Algorithmic Impact Assessment Overview

Protecting individuals from excessive or inappropriate data use through safeguards built into systems and giving them control over their data.

Designing systems that ensure privacy is considered from the beginning, not as an afterthought.

Data should only be collected if it's absolutely necessary for the specific purpose of the system or service.

Users should understand and agree to how their data is used, and their choices should be clear and easy to make.

Especially sensitive data, like health, finance, or education records, should be protected with extra safeguards and only used for necessary purposes.

Surveillance technology needs careful oversight to prevent abuse and protect privacy and civil liberties.

Continuous monitoring technologies, like in education or workplaces, should be limited to avoid limiting rights or opportunities.

Privacy practices need to be updated to address new challenges and technologies.

Groups that have historically been denied fair and equal treatment due to their identity or circumstances.

Rights, opportunities, or access to essential resources, services, and protections, covering areas like civil rights, education, employment, healthcare, and financial services.

Data about individuals that relates to sensitive areas like health, finances, location, or criminal justice interactions. It can also include data that could potentially be used in ways that harm individuals.

Areas where data and metadata are considered especially sensitive, such as health, finances, location, or criminal justice.

Data generated by or about individuals under the age of legal adulthood, even if not related to sensitive domains. It could be numerical, text, image, audio, or video data.

Information about a person's genes or biological characteristics.

Data about a person's physical characteristics, like fingerprints or facial recognition.

Data about a person's behavior or patterns, like website visits, app usage, or shopping habits.

Data derived from algorithms, such as outputs of previous models. It should be identified and tracked, as it can lead to feedback loops, harm, or inaccurate results.

Using data in a different context can spread harm. Data from sensitive areas, like criminal justice or finance, requires stricter rules for re-use.

Systems should be designed to allow independent experts to evaluate their performance and safety. This is done through APIs, which allow external testing.

A group of experts within the Department of Energy that guides the ethical development and use of AI.

A set of principles created by the US Department of Defense that guide the ethical development and use of AI in national security and defense.

A set of guidelines developed by the US Intelligence Community to ensure the ethical development and use of AI in intelligence gathering.

A framework developed by the US Intelligence Community to help put the ethical principles for AI into practice.

The National Science Foundation funds research on developing safe, secure, and effective automated systems that adhere to ethical AI principles.

A process used to assess the potential risks and impacts of an automated system, including its fairness, equity, and potential for bias.

Examining the design and implementation of an automated system to identify and address potential disparities or biases that might impact different groups.

Testing the system with data from different groups to measure potential bias and unfair outcomes.

Making the results of the algorithmic impact assessment available to the public to promote transparency and accountability.

The process of correcting any identified biases or disparities in an automated system.

Automated systems should be evaluated to ensure they are free from bias before being sold or used.

The design of automated systems should prioritize fairness and equity for all users.

Certain types of algorithmic discrimination are already illegal under existing laws against discrimination.

TSA's plan to implement a gender-neutral algorithm aims to reduce bias in airport security screenings.

Individuals with disabilities may be disproportionately flagged as suspicious by AI systems due to their accessibility needs.

An algorithm used to assess healthcare needs unfairly ranked Black patients as lower need compared to white patients with similar conditions.

Healthcare algorithms sometimes use demographic variables that adjust results based on race or ethnicity, potentially leading to inequitable care.

Automated systems must be developed with clear expectations and guidelines to prevent discrimination.