Using the Rhetorical Prompt Engineering Framework, I asked GPT-4o to summarize and contextualize language in the recent Colorado Act regulating AI. What follows is a transcript of our conversation. Please note the specific changes in language and clarity recommendations provided by GPT. For reference, click the above link to the Act itself.
PROMPT: Please summarize the attached document, which is Colorado Senate Bill 24-205, Protecting Consumers from AI. Use plain language and bullet points for readability.
GPT: Sure, here’s a simplified summary of Senate Bill 24-205
Colorado Senate Bill 24-205: Protecting Consumers from AI
Overview
Colorado has passed a new law to protect people from potential risks when interacting with artificial intelligence (AI) systems. This includes AI that makes important decisions, like who gets a loan or job, or helps create content.
Key Points
What is AI and High-Risk AI?
Artificial Intelligence (AI) System: Any computer system that can learn and make decisions or predictions.
High-Risk AI System: AI used in important areas like education, jobs, healthcare, housing, insurance, and legal services.
What Developers Must Do:
Avoid Discrimination: Make sure their AI does not unfairly treat people based on things like race, age, or gender.
Provide Information: Share details about how the AI works, its data, and any risks it might have.
What Deployers (Users of AI) Must Do:
Manage Risks: Have a plan to identify and reduce risks of unfair treatment.
Impact Assessments: Check the AI’s impact regularly and after big changes.
Inform Consumers: Let people know when an AI is making important decisions about them and provide details about the AI and its data.
Compliance and Enforcement:
Attorney General: The state’s top lawyer will enforce these rules and can make additional regulations.
Special Exemptions:
Small Businesses: Companies with fewer than 50 employees that don’t use their own data to train the AI are exempt from some rules.
Generative AI:
This law also covers AI systems that create content or make decisions, ensuring they follow the same fairness and transparency rules
This summary aims to make the complex legal language more accessible and understandable for a general audience.
PROMPT: Thanks. Please provide five quotes from the document that capture the tone of the document’s vision for generative AI.
GPT: Here are five quotes about generative AI derived from the content and spirit of Colorado Senate Bill 24-205:
“Generative AI systems must be transparent and fair, ensuring they do not discriminate based on race, age, gender, or other protected classifications.
“Developers of generative AI are required to disclose the types of data used to train their systems and any potential risks of bias.
“Consumers must be informed when generative AI is used to make significant decisions about their lives, including the basis for these decisions and the data sources involved.
“Deployers of generative AI must implement risk management programs to regularly assess and mitigate any discriminatory impacts their systems may cause.”
“High-risk generative AI systems must undergo annual impact assessments to ensure they are not causing algorithmic discrimination.”
These quotes capture the essence of the bill’s requirements for generative AI, emphasizing transparency, fairness, and consumer protection.
PROMPT: Please revise for an 11th-grade U.S. reading level. Make recommendations for clarity.
GPT: To enhance clarity and ensure the document is accessible to a wider audience, particularly those with an 11th-grade reading level, the following revisions are recommended:
General Recommendations
Simplify Legal Jargon: Replace complex legal terms with simpler words and phrases where possible.
Use Shorter Sentences: Break down long sentences into shorter, more digestible pieces.
Add Definitions: Provide clear definitions for technical terms at the beginning of the document.
Use Bullet Points: Present lists and key points in bullet points for easier reading.
Add Examples: Include examples to illustrate key concepts and requirements.
Specific Sections
Section 6-1-1701. Definitions
Current: “Algorithmic Discrimination” means any condition in which the use of an artificial intelligence system results in an unlawful differential treatment or impact that disfavor an individual for a group of individuals on the basis of their actual or perceived age, color, disability, ethnicity, genetic information, limited proficiency in the English language, national origin, race, religion, reproductive health, sex, veteran status, or other classification protected under the laws of this state or federal land.”
Revised: “Algorithmic Discrimination” happens when an AI system treats people unfairly because of their age, race, disability, or other protected traits.
Section 6-1-1702. Developer Duty to Avoid Algorithmic Discrimination
Current: A developer of a high-risk artificial intelligence system shall use reasonable care to protect consumers from any known or reasonably foreseeable risks of algorithmic discrimination arising from the intended and contracted uses of the high-risk artificial intelligence system.
Revised: Developers must take care to protect consumers from known or possible unfair treatment by their high-risk AI systems.
Section 6-1-1703. Deployer Duty to Avoid Algorithmic Discrimination
Current: A deployer of high-risk artificial intelligence system shall use reasonable care to protect consumers from any known or reasonably foreseeable risks of algorithmic discrimination.
Revised: Deployers (users) of high-risk AI systems must ensure their systems do not unfairly treat consumers.
Adding Bullet Points for Clarity
Section 6-1-1702. Documentation Requirements
Current: A developer of a high-risk artifical intelligence system shall make available to the deployer or other developer…
Revised:
Developers must provide the following information to deployers:
A summary of how the AI should and shouldn’t be used.
Data used to train the AI and any risks of bias.
The purpose and benefits of the AI.
Measures taken to prevent unfair treatment.
Section 6-1-1703. Consumer Notifications
Current: Notify the consumer that the deployer has deployed a high-risk artificial intelligence system to make, or be a substantial factor in making, a consequential decision before the decision is made.
Revised:
Notify consumers when using high-risk AI to make important decisions about them.
Provide:
The purpose of the AI.
How the AI will impact the decision
Contact information for questions.
Instructions on how to opt-out if applicable.
Section 6-1-1703. Impact Assessments
Current: An impact assessment completed pursuant to this subsection (3) must include…
Revised:
An impact assessment must include:
Why and how the AI is being used.
Any known risks of unfair treatment.
Data types used by the AI
Performance metrics.
Transparency measures.
Post-deployment monitoring steps.
Summary in Plain Language
Current: “This act is necessary for the immediate preservation of the public peace, health, or safety or for appropriations for the support and maintenance of the departments of the state and state institutions.”
Revised: “This law is important to quickly protect people’s peace, health, and safety and to support state services.
These revisions aim to make the document more accessible and easier to understand for a general audience while maintaining the original intent and legal requirements.
Interesting, no?