Constituent Engagement Related to Use of AI

Instrument 2: Focus groups with public benefit participants

Intro text

Good morning/afternoon and thank you for joining us.

We are from a research organization called Mathematica and we will be guiding the conversation today. The purpose of this focus group is to get your perspectives on how public benefit programs may consider using artificial intelligence, or AI.

I’ll explain a little more about what we mean by AI in a minute. We are hosting this session on behalf of the Administration for Children and Families. The Administration for Children and Families, or ACF, is a federal agency that funds a number of public benefit and other programs, including Temporary Assistance for Needy Families, Head Start and Early Head Start, teen pregnancy prevention programs, and many others.

The Administration for Children and Families is interested in hearing the perspectives of organizations and people who could be affected by the use of AI in public programs. This is one of several focus groups being conducted with different groups. ACF will compile all the feedback received from the sessions to help it prioritize issuing guidance on AI use. ACF might also use the feedback to develop resources to support responsible AI use among the organizations it funds.

Before we start our conversation, I just want to let you know that nothing you say here will have an impact on any public benefits you receive, now or in the future. And it won’t affect your ability to participate in any programs now or in the future.

There are no right or wrong answers to the questions we’ll ask today, ACF is really interested in hearing your opinion, whatever it is. Information collected will be kept private to the extent permitted by law. Your participation is voluntary, and you can participate in this session to the extent you feel comfortable, you do not need to answer any questions you don’t want to. There are no risks associated with participating in this session.

I’m going to read a statement for your awareness: An agency may not conduct or sponsor, and a person is not required to respond to, a collection of information unless it displays a currently valid OMB control number. The OMB number and expiration date for this collection are OMB #: ADD, Expiration: ADD.

We would like to record this conversation to ensure we capture the key points. We will also transcribe it and share with people at ACF. In the transcript, we will remove names and another other personal identifiers to protect your privacy.

Background

First let’s set the stage so we all know what we mean when we say “AI.” In broad terms, AI is technology that enables computers and machines to simulate human intelligence and problem-solving abilities. When we say “AI,” we aren’t referring to far-fetched future technologies you might see in movies, but rather technology that currently exists and is accessible to most people. There are lots of different kinds of AI, and you have probably interacted with some of them.

Examples are talking to digital assistants, like Alexa, Siri, or ‘Hey Google.” You ask them a question and they return an answer to you. Other examples include personalized recommendations and content offered up by Netflix, Instagram, YouTube, or Amazon.

As a warm-up we wanted to learn about what types of AI you may already have experience using. We will use the polling feature to conduct a quick pulse check. [Provide additional instructions for using poll feature once platform is confirmed.]

Which of the following types of AI have you used or interacted with? Select all that apply.

1. Digital assistants (e.g., Alexa, Siri, Google Assistant)

2. Personalized recommendations (e.g., Netflix, YouTube, Amazon)

3. Social media algorithms (e.g., Facebook, Instagram, Twitter)

4. Language translation tools (e.g., Google Translate, Microsoft Translator)

5. Facial recognition (e.g., for phone unlocking, tagging photos)

6. Chatbots for customer service or support

7. Navigation apps with real-time traffic updates (e.g., Google Maps, Waze)

8. Voice-to-text dictation (e.g., for sending messages or writing emails)

9. None of the above

10. Other (please specify)

For our conversation today, we want to think about that sort of activity in the context of public benefits and services.

I’m going to present a few scenarios and I’d love to hear your thoughts about how you might respond. There are no right or wrong answers here—we are just interested in what you have to say. For each scenario, take a moment to consider it and imagine that you are in each situation, even if the specifics do not apply to you. When you are ready to share, raise your hand or, if you prefer, type your thoughts into the chat.

Scenario 1

Imagine that your child attends a public early childhood education or child care center that uses virtual assistants like Alexa or Google Home. These work by capturing short snippets of audio when activated. The staff use these virtual assistants to help manage the physical environment (like the temperature and lighting or playing music) or as a teaching tool (like to look up facts about bugs).

• What are some positive aspects of this scenario? What might work really well?

• What are some aspects of this scenario that would raise concerns for you or might not be so good?

• Overall, would you be comfortable with this use of AI in a child’s classroom? How would you describe your comfort or discomfort?

  • Probes: enthusiasm, fear, nervousness, no strong feeling

Scenario 2

Imagine that you are applying online for assistance from the government. For example, you are applying for cash assistance or SNAP benefits. Immediately after completing the application and uploading required documents, you receive an AI-generated notice that you are eligible and will be auto-enrolled into the program you applied for. This means you will begin to receive the public benefit as soon as you complete needed requirements. There is a note that there is a chance your eligibility may be re-determined upon later review from a human.

• How does this compare with your past experiences applying for assistance from the government?

• Would this approach feel better, worse, or the same?

• Talk a bit about why you feel this way.

Scenario 3

Imagine that you are enrolled in a publicly funded program, like TANF or SNAP, and you are assigned a new case manager. Your new case manager receives an AI-generated summary of the case notes entered by your prior case managers in the program, in addition to all of the detailed notes. Lacking time to read the detailed notes, they read only the summary before meeting with you.

• What are some positive aspects of this scenario? What might work really well?

• What are some aspects of this scenario that would raise concerns for you or might not be so good?

• Overall, would you be comfortable with this use of AI? How would you describe your comfort or discomfort?

  • Probes: enthusiasm, fear, nervousness, no strong feeling

Scenario 4

Imagine that you call a hotline to seek immediate help because either you or someone you know is in a dangerous situation. The phone call is recorded as usual. The hotline staff who speaks with you gets real-time, AI-generated recommendations including resources that you can access in your community and whether to dispatch first responders.

• What are some positive aspects of this scenario? What might work really well?

• What are some aspects of this scenario that would raise concerns for you or might not be so good?

• Overall, would you be comfortable with this use of AI? How would you describe your comfort or discomfort?

  • Probes: enthusiasm, fear, nervousness, no strong feeling

Scenario 5

Imagine that you are enrolled in a program to learn new job skills. As part of the program, you can work with a human job skills coach for up to a certain number of hours a month. The human coach could help you with things like resume building and doing practice interviews. You can also use an AI job skills coach for additional practice as much as you'd like. For example, the AI coach could ask you common job interview questions and provide you with feedback about your responses, much like the human would do.

• How likely would you be to work with the AI job skills coach?

• What would you need to know before you started working with the AI coach?

• If you started working with the AI coach, what would it need to do to convince you to keep working with it?

Quick feedback scenarios

Before we wrap up, we’re going to give you a few more scenarios for quick feedback. After I share each scenario, please show with a thumbs up/down or a “yes” or “no” typed into the chat whether you think AI would be useful in these scenarios.

1. Imagine that you are attending a program orientation where the main language being spoken is not your native language. AI provides live translation so you can interact with social service providers in your preferred language. Indicate with a thumbs up if you like this idea. Or enter “yes” in the chat. If you don’t like it, give it a thumbs down or a “no” in the chat.

2. Imagine that you are enrolled in a public benefit program, like SNAP or Head Start for your kids. The staff use AI to assist with administrative tasks, like identifying duplicate documents, matching signatures, taking notes in meetings, and filing electronic documents.

3. Imagine that you are applying for public benefits, like cash assistance or Medicaid. AI provides recommendations to staff about whether you are eligible for the benefits you applied for. Staff are trained to use the recommendations as a consideration but make the final decision, not just follow the AI recommendation.

4. Imagine you are participating in a public benefit program that includes case management—that means you regularly talk to a staff person about your needs and goals. Staff ask if you are okay with them using an AI assistant to record your conversation together so that the AI can take notes and recommend services that might meet your needs.

5. An AI program trained on risk assessment scans all the case files for people receiving public benefits and flags files with perceived risks related to mental health or child abuse.

Closing questions

We’ve given you some examples of how AI might be incorporated into public benefit programs.

• Are there other ways in which you think AI, or better technology or automation more generally, would make your experience accessing public benefits and services better? What are they? Feel free to say them out loud or type into the chat.

• What situations related to accessing public benefits and services do you really not want AI to have any part in, compared with working exclusively with people?

• Are there any types of resources/information on the use of AI within public benefits programs that you think would be useful or would like to see developed?

• Any additional thoughts or questions related to the use of AI by public benefit agency or provider staff?

Thank you

Thank you so much for our time today. These insights will be really valuable to public benefit administrators and service providers in the future.