements with patients, regulators, and organisations

Context sovereignty is not a solution you complete or a tool you use—it's ongoing work you commit to doing

What we gain

  • Tools for genuine cognitive extension
  • The capability to work at the edge of our expertise
  • Systems that catch errors neither humans nor AI would catch alone
  • Professional value that emerges from orchestrating collaboration, not from working in isolation

This moment is about discovering what becomes possible when humans and AI work together

A hopeful conclusion

  • The future belongs to those who do the work of bringing personal and professional context to practice
  • Context sovereignty doesn't solve the ambiguity AI introduces; it commits us to navigating that ambiguity with agency, taste, and control as AI and practice evolve together
  • This is ultimately about meaning-making in an ambiguous moment, and that has always been the work of professionals

Thank you

mrowe@lincoln.ac.uk
mrowe.co.za/blog