Deep Dive Breakout 2: Navigating the Choppy Waters of AI

By now, most of us have likely dabbled in using AI tools, but navigating practical ways to incorporate AI into our work can be intimidating. How have practitioners started responsibly utilizing AI at our workplaces and for advocacy campaigns? What are some ways that we’ve seen AI impact our communication and relationships with lawmakers and our communities?

Laura Brigandi, Public Affairs Council
Zachary Keita, National Association for Music Education
James Servino, Human Rights Campaign [OFF THE RECORD]
Mike Panetta, Beekeeper Group (moderator)

Notes:

Learning Objective 1: What is AI? How have practitioners used it in their workplaces?

• Narrow/”weak” AI
  • Image generation, responds to prompts (Siri, Alexa, ChatGPT)
• General/”strong” AI
  • Machine learning – indistinguishable from the human mind
  • No currently available forms of AI fall into this category
• What is it?
  • No 100% agreed-upon definition, but anything when a machine is doing something that requires human intelligence
  • We are only a subset of a subset when looking at ChatGPT-style generative AI
  • Improve over time without human intervention
• Use in Public Affairs Survey
  • April – July 2024, used at June 2024 DMAS, 48 respondents
  • Wide-ranging types of organizations, as well as involvement
  • Small subset with AI integrated into proprietary
  • More (29%) use outside vendors with outside AI integration
  • 56% use tools like ChatGPT/DALL-E, which are not integrated into the organization system
  • 31% use no AI tools – some not allowed by organization structure/IT
• Popular uses among respondents
  • Brainstorming 54%
  • Proofreading/editing 52%
  • Content generation 49%
  • Legislative Bill Summary 23%
  • Others: Data analysis, tracking legislation, data visualization, notetaking, stakeholder mapping, event planning
• Use case – Policy Analysis
  • Title II ESEA – Ask generative AI to summarize
  • Be sure to double-check and verify

Learning Objective 2: Takeaways around the use of AI on the Hill, how it’s regarded and used

• Timeline of use on the Hill
  • April 2023 – ChatGPT Plus licenses – intern developed a tool to allow easier drafting of congressional memos – automate and simplify (later bought by FiscalNote)
  • Later cracked down – limited only to a paid version
  • Senate issues guidance allowing the use of ChatGPT+, Claude
  • Used to respond to constituent correspondence, summarize/jumpstart communications
  • January 2024 – The app (Turbo Legi) acquired
  • March 2024 – House banned Microsoft Copilot due to data security concerns
    – Q: What caused the concern with Copilot? Not a data breach – determined that the way Copilot is integrated raised concerns about constituent data/confidentiality – but Senate allows it
• Can lawmakers detect AI use?
  • Not according to the March 2023 Cornell University study
  • Examined responses written by people vs. responses written by AI – 2% difference
  • Perception changing – look at any email campaign suspiciously as though it is written by AI. Not counted as seriourly.
  • Technology that Hill is using to receive communications can detect if it had been generated by AI? Some might have preferences/detection thresholds for how much variation there is in the message.
• Cautionary tales
  • AI-generated voicemails from deceased gun violence victims. Giving a voice to the victims – some members thought this was a compelling message, others called it fraudulent/creepy.
    • Original use of AI, but a very risky move – was it worth it? Did it move the needle?
  • Pro-ceasefire emails – clogged inboxes of Congressional staff. AI was less problematic than automation. Put in information, generate multiple messages multiple times per day, and email them automatically to legislators.
• AI From 107th Congress to Today
  • Pre-2001: AI to automate and assist with casework
  • House Office of the Clerk Comparative Print Suite – see how amendments/laws changed with new language.
  • GAO using custom Project Galileo began with apps Sia, Wordworkr, Titan
• Reactive Fear – OTR
  • The new iteration of past strategies – pre-authorized mailgrams in the 80s, today it is AI
  • AI eliminating barriers – offices weighing the amount of effort required in constituent communication.