AI Bill of Rights
This week, the White House published the Blueprint for an AI Bill of Rights. To directly quote the document, the blueprint “is a set of five principles and associated practices to help guide the design, use, and deployment of automated systems to protect the rights of the American public in the age of artificial intelligence.” With the potential to affect every sector of our lives, the need for AI to have clear boundaries around that influence has been apparent since the first AI-driven robots started communicating with humans in labs through the keyboard. To that end, the White House assembled a set of experts from the public and private sectors to come up with a way to help the current Administration carry out their mission to uncover and systematically rule out inequities in the use of technology, whether deliberate or unintentional, to make sure opportunities are available to everyone and is safe to use by everyone.
The White House Office of Science and Technology Policy has created five principles to guide the design of AI software (and in turn, devices that interface with it) that will affect people and their opportunities, access to critical services and locations, and rights. These principles are:
Safe and Effective Systems - “You should be protected from unsafe or ineffective systems”
A well-designed and monitored system mitigates risk and means that data created by that system that pertains to you will be accurate and safely stored and used. This principle was created in response to the prevalence of malicious software and systems.
Algorithmic Discrimation Protections - “You should not face discrimination by algorithms and systems should be used and designed in an equitable way.”
Algorithms shouldn’t create or amplify inequities to the detriment of anyone or any group of people. Even with thorough testing, the developers may not even realize there is bias built into the model until hundreds of iterations later, when a graph shows that the output data is skewed in a different direction than the original data. This principle will ensure that developers constantly work to eliminate biases.
Data Privacy - “You should be protected from abusive data practices via built-in protections and you should have agency over how data about you is used.”
Your privacy should be written into the code and designed into the architecture of systems where your private data resides. You should be able to retrieve it at will, and control who has access to it.
Notice and Explanation - “You should know that an automated system is being used and understand how and why it contributes to outcomes that impact you.“ It should be transparent that you are interfacing with a computerized system making determinations that will affect you and your family. Social media browsing choices, or any web tracking, being fed back to marketing companies is one example.
Human Alternatives, Fallback and Consideration - “You should be able to opt out, where appropriate, and have access to a person who can quickly consider and remedy problems you encounter.”
If you decide you want to interface with a human to help, the AI you’ve been talking to will disengage and connect you to a human - no ‘deep fake’ VR or AI audio. If you want to opt-out completely, that option should always be available. Additionally, if an AI system fails, it should immediately fall back to a human-interactive protocol. Finally, it is acknowledged that there are many situations where human oversight is absolutely critical, such as in healthcare.
If you click on any one of these principles, their webpage has detailed descriptions stating why the principle is important (including examples), what will be expected of automated systems moving forward, and how the principle can be applied. The framework this Bill of Rights applies to includes automated systems that have the potential to meaningfully impact the American public’s rights, opportunities, and access to critical resources and services. This overlap is deliberate, to cover the wide scope of American life taken as a whole instead of compartmentalized.
You can download a PDF of the entire blueprint, including appendices, at this link.