Responsible AI implementation
We partner with ambitious attorneys and forward-thinking law firms who understand how to use AI responsibly.
Our approach mirrors the legal process, ensuring thorough implementation and risk mitigation.
Our approach mirrors the legal process, ensuring thorough implementation and risk mitigation.
1
Step 1: Pleadings
We understand the risks associated with language models, and how they relate specifically to the legal industry.
"One of AI’s prominent applications made headlines this year for a shortcoming known as 'hallucination,' which caused the lawyers using the application to submit briefs with citations to non-existent cases. (Always abad idea.)"
"However, these tools have the welcome potential to smooth out any mismatch between available resources and urgent needs in our court system."
- Chief Justice John Roberts, 2023 Year-End Report on the Federal Judiciary
That's why we begin with identifying low-risk, high-impact areas for AI integration, ensuring a solid foundation for your AI strategy. For example, we've automated call transcriptions and report generation for legal intake and pre-qualification, providing you with structured summaries of phone calls that streamline research initiation and matter preparation.
"One of AI’s prominent applications made headlines this year for a shortcoming known as 'hallucination,' which caused the lawyers using the application to submit briefs with citations to non-existent cases. (Always abad idea.)"
"However, these tools have the welcome potential to smooth out any mismatch between available resources and urgent needs in our court system."
- Chief Justice John Roberts, 2023 Year-End Report on the Federal Judiciary
That's why we begin with identifying low-risk, high-impact areas for AI integration, ensuring a solid foundation for your AI strategy. For example, we've automated call transcriptions and report generation for legal intake and pre-qualification, providing you with structured summaries of phone calls that streamline research initiation and matter preparation.
2
Step 2: Discovery
Concerned about client confidentiality and data security? Good.
As it's being discovered, frontier language models were trained on IP-protected data The New York Times Company v. Microsoft Corporation (1:23-cv-11195). This is a nightmare for legal professionals whose livelihoods depend on their work product.
Some, but not all, AI companies will enter into a Data Processing Agreement (DPA), which is a legally binding contract between a Data Controller ("Controller") and a Data Processor ("Processor"). We help you navigate this process and assesses your unique security needs. With this information, we're able to determine the safest deployment strategy, from web apps hosted on our servers, servers of your choice, or completely on-premises local solutions that don't even require an internet connection to work. Our platform is model agnostic allowing for the use of one or more models simultaneously.
As it's being discovered, frontier language models were trained on IP-protected data The New York Times Company v. Microsoft Corporation (1:23-cv-11195). This is a nightmare for legal professionals whose livelihoods depend on their work product.
Some, but not all, AI companies will enter into a Data Processing Agreement (DPA), which is a legally binding contract between a Data Controller ("Controller") and a Data Processor ("Processor"). We help you navigate this process and assesses your unique security needs. With this information, we're able to determine the safest deployment strategy, from web apps hosted on our servers, servers of your choice, or completely on-premises local solutions that don't even require an internet connection to work. Our platform is model agnostic allowing for the use of one or more models simultaneously.
3
Step 3: Trial
Next, we develop and rigorously test your custom AI tools.
With any new technology there comes known and unknown risks. We first address the known risks, namely data privacy and security, human error, and hallucinations. We help you sign DPAs to reduce data risks, offer practical training to reduce the risk of human error, and constantly improve our citation checker to reduce the risk of hallucination. No platform is perfect, but our team is obsessed with finding solutions.
For example, We analyzed nearly 5,000 zero-shot case law citations generated by GPT 3.5 Turbo. Our goal was to identify accuracy patterns across various jurisdictions and practice areas. We gave the language model as little help as we could in terms of prompting and grounding. The results show that GPT case law citations vary widely across jurisdictions and practice areas, with Federal Constitutional Law being most accurate, and Maine Bankruptcy Law being the least accurate.
View case law hallucinations by jurisdiction and practice area here.
With any new technology there comes known and unknown risks. We first address the known risks, namely data privacy and security, human error, and hallucinations. We help you sign DPAs to reduce data risks, offer practical training to reduce the risk of human error, and constantly improve our citation checker to reduce the risk of hallucination. No platform is perfect, but our team is obsessed with finding solutions.
For example, We analyzed nearly 5,000 zero-shot case law citations generated by GPT 3.5 Turbo. Our goal was to identify accuracy patterns across various jurisdictions and practice areas. We gave the language model as little help as we could in terms of prompting and grounding. The results show that GPT case law citations vary widely across jurisdictions and practice areas, with Federal Constitutional Law being most accurate, and Maine Bankruptcy Law being the least accurate.
View case law hallucinations by jurisdiction and practice area here.
4
Step 4: Appeal
Our commitment extends beyond initial deployment. Through regular consultations, we work to improve your practice assistants. This process can include prompt engineering, creating custom datasets to enhance retrieval, accuracy and relevance, or advanced techniques like Reinforcement Learning from Human Feedback (RLHF). We also offer training sessions keep your team updated on best practices and new features.
