AI and Health Services: Benefits, Risks and Implications for Disabled People
Artificial intelligence is becoming increasingly influential in developing and delivering health services. It is being used to analyse medical images, support diagnosis, predict health risks, manage appointments, automate administrative tasks, personalise treatment and provide information through digital assistants and chatbots.
For people with disabilities, these developments could bring major benefits. AI may make health services more accessible, responsive and personalised. It could support earlier diagnosis, reduce the burden of navigating complex systems and help people manage their health more independently.
However, AI can also reproduce existing inequalities, introduce new barriers and reduce human accountability. The impact will depend not only on the quality of the technology, but also on how systems are designed, tested, governed and implemented.
More personalised and responsive healthcare
One of the most promising benefits of AI is its ability to analyse large amounts of information and identify patterns that may be difficult for clinicians to detect quickly.
For disabled people with complex or multiple health conditions, this could support more personalised care. An AI system might bring together information from medical records, medication histories, diagnostic tests and data from monitoring devices to help clinicians understand a person’s needs more fully.
AI could also improve early identification of health problems. Changes in movement, speech, breathing, sleep or behaviour may indicate that a person’s condition is changing. When used carefully, monitoring tools may identify these changes before they become a crisis.
This could be particularly valuable for people who have difficulty recognising, describing, or communicating symptoms. However, such systems should support clinical judgement and communication, rather than replace direct engagement with the person.
Improving communication and access
Many disabled people face communication barriers in healthcare. Information may be provided in inaccessible formats, appointments may rely heavily on spoken communication, and digital services may not work well with assistive technologies.
AI can reduce some of these barriers. Speech recognition and text-to-speech tools can support communication between patients and professionals. Live captioning may help deaf and hard-of-hearing people. Image descriptions may improve access for blind and partially sighted people. Language simplification tools can help create Easy Read or plain-language health information.
AI may also support people who use augmentative and alternative communication. For example, predictive language systems could help users prepare questions, explain symptoms or communicate more quickly during an appointment.
These benefits are not automatic. Systems must be compatible with screen readers, switch access, alternative keyboards, eye-gaze systems, voice control and other assistive technologies. People must also be able to choose how they communicate rather than being forced to use a standard digital interface.
Greater independence and self-management
AI-enabled health applications may help disabled people manage medication, monitor symptoms, prepare for appointments and understand treatment plans.
Virtual assistants could provide reminders in a person’s preferred format. Wearable devices might detect falls, seizures, changes in heart rate or other indicators of concern. Smart-home technologies could connect health monitoring with environmental control, helping a person call for assistance or adjust their surroundings.
Used well, these tools could increase independence and reduce avoidable hospital admissions. They may also help people participate more actively in decisions about their health.
However, there is a risk that “independence” becomes a justification for withdrawing human support. Technology should expand choice and control, not replace care workers, interpreters, advocates, personal assistants or trusted relationships when these remain necessary.
Reducing administrative barriers
Health services can be difficult to navigate, particularly for people who interact with multiple clinics, specialists, social care agencies and support services.
AI could reduce some of this administrative burden by coordinating appointments, identifying missing information, summarising records and helping patients understand what will happen next. It may also free health professionals from repetitive administrative work, creating more time for direct care.
Yet automated administration can create serious problems when systems are inflexible. A chatbot may not understand an unusual request. An appointment system may not record that someone requires extra time, accessible transport, a sign-language interpreter or the presence of a supporter.
Human assistance must remain available, especially when a person cannot use the digital route or when the situation is complex.
Bias and discrimination
One of the most significant risks is that AI systems may reflect bias in the data used to develop them.
Get David Banes’s stories in your inbox
Join Medium for free to get updates from this writer.
Disabled people are often underrepresented in health datasets. When included, their experiences may be recorded through assumptions that treat disability primarily as illness, dependency or reduced quality of life.
An AI system trained on incomplete or biased data may be less accurate for people with particular disabilities. Speech recognition may perform poorly for people with dysarthria or non-standard speech. Systems using facial expression to assess pain or emotion may misinterpret people with facial paralysis, autism, cerebral palsy or other conditions.
Risk prediction systems may also reach unfair conclusions if they confuse disability with poor health outcomes or lower treatment value.
Bias can be especially harmful when AI influences access to treatment, rehabilitation, mental health support, organ transplantation or other scarce resources. Decisions must never be based on assumptions that a disabled person’s life is less valuable.
Privacy, surveillance and consent
Health information is highly sensitive. AI systems may use medical records, genetic information, behavioural data, location data, voice recordings or information collected by wearable devices.
Disabled people may be particularly exposed to surveillance because they often interact with health, care, education, employment and benefits systems. Data collected for one purpose could potentially be used for another, such as assessing eligibility for support, insurance or employment.
Consent may also be difficult when systems are complex and people are not clearly told how their information will be used. Some individuals may feel they have little real choice if refusing data collection means losing access to a service.
Health organisations must minimise data collection, explain systems clearly and provide meaningful choices. People should be able to challenge decisions, correct inaccurate information and withdraw consent where appropriate.
The danger of losing human judgement
AI can identify patterns, but it does not understand a person’s life in the way that the person, their family or a skilled professional can.
A system may not recognise that a recommended treatment is inaccessible, that medication side effects affect someone’s ability to communicate, or that a person’s support arrangements make a particular care plan unrealistic.
Health professionals may also place too much trust in automated recommendations. Human oversight must involve more than simply approving what the system suggests. Professionals need enough knowledge, time and authority to question the output.
Patients must also be told when AI has influenced a decision and how they can request a human review.
Designing health AI with disabled people
The strongest way to reduce risk is to involve disabled people throughout AI development and implementation.
Participation should begin when the problem is being defined, not after a system has already been built. Disabled people and their representative organisations should help decide what the technology is intended to achieve, what data it uses, how success is measured and what safeguards are needed.
Testing should include people with different physical, sensory, cognitive, communication and psychosocial disabilities. It should also reflect differences in age, gender, ethnicity, language, income and access to technology.
Accessibility, equality and human rights should be treated as essential measures of quality, alongside safety, accuracy and efficiency.
A tool for better healthcare, not a substitute for inclusion
AI could help create health services that are more accessible, predictive and personalised. It could support communication, independence and earlier intervention while reducing unnecessary administrative barriers.
But it could also intensify discrimination, surveillance and exclusion if disabled people are missing from the data, the design process and the decisions that shape implementation.
The central question is therefore not whether AI should be used in healthcare. It is what kind of health service AI is being used to create.
AI should strengthen relationships, support professional expertise and give disabled people greater control over their health. It should not be used to reduce rights, remove human support or make inaccessible systems appear more efficient.
The future of AI-enabled healthcare must be shaped with disabled people, not simply delivered to them.
