AI must benefit people with intellectual disabilities

When I first heard the term artificial intelligence, the first person I thought of as a mother was my child with an intellectual disability. If this technology was created to support human intelligence, I wondered, should it not first help people who need more support to understand, learn, and make judgments?

Rather than making the most capable people move even further ahead, AI should expand understanding and choice for people who have historically been left behind by technology. I believe that is one of the first things artificial intelligence should do.

A mother’s question

I do not want to offer my child’s name or private life as evidence for this argument. The question is clear without exposing one person’s privacy.

Who should artificial intelligence serve first?

Today’s AI summarizes writing, answers questions, creates images and video, and processes complex tasks quickly. For people who can already navigate large amounts of information, it offers even more speed and convenience. The progress is remarkable. But if the people best positioned to use technology are always the ones who move furthest ahead, can we call that progress for everyone?

Intellectual disability involves significant limitations not only in intellectual functioning such as learning, reasoning, and problem-solving, but also in adaptive behavior used in everyday life, including language and literacy, money and time, self-direction, schedules, and safety.4 This does not mean that people with intellectual disabilities are unable to make choices. The answer is not to remove choice, but to provide support tailored to each person so they can understand, express themselves, and make judgments.

Nalja Books’ declaration

AI must benefit people with intellectual disabilities. AI should expand their understanding, choice, and participation.

This is not a vague hope that technology may someday do something good. It is the standard Nalja Books uses when judging AI and educational materials.

The United Nations Convention on the Rights of Persons with Disabilities recognizes dignity, individual autonomy, freedom to make one’s own choices, and full and effective participation in society as foundational principles. It also calls for access to information and communication technologies, information in accessible formats, and support for communication and the exercise of legal capacity.1 These rights existed before AI. AI does not grant them. It should help people exercise rights that should already be guaranteed in everyday life.

What practical benefit means

Here, benefit does not mean only making work faster for families or institutions. It should create direct changes for people with intellectual disabilities:

They should be able to understand necessary information in a form that works for them.
They should be able to express their questions, feelings, and intentions.
They should be able to consider available options and their consequences and make a choice.
Their opportunities to participate in education, everyday life, and society should grow.

AI may help turn a long notice into clear sentences organized around its central meaning. It may help someone who finds it difficult to organize a spoken or written question identify what they want to say. It can compare the similarities, differences, and possible consequences of choices one step at a time. When one explanation is not enough, it may repeat the idea using different words and examples or help someone check information about schedules, travel, money, and safety.

Making information easier to read and understand is not the same as shortening it indiscriminately. Inclusion Europe’s easy-to-read standards call for words, sentences, structures, and forms of presentation that the intended reader can understand.3 If AI is used to produce easy information, the result should be checked with people with intellectual disabilities. A sentence may look simple while still being inaccurate or missing an important condition.

These are possible directions for support. They are not evidence that every current AI product works safely or effectively for every person with an intellectual disability. Language, experience, interests, sensory environments, and support needs differ. AAIDD likewise emphasizes considering a person’s limitations and strengths together and providing appropriate, individualized support over time.4

Support is not substitute decision-making

Explaining options in understandable language supports a decision. Letting AI, a family member, or an institution choose on someone’s behalf merely for convenience takes the decision away.

For example, AI may help explain the schedules and differences between two educational programs and help a person ask about the conditions that matter to them. But if it uses past records to decide, “This is the right option for this person,” and hides other possibilities, it becomes a form of control rather than support.

The Convention on the Rights of Persons with Disabilities states that people with disabilities enjoy legal capacity on an equal basis with others and should have access to the support they may require in exercising it.1 The central principle is to support a person to decide, not to take over the decision.

An AI answer can sound natural and confident while being wrong. It may omit important information or misread a person and their situation. Decisions with major consequences in education, welfare, health, and law require verifiable sources and review by accountable people. AI can assist with an explanation; it cannot assume responsibility.

Five principles for beneficial AI

1. Build it with the people who will use it

People with intellectual disabilities should not be the last users invited to test a finished technology. They should participate when important problems are selected, when language and interfaces are designed, and when real-world outcomes are evaluated. Convenience imagined without the people concerned may become inconvenience or risk for them.

2. Make it understandable and operable

Products need clear sentences, visually distinct controls, sufficient time, and an obvious way to undo an action. They should support multiple ways to provide and receive information, including speech, writing, images, and read-aloud output. Accessibility should be part of the basic structure from the beginning, not an optional feature added later.

3. Allow support to be adjusted to the person

People described by the same diagnostic term understand information in different ways and require different support. The length of an explanation, vocabulary, examples, repetition, and amount of information should be adjustable. Personalization, however, must not be used to predict low potential or close off new experiences.

4. Protect privacy and dignity

Disability information, conversation records, and daily habits are highly sensitive. A service should not collect information it does not need, and it should explain what it uses and why in an understandable way. Safety must not become a justification for continuous surveillance or the reuse of data without consent.

5. Preserve human choice and accountability

People should be able to see what information informed an AI answer, correct an error, refuse the technology, and ask a person for help. Responsibility for important decisions remains with the people and institutions that build and use the service, not with AI.

UNESCO’s Recommendation on the Ethics of Artificial Intelligence identifies human rights and dignity, diversity and inclusion, fairness and non-discrimination, privacy, human oversight, and accountability as principles that should apply throughout the AI life cycle.2 The recommendation does not directly endorse Nalja Books’ declaration. It does show that any serious account of AI’s benefits must consider rights and risks together.

Possibility and risk

I believe AI can open new possibilities for people with intellectual disabilities. It may explain difficult information in another way, help organize something a person wants to say, and support repeated practice at an individual’s pace. It may expand opportunities for learning and participation for people who have not received enough materials or support.

Believing in that possibility is not the same as ignoring risk. We must also recognize that:

AI can state incorrect information with confidence.
Training data and evaluation may not adequately represent the language and experience of people with intellectual disabilities.
Protection and efficiency can be used to justify surveillance and behavioral control.
Disability information and private conversations may be collected or exposed.
A service may prioritize the convenience of families and institutions over the needs of the person using it.
AI may be used as a cheap substitute for adequate human support.

Benefit is not achieved simply because the positive side of a scale outweighs the negative side. We can call AI beneficial only when serious risks are reduced, people can understand, refuse, and correct it, and they retain agency in their own lives.

Nalja Books’ commitment

Nalja Books will not treat people with intellectual disabilities as users to consider after a technology is complete. We will regard them as people who should benefit from AI from the beginning.

As we evaluate AI-related content, educational materials, and products, we will repeatedly ask:

Does this technology help people with intellectual disabilities understand more clearly?
Does it expand their ability to express intentions and make choices?
Does it protect their dignity, privacy, and agency?
Does it provide a direct benefit to them rather than stopping at convenience for families, teachers, or support staff?

As a mother, I do not want new technology to decide my child’s life. I want it to help my child understand, express, and choose more fully. I will hold the same standard as the publisher of Nalja Books.