The Rise of AI Regulation

Breaking down how states are considering regulating AI

Author

Greg Miller
July 15, 2024

Welcome to my first newsletter!

I plan to use this to analyze policy trends across the United States. I hope these are interesting and informative.

This week, I am looking into nationwide trends of proposed AI regulation.

In eras of technological innovation, policy responses shape the future. A.I. always had the potential to disrupt our society. Still, it took ChatGPT’s seemingly magical abilities to raise the flags to citizens and policymakers alike on the true potential (I’m not an exception—my “aha” moment with ChatGPT led me to leave my government job to build AI software).

ChatGPT was released in November of 2022. The following year, state legislation mentioning “artificial intelligence” in the title more than tripled. So far, 2024 has surpassed last year’s number with 67 bills.

A quick analysis of legislation since 2017

California’s SB1047 is one example of a bill gaining significant attention across the tech world (even soliciting a response from Y Combinator). SB1047 would prohibit developers of large AI models (those that cost greater than $100 million of computing power to train) from commercializing the models if there is an “unreasonable risk” that the model will cause “critical harm.”

California has at least 20 different bills with Artificial Intelligence in the title and 64 with AI in the bill text.  

States across the US are considering AI legislation. Below, I classify ways states are considering regulating AI with associated pieces of legislation being considered (excluding CA SB1047).

I focus primarily on AI bills that may impact the private commercialization of AI—there are plenty of bills regarding government use of AI—with no prejudice on the merits of the bills themselves. This list introduces how states consider regulating AI and is not intended to be holistic.

Transparency and Documentation

  1. Training Data Transparency

    • California (AB2013): Requires developers to disclose detailed information about the datasets used for training AI models.

  2. Disclosure of AI Use

    • New York (A10103, S9450): Mandates warnings on generative AI systems about potential inaccuracies.

    • Massachusetts (HD4788), California (AB3050): Require clear disclosures on AI-generated content.

  3. Annual Reporting and Inventories

    • Louisiana (SB118): Requires registration of foundation models.

    • New York (A8129): Requires public reporting of AI systems by developers in plain language.

Ethical and Equitable AI

  1. Algorithmic Discrimination

    • New York (A8129): Protects against algorithmic discrimination. Requires developers’ proactive action to protect from discrimination.

    • Massachusetts (S31): Ensures AI models do not engage in discrimination or bias. Requires registration with the Attorney General.

      The intro to this bill is apt: “An Act drafted with the help of ChatGPT to regulate generative artificial intelligence models like

      ChatGPT.”

  2. Consumer Protection

    • Utah (SB 149): Establishes liability for AI misuse.

    • California (SB970): Requires consumer warnings for AI-generated content.

    • California (AB3050): Prohibits unauthorized deepfakes.

    • New York (A10103, S9450): Establishes penalties for non-compliance with disclosure requirements.

Accountability and Governance

  1. Creation of Oversight Bodies

    • Utah (SB 149), New York (S9104), California (SB313), Connecticut (SB00002): Establish Offices of Artificial Intelligence and/or Chief AI Officers.

  2. Ethical Guidelines and Standards

    • New York (A8195): Establishes an ethical code of conduct for AI.

    • Massachusetts (S31): Requires regular risk assessments and ethical guidelines.

Impact Assessments and Risk Management

  1. Impact Assessments

    • Connecticut (SB00002), Colorado (SB24-205), California (AB2885): Mandate documentation and impact assessments for high-risk AI systems.

Data Privacy and Security

  1. Data Protection

    • Massachusetts (S2770), Utah (SB149): Require clear affirmative consent for data with opt-out rights.

If you have any questions about AI legislation nationwide, please don't hesitate to reach out. As a co-founder of Aidy, I leverage our platform for in-depth research and analysis of these bills. Looking ahead, what topics should I prioritize in the coming weeks?

If you found this article interesting, consider subscribing to this newsletter.