Legislation We Worked On

On September 20, 2023, Virginia Governor Glenn Youngkin announced Executive Directive Number Five (2023) on Artificial Intelligence (AI), acknowledging the critical role the state government must play in ensuring effective oversight of this rapidly evolving technology. The executive order was an important first step to lay the ground work for understanding the opportunities and challenges that AI poses for government modernization, economic growth and education. However, legislative action by the General Assembly is also needed to both incentivize proper use and development, while mitigating harms related to manipulation of data, privacy and bias. We have prepared this brief overview of various legislative options.

Legislative Option: Watermarking and/or Disclaimers

At the federal level, U.S. Senator Pete Ricketts (R-Nebraska) introduced legislation to regulate artificial intelligence. The Advisory for AI-Generated Content Act would require a digital watermark on content generated by Artificial Intelligence. Additionally, it would direct the Federal Trade Commission, the Federal Communications Commission, the Department of Justice, and the Department of Homeland Security to create federal guidelines for enforcing the requirement, including on political advertising.

Several products on the market, political and apolitical, have disclaimers on them already. Cigarettes must warn of their health effects, wine bottles warn of the dangers the beverage has on pregnant women, and political committees must provide a disclaimer.

Legislation should include a required statement, such as: “This communication was generated using artificial intelligence" on all political physical/digital/audio media that has AI involvement. It should also set forth civil penalties for violations of the act.

Voter intimidation online, especially through the use of AI could be dangerous and should be treated as so. There should be a clear and separate provision about misinformation relating to polling places and times, and discouraging people from voting.

Model Legislation: New York’s 2023 NY S 7592 

“Any political communication, regardless of whether such communication is considered a substantial or nominal expenditure, that uses an image or video footage that was generated in whole or in part with the use of artificial intelligence, as defined by the state board of elections, shall be required to disclose that artificial intelligence was used in such communication…”

Model legislation: NY SB6859

While this bill is not specifically targeting political communications, it is more broad, and could be helpful because deep fakes and false information, even if they aren’t directly created by specific political campaigns, can have repercussions on elections. It includes a $1000 fine for first violation and $5000 for subsequent violations. 

Legislative Option: Registry

Legislation could also require businesses and political candidates using AI to disclose that they do so in an Artificial Intelligence Registry. The registry should include IP address, type of software and intent of use. 

Provisions of such a legislation could include: 

• Directing appropriate state agency to establish a registry of businesses operating artificial intelligence systems in the State.  

• Registry would include 

• The name of the business operating artificial intelligence systems; 

• The IP address of the business; 

• The type of code the business is utilizing for artificial intelligence; 

• The intent of the software being utilized;

• The personal information and first and last name of a contact person at the business;

• The address, electronic email address and ten-digit telephone number of the contact person; 

• A signed statement indicating that the business operating an artificial intelligence system has agreed for the state agency to store the business' information on the registry.

Virginia would publicly disclose: 

• The total number of businesses operating artificial intelligence systems registered with the state

• The number of registrants received by the state 

• Names of businesses operating AI systems and categories of purpose

• Expenses incurred by the state

Model legislation from Pennsylvania.

Legislative Option: Establish an agency, task force, commission and/or legislative committees to understand and govern AI capabilities

The importance of legislators and the public fully understanding the evolving capabilities of AI cannot be overstated.  At least 12 states — including Alabama, California, Colorado, Connecticut, Illinois, Louisiana, New Jersey, New York, North Dakota, Texas, Vermont, and Washington — have enacted laws that delegate research obligations to government or government-organized entities to increase institutional knowledge of AI and better understand its possible consequences. Virginia can be at the forefront by creating committees to investigate, and guardrails for generative and other AI in order to better understand how it impacts job markets, society and democratic institutions. AI is rapidly evolving and will only grow in ability and power. Establishing an agency, commission or legislative committees would be an important first step in understanding the impact of AI through firsthand investigation and could precede any AI legislation. Such an effort should include developing an AI Bill of Rights. 

Texas HB 2060 provides a model for such an approach as it “creates an advisory council to study the effects of AI and report findings to the legislature” which will “assess the need for a state code of ethics for artificial intelligence systems in state government.”

Provisions of legislation establishing an agency, commission or legislative committee should encompass:

• developing and establishing policies and procedures concerning the development, procurement, implementation, utilization and ongoing assessment of systems that employ artificial intelligence and are in use by state agencies. 

• developing policies and procedures regarding the governing of the procurement, implementation and ongoing assessment of such systems by state agencies; 

• ensuring that policies and procedures do not result in any unlawful discrimination against any individual or group of individuals, or any unlawful disparate impact on any individual or group of individuals on the basis of any actual or perceived differentiating characteristic, including, but not limited to, age, genetic information, color, ethnicity, race, creed, religion, national origin, ancestry, sex, gender identity or expression, sexual orientation, marital status, familial status, pregnancy, veteran status, disability or lawful source of income…

• Making recommendations concerning, and develop best practices for, the ethical and equitable use of artificial intelligence in state government;

• Assessing options for an AI Bill of Rights and make recommendations concerning the regulation of the use of artificial intelligence in the private sector based, and

• Making recommendations concerning the adoption of other legislation concerning artificial intelligence.

Cost: Similar creations of task forces in Virginia were “not expected to create a material fiscal impact.” JLARC finds that agencies “can absorb the cost of the study in its existing budget.” It is likely that the budget for a task force could come from an existing budget. If a task force requires “additional documentation, preparation of a written recommendation to the Secretary, and an annual report,” these activities could be performed using existing staff resources.” (JLARK, 2019 Fiscal Impact Review) 

Legislative Option: Privacy, Transparency & Accountability

The problems of bias in algorithms and automated system are well documented. Algorithmic transparency, accountability, and fairness are the touchpoints of most AI governance frameworks and regulatory proposals. There is also broad agreement that AI should be “ethical” and a global convergence around five ethical principles (transparency, justice and fairness, non-maleficence, responsibility and privacy). However, there is divergence in how these principles are interpreted and how they should be implemented. Key provisions of state legislation should include: 

• Clear definitions of ethical principles for AI and how they should be implemented

• Boundaries on collection, use, and sharing of personal information, requiring that such process be “necessary and proportionate” to what is necessary to provide a products or service or for other enumerated purposes

• Prohibiting the use of personal information in ways that discriminate

• Requiring transparency and accountability of AI systems and algorithmic assessments 

• Adopt a similar approach to that of the European Union and require a thorough risk assessment for AI systems and categorize AI systems by risk level (e.g., minimal, limited, high). In this approach, not all AI applications would be painted with the same regulatory brush.

Legislative Option: Moratorium on AI for Political Ads in 2024

Based on the bipartisan Protect Elections from Deceptive AI Act, state legislation could prohibit the distribution of deceptive AI-generated audio, images, or video relating to federal candidates in political ads or certain issue ads. The legislation could be time-limited to 2024 in order to allow for reassessment and implementation of other protective measures beyond, while safeguarding the 2024 elections. Such a bill would provide an exception for parodies and content in news broadcasts (would need to be clearly defined).  It should also set forth civil penalties for violations of the act.

Legislative Option: Prevent Data Profiling

To protect the public from data profiling by companies using AI systems, such a bill should include: 

• Creating a complaint filing system, similar to reporting available for consumers currently for bad business practices

• Provide authority to the state Attorney General who may then subpoena information and investigate: “This bill authorizes the attorney general to issue a civil investigative demand if it has reasonable cause to believe that an individual, controller, or processor has engaged in, is engaging in, or is about to engage in a violation of this bill. “

• Include provisions for enforcement, civil penalties and expenses 

Model Legislation: Tennessee SB0073 

• Current VA legislation related to AI and data profiling, “The VCDPA gives consumers the right to opt-out of profiling in furtherance of automated decisions. It also requires a data protection assessment for activities that pose a “heightened risk of harm,” including targeted advertising and some types of profiling.” 2021 Session, patroned by Dunnavant, Boysko, Marsden https://lis.virginia.gov/cgi-bin/legp604.exe?211+ful+SB1392+pdf