Artificial Intelligence as Aid or Appliance

Artificial intelligence (AI) can function as an aid or appliance by reducing barriers, compensating for limitations and making activities more accessible. However, removing effort is not always beneficial: some forms of effort contribute to learning, judgement and the development of independent capabilities. Drawing on disability studies, Nassim Nicholas Taleb’s concept of antifragility, research on productive difficulty, Donna Haraway’s cyborg theory and an exploratory epistemological analogy with transgender experience, this article proposes the concepts of anti-aid and anti-appliance for uses of AI that deliberately preserve or introduce productive friction. It argues that AI should neither be understood as inherently beneficial assistance nor rejected as an obstacle to human capability. Its role should depend on the person, the task, the circumstances and the capabilities that need to be supported or developed.

Contents

1. Aid and appliance
2. Antifragility and friction
3. The friction paradox in AI
4. Anti-aid and anti-appliance
5. Desirable difficulties and learning
6. The cyborg and ordinary disability
7. Transgender epistemology
8. Changing needs, changing roles
9. Rehabilitation and the challenge point
10. An optimal strategy for AI
11. Conclusion
12. References

Aid and appliance

Assistive technology is commonly understood as a broad category of products, systems and services that help people maintain or improve functioning, independence and participation. The World Health Organization’s work on assistive technology includes products associated with mobility, vision, hearing, communication, cognition and self-care.

An aid or appliance may compensate for an impairment, reduce the consequences of an inaccessible environment or enable an activity that would otherwise be difficult or impossible. A wheelchair, hearing aid, prosthesis, screen reader, communication device and pair of spectacles can all alter the relationship between a person and the world.

These technologies do not belong exclusively to a particular stage of life. Someone born with a disability may use an assistive product throughout their life. Another person may need one temporarily after an injury, illness or operation. Others may begin using aids or appliances as they age or approach the end of life. The same product may be essential in one context, useful in another and unnecessary in a third.

The distinction between an aid, an appliance and a prosthesis is not always sharp. An aid generally suggests assistance or support. An appliance often refers to a device used to perform a particular function. A prosthesis more specifically replaces or substitutes for a missing body part or bodily function, although the term is also used in broader discussions of technological extension. These terms overlap, but they do not mean exactly the same thing.

AI can perform functions associated with each category without necessarily being a prosthesis in the conventional medical sense. A speech-to-text system may assist a person who cannot easily type. An AI interface may make information more accessible to a person with a cognitive or sensory disability. A language model may help someone organise thoughts, understand unfamiliar material or communicate across a language barrier. In these cases, AI can operate as an aid or appliance even when it does not replace a bodily part.

The usefulness of the category depends on what the technology does in practice, rather than on whether its underlying mechanism is biological, mechanical or computational.

Antifragility and friction

Nassim Nicholas Taleb’s concept of antifragility describes systems that can benefit from volatility, disorder or stressors, rather than merely resist them or return to their previous condition. The concept invites a question about technologies designed to remove difficulty: can the elimination of an immediate burden also remove conditions that help a person develop?

Friction has more than one meaning. In physical systems, it describes resistance between surfaces. In ordinary life, it can refer to the effort, delay, uncertainty or resistance encountered while carrying out a task. Some friction is wasteful. Bureaucratic obstacles, inaccessible interfaces, repetitive paperwork and unnecessary administrative delays can consume time without developing any useful capability. Removing them is often desirable.

Other forms of friction may be productive. Working through a difficult problem, trying to recall information, learning to navigate an unfamiliar place or revising an unsuccessful argument can help develop knowledge and skill. The effort is not automatically valuable merely because it is difficult. Its value depends on whether it contributes to a capability, understanding or outcome that matters.

The distinction can be described as one between negative friction and positive friction. Negative friction imposes costs without a corresponding developmental benefit. Positive friction requires effort that can contribute to learning, judgement, resilience or competence. These are analytical terms proposed here, rather than established technical categories in the literature.

The relationship between effort and capability is not simple. Excessive difficulty can discourage learning, cause harm or make an activity inaccessible. The purpose is not to maximise friction, but to distinguish the obstacles that should be removed from the challenges that may be worth preserving.

The friction paradox in AI

Ravi Venkatesan’s article “Artificial intelligence and the friction paradox”, published in Hindustan Times on 28 October 2026, makes a closely related argument. It draws attention to the possibility that AI can improve immediate performance while reducing opportunities to develop the underlying capabilities that would allow a person to perform independently.

The article discusses examples involving navigation, education, customer service, medicine and early-career work. Its central concern is that the immediate benefits of automation are easy to observe, while the possible loss of skills may emerge slowly. It distinguishes work that creates needless frustration from challenges through which people become more capable. It also argues for AI systems that coach users and provide hints rather than simply supplying finished answers.

This argument is especially relevant to education. A student who asks AI to complete a mathematics assignment may submit more correct answers without necessarily learning how to solve the problems. The same student could instead ask AI to generate a personalised quiz, attempt each question independently and request hints only when necessary. In the second case, AI helps structure the learning environment without removing the student’s opportunity to practise.

The distinction is not between using AI and refusing to use it. It is between using AI to bypass the process through which a capability develops and using AI to support that process.

Venkatesan’s article is an important contemporary contribution to this discussion, but an opinion article is not a substitute for examining the underlying studies. The reported findings need to be checked against their original publications, including their methods, limitations and applicability to different users. The broader argument also requires qualification: reduced practice does not necessarily lead to a measurable loss of capability in every person or every task.

Anti-aid and anti-appliance

If an aid reduces a barrier, an anti-aid in the sense proposed here is a tool deliberately used to preserve or introduce a productive challenge. An anti-appliance is an AI system or configuration designed not simply to perform a task for the user, but to help the user develop or retain the capability to perform it.

These terms are not intended to describe technologies that obstruct disabled people, deny reasonable assistance or make essential activities unnecessarily difficult. They describe a particular mode of use, not a universal property of a product.

The same AI system can be an aid in one situation and an anti-aid in another. A person may use AI to summarise a long document because a disability, fatigue or time constraint makes reading the entire document difficult. The same person may later ask AI to quiz them on the document, withhold the answers initially and identify gaps in their understanding. In the first situation, AI reduces cognitive load. In the second, it creates a structured opportunity to practise recall.

The distinction is not between a good technology and a bad technology. It is between different relationships among a technology, a person, a task and a desired outcome.

An anti-aid should therefore be configurable. It might offer hints before answers, ask the user to attempt a task first, request an explanation of the user’s reasoning, reveal solutions in stages or provide feedback only after an independent attempt. It might also allow the user to choose when these restrictions should apply and when they should be removed.

For a beginner learning to programme, an AI system might explain a concept and provide a small exercise without generating the complete solution. For an experienced programmer working under a deadline, generating a routine section of code may be a sensible use of assistance. The difference is partly a matter of expertise and partly a matter of purpose.

An anti-appliance is not necessarily an additional product. It may be a deliberate way of configuring and using an existing AI tool.

Desirable difficulties and learning

Research on learning provides a basis for distinguishing productive challenge from unnecessary hardship. Robert and Elizabeth Bjork’s work on desirable difficulties examines how conditions that make practice more demanding can sometimes improve long-term retention and transfer, even when they make performance during practice less fluent. Manu Kapur’s work on productive failure explores how attempting to solve problems before receiving instruction can support subsequent learning under suitable conditions.

Neither line of research establishes that difficulty is always beneficial. A task can be so demanding that a learner lacks the knowledge or support needed to make progress. Nor does a difficult experience necessarily produce learning merely because the person has struggled.

The relevant question is whether the task is designed so that effort can contribute to a later capability. A student who attempts a problem, receives a hint, tries again and then examines a worked solution may benefit from the sequence. A student who is denied necessary explanation or accessibility support may simply be excluded.

This distinction matters for AI because the technology can provide different levels of assistance. It can deliver a complete answer immediately, guide a user through a sequence of questions, offer a hint or wait for the user to explain their reasoning. Each configuration changes the relationship between the learner and the task.

An anti-aid approach does not require AI to be less intelligent or less helpful. It asks the system to be helpful in a way that is consistent with the user’s longer-term objective.

The cyborg and ordinary disability

Donna Haraway’s “A Cyborg Manifesto” challenges rigid distinctions between human and machine, organism and technology, and other categories that are often treated as natural or self-evident. Her cyborg figure offers a way to think about human identity and capability as already entangled with technological systems.

Spectacles provide a familiar example. They can compensate for a visual impairment, but their everyday use also complicates the distinction between disabled and non-disabled people. A person who relies on spectacles may move between social categories depending on the context, the task and the availability of the device. The technology can become so ordinary that its assistive role is no longer noticed.

This does not mean that all people are disabled in the same way or that the material realities of disability can be dismissed. Disability involves different bodies, environments, barriers and social arrangements. Treating every human limitation as equivalent would obscure those differences.

It does, however, suggest that dependence on technology is not an exceptional condition confined to a clearly defined group. Human activity has long relied on tools, infrastructure, written language, communication systems and other forms of technological mediation. AI extends this relationship into tasks associated with language, memory, reasoning, planning and knowledge work.

AI may therefore operate as an aid or appliance for a person with a disability, an ageing person, a temporary patient, a student, a professional or someone facing an unfamiliar task. These uses are not identical, but they belong to a broader history of people using technologies to alter what they can do.

The cyborg perspective makes it harder to maintain a simple opposition between independent humans and dependent technology users. The more useful question is how technological dependence is organised, whose needs it serves and which capabilities it enables or displaces.

Transgender epistemology

The idea of transgender epistemology is proposed here as an exploratory analogy, not as an established term in epistemology or transgender studies. Its intended purpose is to question the assumption that a technology must have one fixed role: either an aid or an obstacle, either beneficial or harmful, either enabling or disabling.

Transgender experience can challenge assumptions that gender must be understood through an immutable binary or that the relationship between identity, embodiment and social recognition is always simple . However, an analogy between gender and knowledge must be handled carefully. Transgender lives are not merely metaphors for intellectual flexibility, and the diversity of transgender experience cannot be reduced to an epistemological model.

The proposed analogy concerns the limits of fixed categories. A person’s relationship with a technology may change over time, and different descriptions may be appropriate at different moments. This does not mean that every claim is equally true or that evidence becomes irrelevant. A finite world can contain many true statements about the same person or system because the statements concern different circumstances, times, needs or relationships.

For example, a language model may reduce the cognitive burden of drafting an email while making it harder for a user to practise writing unaided if it always supplies the finished text. Both observations can be true. They describe different effects of the same technology under different conditions.

The term remains open to criticism. It may be more precise to describe the argument as contextual epistemology or situated technological judgement. The term transgender epistemology should be retained only if the analogy proves analytically useful and can be developed without appropriating or misrepresenting transgender experience.

Changing needs, changing roles

The role of an aid can change within a single day. A person may use AI in the morning to reduce the effort required to understand a complex document, in the afternoon to generate practice questions and in the evening to test their unaided recall. The first use reduces cognitive load; the second creates a learning activity; the third checks what the person can do without immediate assistance.

The same variation applies across the life course. A person recovering from an injury may initially need technology to remove effort and support safe participation. Later, depending on the injury, treatment and professional guidance, the person may engage in carefully graded activities intended to regain function. An older person may use AI to simplify a task that has become burdensome while continuing to practise a different skill they wish to maintain.

These examples should not be interpreted as medical advice. The appropriate balance depends on the individual, the activity and the relevant clinical or accessibility needs. In some circumstances, attempting a task without assistance can be unsafe or needlessly exhausting. In others, independent practice may be an important goal.

A single person can therefore need assistance and challenge at the same time. The objective is not to classify the person as dependent or independent, nor to classify AI as beneficial or harmful. It is to decide which role best serves the particular need.

Rehabilitation and the challenge point

Research on motor learning offers a useful, though limited, comparison. Mark Guadagnoli and Timothy Lee’s challenge point framework proposes that the benefits of practice depend on the relationship between task difficulty and the learner’s skill level. The same practice conditions may be too easy for one learner and too difficult for another.

This framework is relevant to rehabilitation because recovery often involves changes in what a person can safely and effectively attempt. However, it should not be treated as proof that there is one universal optimum between assistance and effort. Rehabilitation outcomes depend on the condition, the person, the task, the stage of recovery and the clinical approach.

The same caution applies to AI. Learning research and rehabilitation frameworks can help formulate questions about AI use, but they do not by themselves establish a clinically validated AI strategy. Evidence about motor learning cannot simply be transferred to every cognitive task, and evidence about classroom learning cannot automatically be applied to disability support.

A useful research question is whether different levels of AI assistance can be matched to the user’s present capabilities and goals. For example, does a system that begins with hints and gradually increases assistance support learning better than one that always supplies complete answers? Does the appropriate level of support change as competence develops? How should users with disabilities be able to control the balance between effort and assistance?

These questions require direct empirical study. Until such evidence is available, the challenge point framework is best treated as a conceptual guide rather than a ready-made prescription.

An optimal strategy for AI

The aim should not be to maximise assistance or maximise friction. Both approaches are too crude. A system that always completes tasks can undermine opportunities for practice, while a system that withholds assistance can make work inaccessible or impose needless burdens.

A more useful strategy begins by identifying the user’s goal. If the goal is to complete a necessary task, reduce pain, communicate effectively or overcome an accessibility barrier, direct assistance may be appropriate. If the goal is to learn, practise or retain a skill, the system may need to preserve some of the work that supports that goal.

The next step is to distinguish necessary effort from wasteful effort. Repetitive administration, inaccessible formats and avoidable interface barriers rarely need to be preserved for their own sake. Reasoning through a problem, practising recall or learning how to evaluate an answer may be worth retaining when those capabilities are important to the user.

The system should then offer a suitable level of support. This may involve asking the user to attempt a task, providing hints before full solutions, making the reasoning process visible, checking understanding or gradually reducing assistance as competence develops. These features should remain adjustable rather than imposing one model of independence on every person.

Finally, the outcome should be evaluated against the original goal. Completing a task quickly is not the same as learning to perform it. Conversely, performing a task unaided is not always the most meaningful measure of success. Accessibility, safety, autonomy, participation, learning and long-term capability may all matter, but their relative importance will differ across contexts.

A good AI strategy should therefore make assistance responsive to the user’s circumstances, rather than treating frictionlessness as an absolute objective.

Conclusion

AI can be understood as an aid or appliance when it reduces barriers and enables activity. It can also be configured as an anti-aid or anti-appliance when the purpose is to preserve productive effort and support the development or retention of capability. These roles are not mutually exclusive, and the same system may move between them as a person’s needs change.

Disability studies help explain why assistance should be understood in relation to functioning, participation and context. Antifragility and research on learning draw attention to the possibility that some challenges contribute to capability, while Haraway’s cyborg theory challenges the assumption that technological dependence is an unusual departure from ordinary human life. The proposed transgender epistemology analogy explores the instability of fixed categories, although the terminology requires further critical examination.

The practical question is not whether AI should make human life easier or harder. It is which difficulties should be removed, which should be preserved and who should decide. The answer should be shaped by evidence, accessibility, individual goals and changing circumstances. AI is most useful when it can provide assistance without assuming that every task should be completed on the user’s behalf, and when it can introduce challenge without confusing unnecessary hardship with meaningful development.

References

1. World Health Organization. (2016). Priority Assistive Products List. Published 1 February 2016. Accessed 28 October 2026.
2. World Health Organization and UNICEF. (2022). Global Report on Assistive Technology. Published 15 May 2022. Accessed 28 October 2026.
3. Taleb, Nassim Nicholas. (2012). Antifragile: Things That Gain from Disorder. Random House. Accessed 28 October 2026.
4. Haraway, Donna. (1985). “A Cyborg Manifesto: Science, Technology, and Socialist-Feminism in the Late Twentieth Century.” Reprinted in Simians, Cyborgs, and Women: The Reinvention of Nature. University of Minnesota Press. Accessed 28 October 2026.
5. Bjork, Elizabeth Ligon, and Robert A. Bjork. (2011). “Making Things Hard on Yourself, But in a Good Way: Creating Desirable Difficulties to Enhance Learning.” In Psychology and the Real World: Essays Illustrating Fundamental Contributions to Society. Author-hosted PDF. Accessed 28 October 2026.
6. Kapur, Manu. (2008). “Productive Failure.” Cognition and Instruction, 26(3), 379–424. DOI link. Published online 7 July 2008. Accessed 28 October 2026.
7. Guadagnoli, Mark A., and Timothy D. Lee. (2004). “Challenge Point: A Framework for Conceptualizing the Effects of Various Practice Conditions in Motor Learning.” Journal of Motor Behavior, 36(2), 212–224. DOI link. Published June 2004. Accessed 28 October 2026.
8. Stanford Encyclopedia of Philosophy. “Trans Philosophy.” Entry. Accessed 28 October 2026.
9. Venkatesan, Ravi. (2026). “Artificial intelligence and the friction paradox”. Hindustan Times, 28 October 2026. Accessed 28 October 2026.

📄 This page was created on 28 October 2026. You can view its history on GitHub, preview the fileTip: Press Alt+Shift+G, or inspect the underlying source.