In recent years, the rise of generative AI has become undeniable across virtually all sectors – and the legal world is no exception. The latest generation of AI tools, focusing on advanced reasoning models and Retrieval-Augmented Generation (RAG), promises to fundamentally change how lawyers and legal professionals work. In this article, we dive into concrete experiments and results from recent research and explain how these technologies can transform legal practice in the future.1
In this blog, we explore the two main innovations in legal AI: advanced reasoning models and Retrieval-Augmented Generation (RAG). We discuss the results of a recent experiment with law students, analyze the impact on productivity and quality of legal work, and look at the future implications for legal practice.
Figure: Score distribution for legal memos clearly shows higher average scores for both o1-preview (red line) and Vincent AI (green line) compared to the control group without AI (blue line).
The Two Main Parts of Innovation
The recent breakthroughs in AI for legal work can be divided into two main categories. Both categories have their unique benefits and applications in practice.1
1. AI Reasoning Models
Traditional AI models, such as earlier versions of ChatGPT, already took considerable work off our hands. However, with the advent of AI reasoning models – such as OpenAI's o1-preview – a completely new dimension emerges. These models are specifically developed to comprehend complex, multi-step legal issues.
In concrete terms, this means that the model internally builds a "chain of reasoning," similar to how a lawyer first makes a plan before approaching a complex legal problem.5 This results in answers with greater analytical depth, which is essential when crafting well-founded legal arguments.
The power of these models lies in their ability to:
Break down complex legal concepts step by step
Weigh contradictory arguments against each other
Draw logical conclusions based on precedents
Formulate nuanced legal advice that takes multiple factors into account
2. Retrieval-Augmented Generation (RAG)
On the other hand, we have RAG technology, illustrated by tools such as Vincent AI. This technology combines the power of generative AI with advanced search and document retrieval systems.4 This allows the answers to be anchored in current, reliable legal sources, such as case law, statutes, and other primary documents.
This is especially important because traditional models often tend to generate "hallucinations" – that is, making up facts or sources – which is unacceptable in legal practice.4 By using RAG, lawyers can always verify the output by consulting the underlying sources.
Technology
Core Advantage
Practical Application
AI Reasoning Models
Deep analytical capacity
Complex legal memos, argumentation structures
RAG Technology
Factual accuracy & source referencing
Case law research, statutory analysis
The Experiment: A Practice-Oriented Approach
To measure the actual impact of these AI tools on legal work, a randomized controlled trial was conducted with 127 law students from the University of Minnesota and the University of Michigan.1 The setup of the experiment was as follows:
Three Groups:
No AI Support: The students were given access to traditional legal sources, such as Westlaw or Lexis, but were not allowed to use generative AI tools.
AI Reasoning Model (o1-preview): This group used an advanced reasoning model that performed step-by-step legal analyses.
Vincent AI (RAG-based): This group worked with a tool that combines AI with automatic retrieval of legal sources and integrated prompting.
Six Realistic Legal Tasks:
The students were given six assignments, developed in collaboration with experienced lawyers. Concrete examples of these include:
Assignment 1: Drafting an email to a client (with a time limit of 60 minutes) explaining why a defamation claim cannot be based solely on statements made during a trial. Students had to cite relevant case law and statutory provisions.
Assignment 2: Writing a comprehensive legal memo for a partner, with a time limit of 240 minutes. This task required in-depth analysis and structured argumentation, focusing on both analytical depth and legal accuracy.
The participants received intensive training beforehand, both on the general use of AI in legal practice and on the specific use of Vincent AI. This ensured that all participants, regardless of their assigned group, could utilize the AI tools optimally.
Concrete Results: Speed and Quality Hand in Hand
The results of the experiment were promising and provide a clear picture of how AI can transform legal practice:
Improved Productivity
Speed Gains: Students working with AI tools were significantly more productive. Vincent AI delivered productivity improvements of 38% to 115%, while o1-preview increased productivity by 34% to 140%.1 This meant they could accomplish significantly more work in the same timeframe compared to the control group without AI support.
This productivity gain was not only noticeable for simple tasks but also for complex legal analyses. Even for the most challenging assignments, such as drafting a comprehensive legal memo, the time savings were significant.
Improvements in Work Product
O1-Preview:
This tool resulted in a clear improvement in the analytical depth of legal memos and emails.
Students working with o1-preview produced assignments that were better structured and had a more logical construction.
It was noted that, despite the increased analytical quality, hallucinations occasionally occurred – erroneous additions that could slightly undermine reliability.1
Vincent AI:
Vincent AI excelled particularly in improving the clarity, organization, and professionalism of the assignments.
The number of hallucinations remained approximately the same as in assignments performed without AI, indicating that this tool did not introduce additional errors.4
The combination of AI with automatic retrieval of legal sources made it easier for students to verify their work and substantiate it with current case law.
Aspect
Without AI
With o1-preview
With Vincent AI
Productivity
Baseline
+34% to +140%
+38% to +115%
Analytical Depth
Average
Significantly Higher
Higher
Source References
Limited
Extensive (with risk of hallucinations)
Extensive and verifiable
Structure and Organization
Variable
Consistently Good
Excellent
What Does This Mean for the Future of Legal Practices?
The findings of this research indicate that the combination of AI reasoning models and RAG technologies has the potential to fundamentally improve legal practice:1
Synergy in Use
Combining both technologies can lead to even greater efficiency and accuracy. Consider a situation where a lawyer applies both in-depth analysis (via reasoning models) and real-time verification of sources (via RAG) – this would significantly reduce the risk of errors.
A concrete example: a lawyer analyzing a complex contractual matter can use the reasoning model to explore the various interpretation possibilities, while the RAG technology directly provides relevant case law and statutory provisions that support or refute these interpretations.
Support, Not Replacement
Although AI offers enormous benefits, human expertise remains essential. AI serves as a powerful assistant that supports and enhances the lawyer's work, but the ultimate legal judgment and ethical considerations remain the responsibility of the human.
This aligns with what we see in other sectors: AI is most effective when deployed as a complement to human expertise, not as a replacement for it. The lawyer of the future is not the one who is replaced by AI, but the one who knows how to optimally utilize AI.
Future Developments
As these technologies are further refined, we can expect them to become even better at delivering high-quality legal analyses, which will significantly increase the competitiveness and productivity of legal teams.
The legal firms that now invest in integrating these technologies into their work processes will likely build a significant competitive advantage. This is comparable to the transition to digital documentation in the 1990s – firms that are at the forefront of adopting new technologies can offer their services more efficiently and at lower costs, while simultaneously improving quality.
Practical Implementation: How to Begin?
For legal professionals who want to start integrating AI into their practice, here are some practical steps:
1. Experiment with Different Tools
Start by exploring different AI tools specifically designed for legal work. In addition to the tools mentioned in this article, there are other options available, each with their own strengths. Experiment with different tools to see which best aligns with your specific needs and work style.
2. Start with Simple Tasks
Begin by applying AI to relatively simple, low-risk tasks, such as:
Drafting initial concepts of standard documents
Summarizing lengthy legal texts
Generating checklists for due diligence
As you become more familiar with the technology, you can gradually move on to more complex applications.
3. Integrate AI into Your Existing Workflow
Instead of overhauling your entire work process, look for specific points in your existing workflow where AI can add the most value. This could be, for example, in preparatory research, drafting initial concepts, or checking documents for consistency and completeness.
4. Invest in Training
Ensure that you and your team receive adequate training in the effective use of AI tools. This includes not only technical skills but also insight into the strengths and limitations of the technology, and how to critically evaluate the output.
5. Develop Clear Guidelines
Establish clear guidelines for the use of AI within your practice, with particular attention to:
Confidentiality and data protection
Verification of AI-generated output
Transparency to clients about the use of AI
Implementation Phase
Focus Points
Expected Results
Exploration
Experiment with different tools
Insight into possibilities and limitations
Initial Implementation
Focus on low-risk tasks
Early productivity gains, building trust
Full Integration
Develop workflows and protocols
Systematic productivity improvement
Conclusion
Implementing advanced AI systems in legal work has now become a reality. Practical research convincingly shows that this technology is not merely a time-saving tool - from simple correspondence to complex legal analyses, we see that AI elevates legal services to a higher level, both in terms of efficiency and substantive quality.1
For lawyers, this means a shift in the way of working: AI becomes an essential instrument that helps them work more efficiently, accurately, and ultimately more effectively. The combination of AI reasoning models for in-depth analysis and RAG technology for factual accuracy provides a powerful set of tools that can fundamentally improve legal practice.
As with any technological revolution, there will be early adopters who reap the benefits of increased productivity and competitive advantage, and those who lag behind who risk falling behind. The question is not whether AI will transform legal practice, but how quickly and how profoundly – and whether your practice will be at the forefront of this transformation or chasing after it.
Curious about how your legal practice can benefit from these developments? Follow our blog for the latest insights and practical tips on the use of AI in the legal world. If you have questions or want to know more about specific applications, please don't hesitate to contact us!
