Washington D.C.— Today, House Small Business Committee Ranking Member Nydia M. Velázquez (D-NY) led seven Small Business Committee Democrats in sending a letter to the White House Office of Management and Budget and the Department of Commerce calling on the agencies to study the effectiveness of AI chatbots in serving small businesses interacting with government agencies.
“We are excited to see LLM technology mature and revolutionize the way we as public officials engage and serve entrepreneurs and other constituents,” wrote the lawmakers. “However, as government agencies begin deploying these tools for widespread public use, it is important that they remain reliable and helpful, and minimize implicit biases.”
Recently state and local government have begun to deploy AI large-language model (LLM) chatbots to provide services and advice to small business owners. In October, New York City launched the MyCity Chatbot to answer questions about operating small businesses, including guidance on City small business services and resources, compliance with regulatory requirements, and small business planning.
In the letter, the lawmakers recognize the potential of AI tools have to streamline entrepreneurs’ access to information and resources but raise concerns about the accuracy and reliability of information provided to small businesses by AI chatbots.
“The MyCity chatbot, and others that follow, have great potential to streamline Americans’ navigation of the byzantine layers of government, and with more language features, to make it more accessible to immigrant Americans,” wrote the representatives. “With that said, great uncertainty surrounds LLMs’ ability to share information reliably and accurately, especially of the technical kind needed by small businesses to start up, interact with government agencies, and navigate rules and regulations.”
In their letter, the representatives call on the Biden administration to incorporate the following considerations in upcoming research and guidance on AI:
- What constitutes an acceptable level of information inaccuracies and omissions (both in frequency and significance) in government chatbot outputs
- How to best protect against these inaccuracies and omissions, as well as best practices for implementation
- How to best evaluate, measure, and address bias in chatbot outputs and decision making
- Which questions government chatbots should deem unanswerable and needing a alternative human operator to answer
- What metrics agencies should use to evaluate chatbot output quality and user satisfaction
- How to best optimize ease-of-use and accessibility of small business service chatbots for entrepreneurs
- Potential remedies that agencies serving small businesses can use if any chatbots inaccurately counsel an entrepreneur
In addition to Velázquez, the letter was signed by Reps. Morgan McGarvey (D-KY), Shri Thanedar (D-MI), Dean Phillips (D-MN), Judy Chu (D-CA), Kweisi Mfume (D-MD), Greg Landsman (D-OH), and Hillary Scholten (D-MI).
For a full copy of the letter, click here.
###
