How It Works: The Different Types of Legal AI

The American Bar Association’s guidance is clear: lawyers have an ethical duty to understand the AI technology they use. But with terms like generative AI, machine learning, and foundation models thrown around constantly, how do you know what’s actually happening behind the scenes?

You don’t need to become a computer scientist, but you do need to understand how legal AI works—both to meet your professional obligations and to make informed decisions about which platforms will actually help your practice.

Try Vincent AI Yourself

Why Understanding AI Technology Is No Longer Optional

The Professional Competence Imperative

ABA Model Rule 1.1 requires lawyers to maintain competence, including understanding “the benefits and risks associated with relevant technology.” This is a professional obligation that has been adopted by most states and that courts are beginning to enforce.

ABA Formal Opinion 512 makes this even clearer for lawyers who choose to use GenAI: “lawyers must have a reasonable understanding of the capabilities and limitations of the specific [GenAI] technology that the lawyer might use.” If you decide to incorporate AI into your practice, you don’t need to become an AI expert, you do have a professional obligation to understand the platform you’re using.

The “Black Box” Problem in Discovery

Federal Rule of Civil Procedure 26(g) requires lawyers to make a “reasonable inquiry” into the factual basis of their discovery responses. When that response involves AI technology, courts are demanding transparency.

The recent EEOC v. Tesla case provides a perfect example. Equal Emp't Opportunity Comm'n v. Tesla, Inc., 727 F.Supp.3d 875 (N.D. Cal. 2024). The court approved the use of generative AI for attorney responsiveness review, but only after requiring “disclosures regarding tools and processes necessary to make meet and confers meaningful.” The key principle emerging from case law—parties cannot meet and confer over a “black box.”

In legal technology, a “black box” refers to any system where you can see the inputs and outputs, but you can’t understand or explain what happens in between. If you input documents or prompts and get back results, but can’t explain to opposing counsel how the AI made those determinations, you’re dealing with a black box—and courts won’t accept that.

While Vincent AI is not an eDiscovery platform, it addresses the broader transparency concerns raised by the black box problem in legal research. As vLex Chief Strategy Officer Ed Walters explains: “I like to think of ours as ‘glass box’ AI because you get to look in and have some control over it.” Vincent shows users exactly which legal authorities it relied on, provides confidence rankings for each source, and allows lawyers to see—and even modify—the resources underlying every AI-generated analysis. This transparency ensures lawyers can meet their professional obligations while benefiting from AI’s efficiency.

Essential AI Definitions: What Lawyers Need to Know

Artificial Intelligence (AI)

At its core, AI is technology that performs tasks typically requiring human intelligence, like pattern recognition, decision-making, and learning from data. In legal practice, this means automating tasks like document review, legal research, and analysis.

While it may seem like AI is a new technology, it was actually invented back in the 1960s and has been a bigger part of your life than you realize. For example, Microsoft’s “Clippy” was a rudimentary AI assistant. Legal discovery platforms, like Relativity, have been using AI since the early 2000s.

Think of it like this: AI is like giving your brilliant paralegal superpowers. They never get tired, never need vacation, and can work through thousands of documents in the time it takes you to grab coffee. AI is great at following instructions and spotting patterns, but it still needs you and your team to make the important judgment calls.

Machine Learning (ML)

Machine learning is a subset of AI that learns patterns from data without explicit programming. If you give a machine learning system “Square, Circle, Triangle” and ask what comes next, it probably won’t know. But show it “Square, Circle, Triangle, Square, Circle” and it can start predicting the pattern.

The more training data the system observes, the better it becomes at prediction. Machine learning excels at spotting patterns and identifying outliers in large datasets, which is why it’s so powerful for legal document review.

Think of it like this: Machine learning is like training a new associate. At first, they need you to review their work on every single document. But after they’ve seen enough examples of what makes a document “responsive” or “privileged,” they start making those determinations on their own—and they get better with each case they work on.

Foundation Models

Foundation models are large-scale AI systems trained on massive, diverse datasets that serve as the base for many different applications. These are the general-purpose AI systems like ChatGPT, Claude, or Gemini that know a little about everything but aren’t specialized for any particular field.

Large Language Models (LLMs) are a specific type of foundation model that focuses primarily on understanding and generating human language. This is what allows users to enter prompts in plain language and receive coherent, digestible human-like responses.

Think of it like this: A foundation model is like hiring a brilliant recent graduate who has read everything on the internet. They have broad knowledge across many subjects, but they haven’t specialized in law yet. They can be helpful, but they don’t understand the nuances of legal practice or have deep expertise in legal reasoning. (*Note: Foundation models have not considered the security required of lawyers working with confidential client data. This recent graduate could blab about your client all over town.)

Domain-Specific Models

Domain-specific models take foundation models and enhance them with specialized knowledge, databases, and workflows for particular fields. In legal technology, these models combine the language capabilities of foundation models with legal expertise, legal databases, and legal-specific training.

Think of it like this: A domain-specific legal AI model is like that same brilliant graduate after they’ve completed law school, passed the bar, worked at your firm for several years, and memorized your entire law library. They still have that broad intelligence, but now they understand legal concepts, know how lawyers work, and can apply legal reasoning to complex problems.

Generative AI (GenAI)

Generative AI creates new content based on training data. As a concept, GenAI is like music. While every musical note already exists, songs create new combinations from existing notes. The content is still “new,” just assembled from existing knowledge.

In legal practice, GenAI can draft documents, create research summaries, and assist with legal writing. Due to strict ethics requirements in the legal profession, it is essential that GenAI always be utilized alongside attorney oversight and review.

Think of it like this: Generative AI is like having a ghostwriter who has studied thousands of legal briefs in your practice area. They can draft a motion for summary judgment that sounds like it could have come from you, using language and arguments that feel familiar because they’re based on successful briefs from similar cases. But you still need to review, edit, and put your name on it.

Agentic AI

This is cutting-edge AI that can act autonomously to achieve specific goals through multi-step reasoning and planning. As AI expert, Daniel Hoadley, explains, agentic AI must have two key qualities: “It must be equipped with tools to perform actions, and it must have the ability to reason about how to use those tools.”

Where general AI performs single tasks, agentic AI can handle complex, multi-step processes by dynamically choosing which tools to use and in what sequence. For example, if you asked an agentic AI system to “analyze this contract,” it might:

• Identify what type of contract it is
• Extract the key terms and parties
• Research applicable law based on the contract’s jurisdiction or choice of law clause
• Find relevant case law showing how similar contracts were upheld or invalidated
• Flag unusual or potentially problematic clauses
• Generate a summary of important findings

The key difference is, instead of following pre-programmed workflows, agentic AI dynamically directs its own processes and chooses which tools to accomplish complex tasks.

Think of it like this: Agentic AI is like having a senior paralegal who doesn’t just follow your checklist—they create their own checklist based on what needs to get done. They use their judgment to break down complex tasks into manageable steps.

Retrieval Augmented Generation (RAG)

RAG is a technique that combines the power of large language models with access to specific, authoritative databases. Instead of relying solely on what AI has learned during training, RAG allows AI to search through and retrieve information from designated sources, like legal databases, before generating its response. This ensures answers are grounded in current, verified information rather than potentially outdated training data.

In legal applications, RAG is particularly valuable because it allows AI to access the most recent case law, statutes, and legal authorities while generating responses, significantly reducing the risk of hallucinations or outdated information.

Think of it like this: RAG is like giving your research team the ability to instantly access and cross-reference the entire law library while they work. Instead of relying on what they remember from law school, it's as if your research team can instantly pull the relevant books and cases, read them, and then give you an answer based on what they just found.

Zero Data Retention

Zero data retention means that AI systems don’t store, learn from, or retain any of the confidential information you input during your legal work. When you upload documents or submit queries, the AI processes your request but immediately discards your data once the task is complete. Nothing is saved, cached, or used to improve the AI model.

This is critical for legal professionals who must protect attorney-client privilege and maintain client confidentiality. Vincent AI operates under strict zero data retention policies, backed by enterprise-grade security certifications including SOC 2 Type II and ISO 27001 compliance.

The importance of zero data retention has been highlighted in recent legal proceedings where AI companies have faced broad discovery requests for user data. In these situations, only those with explicit zero data retention agreements found their data exempt from disclosure requirements, demonstrating why having formal zero data retention policies provides not just privacy, but also legal protection.

Think of it like this: Zero data retention is like having a paralegal with convenient amnesia about confidential matters. They can brilliantly help you with your work while it’s happening, but the moment the task is done, they completely forget every confidential detail they just processed.

How Vincent AI Is Engineered Specifically for Lawyers

Vincent AI represents a domain-specific legal intelligence platform, meaning it takes the power of foundation models and engineers them specifically for legal work. Rather than simply wrapping an interface around general-purpose AI, Vincent combines multiple AI technologies with vLex’s 25+ years of legal expertise and one of the world’s largest legal databases.

Vincent integrates several of the AI approaches we’ve defined:

Machine Learning is used to clean up and classify the underlying data from both Docket Alarm’s court records and vLex’s legal database. This preprocessing makes the raw legal information more structured, creating a stronger foundation that makes Vincent’s responses more effective and accurate.

Foundation Model Intelligence provides the core language understanding capabilities, but enhanced with legal-specific training and knowledge.

Domain-Specific Legal Engineering is what sets Vincent apart. It’s not general AI that happens to work on legal documents, but AI that has been specifically designed to process legal concepts, legal reasoning, and legal workflows.

Generative AI Capabilities enable Vincent to draft legal content, create research summaries, and assist with legal writing, all grounded in authoritative legal sources with proper citations.

Agentic AI Features allow Vincent to handle certain complex, multi-step legal tasks autonomously while maintaining the transparency and oversight that legal professionals require.

Vincent is AI that has been precision-engineered for lawyers, combining the broad intelligence of foundation models with deep legal expertise, comprehensive legal databases, and workflows designed specifically for legal practice.

Building AI Competence: Your Path Forward

Professional competence has always evolved. From typewriters to computers to the internet, the lawyers who understand their tools first are the lawyers who lead their profession. Today, AI literacy is becoming essential to competent legal practice.

Understanding these AI fundamentals helps you meet your professional obligations while positioning you at the forefront of legal innovation. The lawyers who grasp these concepts now will shape how the legal profession adapts to an AI-driven future.

In upcoming articles, we’ll dive deeper into exactly how Vincent’s legal-specific engineering works in practice, including how Vincent utilizes agentic AI, generative AI, and machine learning.

Ready to experience AI that’s been precision-engineered for legal work? Start your free trial of Vincent AI today.

Start Your Free Trial

Authored by

Sierra Van Allen