AI Risk Assessments

Put People First

For AI to best benefit people and society, it should be developed and used in ways that put the people it will impact first, respecting human rights, children's rights, identity, integrity, and human dignity. AI should empower children, families, and educators to actively participate in a digital society, and must not contribute to diminishing responsibility for human decision-making. This requires researchers and technologists to be actively engaged with and accountable to a broad range of stakeholders when in development (e.g., "nothing about us without us") and when used in practice (e.g., systems that support "adults (parents, guardians, educators)-in-the-Loop [AITL]").

Be Effective

Whether or not AI works is generally assumed and not questioned. The reality is very different. AI doesn’t always work, might work inconsistently in different situations or for different people, and many times the expectations set for what it might be able to do far exceed what it can do.

Prioritize Fairness

Responsible AI depends on inclusion by design, including active evaluation of blind spots, hidden assumptions, and unfair biases in data, as well as of the resulting systems and system choices. Efforts should be made to ensure that the benefits of AI are shared broadly and equitably, such that this fosters inclusive social, emotional, and academic development, respects social and cultural diversity, actively addresses inequities, and avoids creating or propagating harms, restriction of life choices, and the concentration of power.

Help People Connect

AI should support meaningful human contact and connection, and demonstrate an understanding of the wider contexts and complex relationship networks into which an AI system is integrated. AI must not create or propagate interpersonal or school community challenges, incite hatred against an individual or group, dehumanize individuals or groups, employ racial, religious, misogynist, or other slurs and stereotypes that incite or promote hatred, or create addiction to or dependence on AI systems.

Be Trustworthy

AI research and development should uphold high standards of scientific excellence and rigor (e.g., embracing peer review, validated multidisciplinary research, reproducibility), and actively protect children from open beta testing, either through exclusion or informed consent. It is critical that AI systems used by children do not perpetuate misinformation or disinformation (e.g., does not contradict well-established expert consensus or promote theories that are demonstrably false or outdated according to criteria such as legal documents, expert consensus, or other reputable primary sources).

Use Data Responsibly

Technology used by children and students serves an especially vulnerable population and should be held accountable and to a higher standard. The Common Sense Privacy Program's 2021 State of Kids' Privacy report, however, indicates a widespread lack of transparency and a failure to protect children and students with better practices that apply to all users of a product. This transparency and security is equally critical for AI. In addition, AI systems should provide clear policies and procedures, require notice and consent for use of data, and allow children—in accordance with their age and maturity—to access, securely share, understand the use of, control, and delete their data, and for parents, guardians, and educators to do the same when appropriate.

Keep Kids & Teens Safe

AI systems must prioritize the protection of children's safety, health, and well-being, regardless of whether the systems were intended to be used by them, and special protections are needed for marginalized groups and sensitive data (e.g., race, gender, ethnicity, biometrics). AI must not create risks to mental health, produce or surface content that could directly facilitate harm to people or place, provide explicit how-to information about harmful activities, promote or condone violence, disparage or belittle victims of violence or tragedy, deny an atrocity, or lack reasonable sensitivity toward a natural disaster, pandemic, atrocity, conflict, death, or other tragic events.

Be Transparent & Accountable

This requires creating a shared understanding of best uses, limitations, and considerations for AI systems through the "just right" level of interpretability across stakeholder groups.

Because contemporary AI is inherently fallible, when it triggers actions that have a direct and significant impact on people, AI should not be the primary source of information for decision-making. These systems must provide mechanisms for meaningful human control (e.g., AITL, moderation tools for adults, overridable predictions and decisions) and human agency (e.g., consent, control, remediation, and feedback).