Artificial Intelligence (AI) principles in action

AI in general

These principles apply across all physiotherapy contexts—clinical, educational, workforce, and research—where AI is used. They reflect the CSP’s commitment to ensuring AI is transparent, effective, ethical, and aligned with physiotherapy values and public interest.

1. Transparency

It should be clear to all stakeholders—patients, clinicians, staff and system partners—when AI is being used, what its role is, and how it may influence decisions.

• Open communication about AI use builds trust and supports informed engagement by patients and professionals alike.

2. Effectiveness

AI tools should be based on robust, real-world evidence and demonstrate tangible benefit in practice—not just theoretical potential.

• Tools should be regularly reviewed to ensure they continue to deliver value, and decommissioned if they do not.

3. Avoid Vendor Lock-In

Procurement of AI tools should ensure flexibility and avoid long-term dependence on a single provider.

• Physiotherapy services should retain autonomy to adapt, switch or reject tools that no longer meet service or ethical needs.

4. Equity

AI systems should be tested and monitored regularly for bias, and designed to produce fair outcomes across all population groups.

• Equity should be embedded throughout AI development, procurement and use, with attention to intersectionality and social determinants of health.

5. Accountability

Even when AI supports a decision, a physiotherapist should remain accountable for its application in care or service delivery.

• There should always be a clear route for human review and explanation—especially where decisions are complex, contested, or affect safety.

6. Intellectual Property

AI tools developed using physiotherapy expertise or patient data should fairly recognise and protect intellectual contributions.

• Organisations and individuals should avoid giving away valuable knowledge or data ownership without clear purpose and oversight.

7. Confidentiality

Any AI system accessing personal or health information must comply with GDPR and uphold the highest data security standards.

• Practitioners should understand how data is stored, processed and shared, and be able to explain this to service users.

8. Risk Management

AI tools should undergo formal risk assessment before use and be continually monitored for emerging risks over time.

• This includes clinical safety, reputational risk, workforce implications, and unintended ethical or operational consequences.

9. Sustainability

Environmental sustainability should be considered in AI procurement and deployment—including energy use, hardware demand, and e-waste.

• The carbon and material impact of AI should align with the NHS and CSP’s broader commitments to climate and planetary health.

AI in service delivery

10. Patient and Public Involvement (PPI)

People who use physiotherapy services should be involved in shaping how AI tools are designed, evaluated, and implemented.

• Involving patients and communities helps ensure that AI tools reflect real needs and values, and that they work in practice—not just theory.

11. Choice in Delivery

Patients should be offered choice, including the ability to receive care without the use of AI tools if they prefer.

• AI should never replace human interaction as the only option —particularly where it may affect trust, understanding, or outcomes.

12. Workflow Integration

AI should be integrated into physiotherapy workflows in ways that enhance—not disrupt —existing systems and processes.

• Deployment should be planned with the clinical pathway in mind, ensuring tools save time, support decisions, and reduce duplication.

13. Interoperability

AI systems should be designed to work with existing digital health records and NHS infrastructure.

• Fragmented or siloed systems waste time and increase risk. Interoperability supports safer and more efficient care.

14. Personalisation

AI tools should support personalised care, recognising the uniqueness of each patient’s goals, preferences, and lived experience.

• Physiotherapists should be able to adjust AI-supported approaches to suit the individual—not the other way around.

15.Digital Clinical Safety Understanding

Services using AI should include processes for identifying, reporting, and responding to digital clinical safety risks.

• Staff should be trained to recognise when an AI tool may be causing harm, making unsafe recommendations, or creating a data issue—and know what to do next.

AI in practice

16.Safety

AI used in clinical settings should comply with relevant medical device regulations and include clear routes for human escalation.

• Physiotherapists should be confident that tools used in care meet safety standards and that decisions can be escalated when concerns arise.

17. Consent

Patients should provide informed consent before AI tools are used in their care, particularly when their personal data contributes to AI models.

• Consent should be meaningful—not hidden in fine print—and patients should understand what the AI is doing and how it affects their treatment.

18. Choice

Patients should be given the option to decline the use of AI-supported services where possible, and be reassured that this will not compromise their access to care.

• Respecting patient choice is key to maintaining trust and therapeutic relationships in a digitally enabled environment.

19. Clinical Accountability

Physiotherapists remain accountable for care decisions, even when they are supported by AI.

• AI may inform decisions, but it should never replace the clinical judgment or ethical responsibility of the practitioner.

20. Scope of Practice

Physiotherapists should only use AI tools that align with their personal scope of practice and professional competence.

• Using AI must not extend practice into areas where clinicians lack training, regulatory permission, or sufficient knowledge to interpret outputs.

21. Clinical Reasoning

AI should support, not override, personalised, patient-centred care based on holistic assessment and shared decision-making.

• Clinicians should retain the ability to tailor care based on clinical reasoning and patient input, even if AI suggests a different route.

22. Digital Clinical Safety Understanding

Physiotherapists should understand the digital safety risks of the AI tools they use and know how to monitor and respond to issues.

• This includes recognising incorrect outputs, technical failures, data quality issues, or signs that a tool is not fit for purpose in a given situation.

AI in research and innovation

23. Research Priorities

AI should be treated as a strategic research priority within physiotherapy, especially where it has potential to improve care quality, access, or workforce sustainability.

• Physiotherapists should be encouraged and supported to lead and shape AI research that reflects the needs and values of the profession.

24. Evaluation of AI Tools

All AI tools should be evaluated using appropriate clinical, operational, and patient-centred outcome measures relevant to physiotherapy.

• This includes testing tools in real-world settings, not just ideal conditions, and publishing results—positive or negative.

25. Innovation Support

AI-related innovation should be supported where it enhances physiotherapy services, care pathways, or patient outcomes.

• Innovation should be responsible, with consideration of unintended consequences, and aligned with professional values and public good.

26. Evidence-Based Practice (EBP)

AI should support, not replace, evidence-based practice. Physiotherapists should use AI as a tool to enhance clinical reasoning—not as a substitute for it.

• Tools should help make sense of data, but decisions must still be grounded in the best available evidence, clinical expertise, and patient preferences.

27. Equality in Research Access

Opportunities to engage in AI research should be inclusive, with attention to representation across geography, protected characteristics, and career stages.

• Funding, authorship, and leadership of AI projects should reflect the diversity of the profession and wider population.

28. Dissemination and Knowledge Sharing

Findings from AI-related research should be shared openly and accessibly, with efforts made to ensure they inform practice, policy, and education.

• Knowledge mobilisation should be prioritised—not just publication. Intellectual property should be protected but not hoarded at the expense of shared learning.

AI in education

29. Support

AI should be used to enhance learning—not replace teaching. Tools that support access, feedback, revision or skills development can add value when used with care.

• AI should serve as an assistive layer to human educators, not a substitute for academic dialogue, mentorship, or critical discussion.

30. Inclusion

AI tools should be designed and implemented in ways that support equity and accessibility, particularly for neurodiverse learners and those facing barriers to traditional education.

• Use of AI should reduce—not reinforce—existing educational disparities, and ensure all learners benefit from innovation.

31. Design

Curricula should be intentionally designed to include AI literacy, ethical awareness, and real-world application in physiotherapy contexts.

• → This includes critical thinking about AI, understanding its limitations, and recognising its appropriate use in clinical practice.

32. Assessment and Coursework

Students should be encouraged to use AI tools ethically, appropriately, and transparently in coursework and learning, with clear boundaries and expectations.

• Assessment methods may need to evolve to reflect real-world digital competence and to ensure academic integrity is upheld in the age of generative AI.

33. Reflective Practice and Learning

Students and educators should be supported to explore how AI can be used in reflection, feedback, and learning conversations—without replacing personal insight or self-awareness.

• AI might help surface patterns or generate prompts, but reflection remains a human, relational process.

34. Simulated Learning

AI-enabled simulation (e.g. avatars, virtual placements, automated scenarios) can supplement traditional learning, particularly when placement capacity or face-to-face time is limited.

• These tools should be used to extend—not replace—hands-on clinical learning, and should reflect physiotherapy values and scenarios.

AI in leadership

35. Design Involvement

Physiotherapy leaders should ensure the profession is actively involved[EM1]  in the design and development of AI tools used in healthcare.

• Inclusion at the design stage helps prevent systems that misunderstand, marginalise, or misrepresent physiotherapy and its patients.

36. Implementation Readiness

Before adopting AI tools at scale, leaders should assess the organisational, cultural, and clinical readiness of their teams.

• This includes reviewing digital infrastructure, workforce skills, clinical risk, and team confidence to ensure safe and sustainable implementation.

37. Leadership Development

AI literacy, ethics and digital confidence should be part of leadership development across the profession—from new graduates to senior decision-makers.

• Leaders should feel equipped to ask critical questions about AI and model curiosity, caution, and courage in equal measure

38. Inclusion in Strategy

AI should be explicitly addressed in service, workforce, education and research strategies—not viewed solely as a digital transformation issue.

• Leaders should integrate AI considerations into wider planning, ensuring coherence with quality, equity, and workforce goals.

AI in work

39. Best Practice and Partnership Working

The introduction of AI in the workplace must involve early consultation with staff and trade unions, with clear risk assessments and partnership approaches.

• Decisions made without workforce engagement are more likely to fail, undermine morale, and create unintended consequences.

40. Disability and Neurodiversity Support

AI systems should be designed and implemented in ways that are inclusive of disabled and neurodiverse staff, supporting accessibility and equitable participation.

• Employers should avoid tools that disadvantage staff based on sensory, cognitive, communication, or any other needs.

41. Legal and Trade Union Advice

Organisations must seek legal and trade union advice when AI tools may affect employment conditions, clinical roles, job security, or performance management.

• This includes AI used in rostering, productivity tracking, or decision support that might influence appraisals, progression or disciplinary processes.

42. AI in Workforce Planning

AI should be used to support long-term workforce planning, helping to model future needs, identify skill gaps, and optimise staff deployment—without reducing roles to data points.

• Workforce models must still consider human factors, team dynamics, and the professional value of physiotherapists.

43. AI in Recruitment and Selection

Where AI is used in recruitment (e.g. CV screening, interview scheduling, aptitude scoring), it should be bias-tested, explainable, and subject to human review.

• Physiotherapists should be confident that AI is not reinforcing discrimination or excluding talent unfairly.

44. Business Continuity Planning

Organisations should account for the risk of AI failure or disruption in their business continuity and risk plans.

• Reliance on AI without contingency plans may compromise service delivery or create clinical risk in times of outage or cyberattack.

45. Efficiency Gains Shared

Where AI delivers measurable efficiency (e.g. time savings, automation), the benefits should be reinvested in improving patient care, workforce wellbeing, or service development.

• Cost savings should not be the sole justification for AI adoption—value should be shared fairly with the people it affects.

46. Workforce Displacement and Job Transition

AI should not be used to displace or devalue physiotherapy roles. Where changes are necessary, staff should be supported through role redesign, 6retraining, or redeployment.

• The introduction of AI should enhance jobs, not replace them. Fairness, dignity, and security must guide any transition.

Related AI in work resources

The institute for the Future of Work (IFOW) have produced the Good Work Algorithmic Impact Assessment (and related resources) to help employers assess the range of impacts that AI and other similar systems may have in work.

How is AI being used?

All AI is not being utilised in the same way and there are broad categories which you could place its potential use in physiotherapy. The following list is not exhaustive but should provide an understanding of how and where it may be used in physiotherapy, along with some perceived risks that the principles will aim to address. 

Generative AI

Generative AI such as ‘Chat GPT’ has the potential to create personalised exercise plans based on a patient’s specific needs, simulating recovery paths to help therapists estimate progress. It might also offer virtual environments for rehabilitation and assist in education for generation of information and interactive learning interactions. It could also assist with non-clinical quality work such as quality improvement, clinical audit, clinical governance documentation, designing, conducting and communicating research. However, there’s a risk that these AI-generated plans or education scenarios may not fully capture the complexities of an individual’s condition or recovery, potentially leading to generic solutions that overlook critical nuances in a patient’s treatment or education of learners. Generative AI tools are also known to “hallucinate” – to give false information, incorrect information returned from internet searching, or false academic references. Members should be aware of this risk and critically evaluate the strength of AI-generated information.

Machine learning (ML) and Predictive analytics

Machine learning can be useful for analysing patient data to predict recovery times, risk of re-injury, or treatment effectiveness. This could help physiotherapists provide more personalised care. Yet, these models rely heavily on the quality of data, and if they’re trained on biased or incomplete datasets, the predictions may be inaccurate, leading to incorrect treatment plans or missed risks, or reinforcement of societal biases. This is also true of non-AI processes and decisions, so should form part of regular risk management discussions and plans.

Computer vision

Computer vision technology is being explored to monitor patient movements and posture, offering real-time feedback during exercises to ensure proper form and technique. It can also support image interpretation which is relevant and already used in other professions in orthopaedics, respiratory and oncology. This could enhance accuracy and safety in physiotherapy. However, these systems may struggle with variability in human movements, potentially giving incorrect feedback, especially in diverse populations with different body types or movement patterns.

Natural language processing (NLP)

NLP could streamline physiotherapy by automating documentation or providing voice-guided exercise instructions for patients in home settings, or by reviewing unstructured records for clinical audit and quality assurance. This can improve efficiency and ensure patients follow their prescribed exercises. A potential risk is that NLP may misinterpret spoken language or patient input, particularly with accents or non-standard speech patterns, leading to misunderstandings or incorrect guidance.

Robotics and automation

Robotic systems, like exoskeletons and assistive devices, show promise in aiding rehabilitation, especially for patients with mobility challenges or those recovering from strokes. These tools offer consistent, guided therapy, which may enhance recovery. However, there’s a risk that over-reliance on robotics could reduce human oversight, and these systems might not adapt well to unexpected changes in a patient’s condition, potentially delaying necessary adjustments to the treatment. The productive and maintenance costs often raise another barrier to implementing as part of business as usual so any new implementations should consider that carefully before deployment.

AI ‘physiotherapists’, virtual assistants and chatbots

Online AI ‘Physiotherapists’, virtual assistants and chatbots could be used to deliver physiotherapy services and support physiotherapy by offering patients guidance for home exercises, answering questions, and adjusting plans based on progress. While they could help improve adherence to treatment, there’s a risk that patients might become too reliant on them, and these systems might fail to escalate issues that require a human therapist’s intervention, potentially leading to delays in addressing serious pathology and complications.

Employment and recruitment

AI is also being used by employers in recruitment, promotion, redundancy and performance monitoring. Automating mundane tasks offers opportunities to free up staff for more fulfilling work but the use of AI carries some risks, particularly regarding equality and diversity which is a generic risk that applies to more than this area.

Equality and diversity risks

There are equality considerations when adopting AI. AI tools are built largely by middle-class, well-educated men from limited racial backgrounds. The data and content utilised by AI algorithms can underrepresent, and often undervalues people of colour, women, working class people, the LGBTQIA+ community, people with impairments, neurodivergence and other marginalised characteristics. 

Due to these factors AI can embed existing biases in automated decision making and lead to discrimination and exclusion, unless these risks are managed. There are principles included above that aim to address these significant and important issues.