LAVA Overview

Mission and Guiding Principles

Legal Aid Virtual Assistant (LAVA) is a chatbot designed by Atlanta Legal Aid Society, Inc. and LawDroid through a grant from the Legal Services Corporation to increase access to justice for the public. LAVA allows users to navigate common civil legal problems in Georgia by providing bilingual, plain-language answers and resources. Our goal is to help Georgians better understand their rights.

Guiding Principles:

• Access First: Every design and development decision prioritizes making legal information more equitable and accessible for users.
• Trustworthy Information: LAVA’s knowledge base was created from articles and resources publicly available on GeorgiaLegalAid.org. Each article is vetted by an attorney and reviewed annually for accuracy.
• Transparency & Trust: The platform will always disclose when a user is interacting with an AI and will clearly state the limitations of the tool.
• Ethical Use of AI: All platforms and tools are vetted to ensure privacy, security, and responsible data practices.
• Plain Language: All content is presented in clear, everyday terms to ensure legal issues are understandable to a broad audience.

Scope and Usage

What LAVA Does

• Provides conversational, text-based responses to user queries.
• Delivers information and resources on civil legal problems in Georgia.
• Links to self-help materials, legal clinics, and other Atlanta Legal Aid/Georgia Legal Services Program services.
• Supports both English and Spanish queries through automatic translation and response.
• Handles a wide range of input, including vague or misspelled questions.
• Suggests follow up questions relevant to conversation and answerable by LAVA.
• Provides the option for voice-to-text questions and text-to-voice responses.

What LAVA Does NOT Do:

• LAVA is NOT a lawyer. It cannot and will not provide legal advice.
• It does not search the internet or use generalized large language model (LLM) knowledge or training data.
• It will not speculate or fabricate answers when a query falls outside of its knowledge base; it is programmed to respond with "I do not have the information to answer your question."

Content and Information Management

Content Source

LAVA’s knowledge base is drawn from static documents that are one-for-one matches to the public-facing information and resources on GeorgiaLegalAid.org.

Content Maintenance

• LAVA's knowledge base is maintained by Atlanta Legal Aid’s Innovation team.
• Content is updated by editing static documents, a process that requires no coding.
• Routine reviews of the knowledge base are conducted in coordination with the review of GeorgiaLegalAid.org’s self-help materials to ensure accuracy and currency.

Data and Privacy

• LAVA operates as a closed system, which prevents the introduction of misinformation and AI "hallucinations."
• Input resembling personal and sensitive information is automatically redacted, ensuring user privacy.
• User feedback and usage trends are monitored anonymously to identify knowledge gaps and guide future improvements.

Ethical Safeguards and Disclaimers

To protect against the unauthorized practice of law (UPL), LAVA operates under strict ethical safeguards.

• Clear Disclaimers: A clear disclaimer is presented at the beginning of every user session, and cautionary language is embedded in responses to reinforce that LAVA is not a legal professional.
• "Cautious Design" Policy: LAVA's programming deliberately avoids speculative responses. This conservative approach, while potentially frustrating for some users, is a core safeguard to ensure interactions remain within ethical and legal boundaries.
• Static, Expert-Approved Content: All LAVA responses are built from pre-approved materials that are routinely reviewed by legal aid staff. Only trained staff can update this content.

Feedback and Accountability

• Users can rate each interaction with a thumbs up or thumbs down.
• A thumbs-down rating prompts a text box with the message: "Help us improve / What could we do better? Your feedback helps us provide more accurate and helpful responses."
• This feedback is stored in real-time and is used by the Innovation Team to identify knowledge gaps, clarify or expand existing answers, and flag queries that should not be addressed by the chatbot.

Future Development Roadmap

• Content Expansion: Add more legal practice areas and updated guidance.
• Data Driven Design: Utilize anonymized usage data to guide improvements to the system's design and functionality.
• Sync Content: Pull resources directly from GeorgiaLegalAid.org to LAVA via LawHelp API.

Our PDF download files do not fully comply with all applicable guidelines for accessible digital documents.