Print 🖨

The AI Stakeholder group was convened for its first meeting on May 14, 2024. Rep. Giovanni Capriglione led the discussions noting the goal of passing AI legislation in one year. The meeting covered the basics on AI, a spotlight on 47 responses received from the Request for Information (RFI), and the framework for possible legislation in the next session, including requests for feedback. Rep. Capriglione concluded the discussion by noting there are 384 days until Sine Die of the 89th Legislative Session.

This report is intended to give you an overview of the discussions on the various topics taken up. It is not a verbatim transcript of the discussions but is based upon what was audible or understandable to the observer and the desire to get details out as quickly as possible with few errors or omissions.

Opening Comments

  • Goal: passing comprehensive AI legislation in a year
  • Steps towards this goal
    • This is the first meeting of the AI Stakeholder Group
    • Created an Advisory Council which addresses AI use in state government
      • There have been 3 meetings already
      • They have a form that is now completed that will be sent to state agencies to report inventory of uses in AI
    • House Select Committee on AI/ET (emerging technology) formed
    • Senate Business and Commerce has an interim charge
    • Innovation and Technology Caucus
      • Goal of the IT caucus is to help educate members what AI is and usage

AI Review

  • Reviewed why we want Texas as the global center of AI including:
    • Job Creation included the semiconductor industry was spotlighted
    • Educational opportunities
    • Biotech and medicine
    • Improved QOL
  • We want AI in Texas so that we have a say in how the rest of the world uses AI
  • Threats of AI also spotlighted and include:
  • Erosion of human autonomy – do not want people to blindly follow AI, value of personal decisions is paramount
  • Biases and discrimination, personal information must be protected
  • One of the biggest fears is not being fed lies, but being told the truth and not believing it

Regulatory Framework of AI

  • The reasons for a regulatory framework include:
    • Regulatory certainty brings investments & order
    • Protection of Privacy and Individual Rights
  • Need a holistic framework to guide AI in the right direction
  • Federal government cannot and will not act
  • Without a Texas-wide bill, smaller bills and local legislation will fill the void
  • A Texas act will unify all of Texas under one law
  • Other regulatory examples include: NIST AT RMF which is active but largely optional, Colorado OECDAI, Connecticut and California
  • The framework Texas is working on will adapt as technology changes, laws pass around the world, receiving feedback

RFI Released & Responses

  • Request for information (RFI) was released January 22nd and closed March 15th
  • 47 responses
  • Mix of businesses, trade associations, consumer groups, non-profits, and academics
  • He anticipates there will need to be exceptions like his privacy bill and will need a single enforcement agency
  • Comments made include:
    • Limit regulatory frameworks
    • Want automated decision-making AI
    • Opposition to private rights of action
    • Internal vs third party mechanism
    • Single enforcement agency
    • No licensing mechanism
    • No rulemaking
    • Align to NIST RMF and OECD
    • Leveraging existing laws
    • Different obligations for developers and deployers
    • Responsibility to downstream deployers

Allowances and Restrictions

  • Points out things that will never be allowed:
  • Manipulation of human behavior
  • Exploiting vulnerabilities
  • Social scoring
  • Untargeted Facial recognition- scraping social media for law enforcement
  • Workplace or Educational Emotional Recognition- see what mood people are in
  • Biometric Categorization
  • Child Sexual Abuse Material
  • Things that need more oversight
  • High Risk Automated Decision-Making Systems (HRADMS)
  • Any AI system that makes consequential decisions- airplanes, hospitals, etc.

Requirements for Deployers of HR ADMS

  • Risk Management Policy and Program
  • Impact assessment
  • Consumer Notice
  • Reporting any harm discovered
  • Statement of Intended Uses to deployers
  • Reporting any harm Discovered
  • May submit impact assessment on behalf of a deployer

Foundational Models/GenAI Requirements

Systemic Risk Model

All AI Systems Requirements

  • Either Consent or notice that a consumer is interacting with a AI system
  • Prompt transformation disclosure
  • Letting the user know when the AI system changes a prompt
  • Additional requirements as recommended by the legislature and stakeholders

Other Provisions

  • Allow AI rulemaking for all agencies with jurisdiction
  • Update definitions of software to include AI
  • Disclosure of AI generated content for campaigns
  • Limit the illegal generation of synthetic content
  • Social Media companies and Digital Advertisers must remove any deep fake that they know violates state law

Fostering Innovation

  • No private right to action
  • No rulemaking
  • Minimum 30-day cure period
  • Single enforcer – Rep. Capriglione noted he wanted input here on which one the stakeholders prefer
  • Attorney General- easier because it has been done before
  • AI Commission- they are flexible, more time to fix problems, more experience in subject
  • Sandbox
    • includes exemptions and lower regulations for autonomous manufacturing facilities
    • Capriglione said this will exist like it does for the privacy bill he worked on
  • Small business Exemption
  • Local preemption – a comprehensive bill prohibits smaller patchwork local bills
  • Update workforce training programs with TWC
  • Create new training, credentials, and degree programs through THECB
  • Creation of AI Innovation Program
    • This would be for both private and non-profits, an incubator type program
  • Amend CUBI to allow training of AI models on images and videos
  • Datacenter Incentive Program

Feedback Requested On

  • What is working and what is not
  • Definitions (namely deployer and developer)
  • Impact assessments
  • Requirements for Foundational/GenAI
  • How to make enforcement meaningful, fair, and efficient
  • What would be helpful in a sandbox
  • Synthetic content and deepfakes
  • Workforce and economic development programs
  • Any clarifications
  • Planning issue-specific stakeholder meetings

Timeline

  • Circulate the first draft this summer
  • Issue specific meetings throughout the year
  • Second draft in the fall
  • Bill filed in November
  • There are 384 days until Sine Die