Identifying and Evaluating GenAI Use Cases

What do GenAI solutions look like, and how do I determine if a GenAI tool is a good fit for my department’s problem?

This section shows common or likely use cases for GenAI in the State, for illustrative purposes. It also lists important guiding questions you should ask well before beginning the procurement process for a GenAI tool.

Introduction

GenAI represents a new category of artificial intelligence. While conventional AI can classify data or make predictions, GenAI can create entirely new content. The table in the next subsection summarizes some business needs already identified as potentially suited for GenAI.

As vendors and software companies push GenAI-enhanced features into production for software your department already owns, it will be important for you to understand common use cases of GenAI. By understanding common use cases, your department can successfully meet its requirement to continuously assess the software by placing it in the right context and asking the right questions.

Identifying use cases

Here are some use cases that may apply to your department’s business needs:

Operational need GenAI outcome Common GenAI use cases
Content generation (text, image, video) Generates completely novel content, instead of remixing and modifying existing content.
  • Generate public awareness campaign materials like fliers, website content, posters, and videos.
  • Generate visualizations of data.
Chatbots Leverages conversational models trained on massive dialogue datasets. Can have coherent discussions and execute tasks via conversation naturally.
  • Build a virtual assistant for common constituent questions.
  • Voice-enabled digital assistance.
  • Create a chatbot to guide users through services in their preferred language.
  • Increase first-call resolution for state customer service centers.
  • Reduce call wait and handle time at state customer service centers.
  • Create greater language access equity for program beneficiaries.
Data analysis Finds insights and relationships in data through learned knowledge about the world, without hand-coded rules or labeled training data.
  • Analyze healthcare claims or tax filing data to detect fraud.
  • Analyze network activity logs, identify cybersecurity anomalies and threats, and propose remediation actions.
  • Ticket triaging.
  • Root cause identification.
  • Resolution recommendation from historical tickets.
Explanations and tutoring Generates natural language explanations and tutoring through dialogue without human-authored content.
  • Explain program eligibility to potential enrollees.
  • Provide interactive tax assistance.
Personalized content Leverages user data, information and/or models to adaptively generate personalized content without explicit rules or large amounts of user data.
  • Auto-populate tax information and filing instructions based on a person's needs.
  • Help auto-populate public program applications based on a person’s situation and household composition.
Search and recommendation Uses contextual cues to improve search relevance and provide useful recommendations.
  • Searching or matching state code regulations concerning specific topics.
  • Recommend government services based on eligibility.
Software code generation Generates code by learning underlying structure and patterns of code, without the need for human written examples. Can expand short descriptions into full programs.
  • Translate policy specifications, such as Web Content Accessibility Guidelines (WCAG) and Americans with Disability Act (ADA) requirements, into software code.
  • Generate data transformation scripts from instructions.
  • Accelerate adoption of human-centered design in state web-based forms and pages.
  • Reduce administrative costs and burden to developing and maintaining best-in-class state government websites.
Summarization Does not require human-written summaries as training data. Can learn underlying patterns of language to generate summaries.
  • Summarize public comments to identify key themes.
  • Summarize public research to inform policymakers.
  • Summarize statutory or administrative codes.
Synthetic data generation Allows generation of new diverse, anonymized data from existing datasets for analysis and experimentation.
  • Generate synthetic patient data for training healthcare AI.
  • Generate simulated tax records for training tax auditing AI.

Source: State of California Benefits and Risks of Generative Artificial Intelligence Report, November 2023.

Evaluating use cases

When evaluating potential GenAI use cases, consider the following questions in order to explore the benefits and risks of deploying this technology:

  • What is the problem statement?
  • What are potential inequities in problem formulation?
  • What are the data inputs?
  • How and when will the solution be implemented and integrated into existing and future processes and delivery of services?
  • What is the return on investment? What are the alternatives to solving the problem?
  • Who will be the GenAI team responsible within the program area to monitor, validate, and evaluate the GenAI tool?
  • How does using the GenAI tool build trust with the end users like State staff or Californians?
  • Is the GenAI tool accessible and culturally appropriate?

Before initiating a GenAI project, it is also critical to consider the communities and users that may be impacted by the use case, whether they are the primary service population or groups that may be unexpectedly affected. A few criteria to begin thinking about the potential impacts of GenAI on communities include severity, scale, duration, and types of potential impacts of GenAI tools across a range of community group sizes.

Severity of risk

The degree and type of impact should be considered in determining an overall assessment of risk level for a GenAI use case.

Scale of risk

The scale and duration of the impact should be a consideration in determining the risk level for a GenAI use case. This should include an impact analysis of various affected groups – including, but not limited to, individuals, workers, communities, businesses, and local governments.

Important considerations in understanding GenAI risks include:

  • The risk of biased outputs from GenAI tools that amplify existing societal biases, particularly in use cases that may impact the delivery of public services.
  • The risk of hallucinated outputs from GenAI tools that can impact the validity, accuracy, or performance stability of State services.
  • The risk of harmful or inappropriate materials generated by GenAI, such as deepfakes that could spread misinformation.
  • The risk of GenAI tools lacking a human-in-the-loop reviewer who can validate the outputs of the system.
  • The risk of automation bias (an over-reliance on automated GenAI systems to make decisions), given the ability of GenAI tools to produce answers that “sound right” despite having no factual accuracy.
  • The risk of security vulnerabilities in GenAI tools exposed through new, natural language interfaces or source datasets.
  • The risk of black box GenAI applications that are unable to explain the rationale behind its recommendations for services that require this capability.
  • The risk of privacy re-identification issues for datasets including vulnerable communities that depend on anonymity for safe data analysis.

Source: State of California Benefits and Risks of Generative Artificial Intelligence Report, November 2023.

The Government Operations Agency (GovOps), Department of Technology (CDT), and Office of Data and Innovation (ODI) are working to develop guidelines for evaluating the impact of GenAI use cases on historically vulnerable and marginalized communities. These guidelines are anticipated to be published by July 31, 2024, and will be added to this toolkit once available.