Q&A: The Future of Artificial Intelligence and Contract Management

As the banking industry continues to embrace artificial intelligence, many bankers are struggling to understand what it means. How does it work? 

This includes developments in contract management. AI holds huge potential for making contract management more cost-efficient and effective – but there is also risk. 

Before risk can be mitigated, it must be understood. That’s why we reached out to Nathan Gonzalez, SVP of Software Development at Ncontracts, to ask about how software powered by artificial intelligence works and what it could mean for the future of contract management.

Table of Contents

• What is an artificial intelligence tool? What does it mean for something to be powered by AI? 
• What is a large language model? 
• Can you give me an example of an LLM? 
• How are AI models trained? 
• How can you use a LLM to improve contract management? 
• Will my data be safe if I use AI software for contract management? 
• How reliable is AI software for contract management? Will it make up answers? 
• What else can Ntelligent Contracts Assistant do? 

Q: What is an artificial intelligence tool? What does it mean for something to be powered by AI? 

Nathan Gonzalez: Artificial intelligence is an ambiguous term. It means a lot of things to a lot of people. 

AI is typically a combination of tools that people are using. It’s very rare that there’s just one tool in the toolbox.  

For example, a tool that can read and analyze a document to answer questions may require multiple AI tools. It might start with optical character recognition (OCR) to identify the letters, numbers, and symbols that make up the document. Then it might use a vectorization and indexing process that recognizes the characters and makes it possible to identify words and their meanings, as well as how similar they are to other words and phrases.  

Vectorization is an algorithm that discovers similarities and relationships between words. It transforms words into something a computer can understand. For example, “king” plus “woman” might equal “queen.” Vectorization is behind a lot of intelligent search engines. You don’t need an exact word match to generate results, as smart search engines determine your intent.  

Finally, the relevant data is fed through a large language model (LLM) which provides human-like analysis and text-based responses.

Related: AI and Risk Management Controls: How to Protect Your Institution

Q: What is a large language model? 

Nathan: An LLM is a type of artificial intelligence trained on large amounts of text data to understand and generate human-like text. Examples include answering a question, summarizing an article, and writing a poem.  

The "large" in its name refers to the size of the dataset used to train it as well as the complexity and capacity of the model itself. An LLM can have millions to hundreds of billions of parameters to help it understand the nuances of language.

Q: Can you give me an example of an LLM? 

Nathan: The most famous example of an LLM is OpenAI’s GPT-4. It’s the large language model that powers ChatGPT (full name Chat Generative Pre-trained Transformer). ChatGPT is a user interface that makes it possible to interact with the LLM.

Q: How are AI models trained? 

Nathan: There’s a common misperception that models like ChatGPT are trained in real-time. People think that it encounters new information and immediately integrates it into its knowledge base. That’s not how it works. 

There is a training process that happens offline, and it’s updated periodically. This training process can take months of time.  Remember, the “P” in GPT stands for “pre-trained.” As of January 2024, its last update was in April 2023, so it has no learnined nothing new after that date. It is stopped at that moment in time. It has no idea what happened yesterday. 

I often compare it to Drew Barrymore’s character in 50 First Dates.  The model “wakes up” when you begin a conversation, and as you continue it will use the information you entered, but when you end the conversation it “goes to sleep” and the next conversation starts the cycle over again.

Related: The Risks of AI in Banking

Q: How can you use an LLM to improve contract management? 

Nathan: We can train an LLM to identify essential information about a contract. Let’s say we want to be able to ask about a contract’s expiration date. You’d train the model by showing it all the types of content that looks like an answer to that question. Then you’d make sure the LLM answers the question using only the contract it was asked to review. So it really makes contract management faster and efficient.

Q: Will my data be safe if I use AI software for contract management? 

Nathan: It depends on the software. We all know that ChatGPT uses data from across the internet to train its models. Especially when using the free model, it will retain anything you enter for future model training, so you won’t want to enter any confidential data into ChatGPT. 

Other AI software, such as the Ntelligent Contracts Assistant add-on for Nvendor, is designed to “forget” user data and keep it segregated and confidential. For example, if you ask the Ntelligent Contracts Assistant a question about the expiration date for a vendor contract, it will find the answer and then forget it. It doesn’t retain the information to train the model. There are guardrails in place to protect data. It’s a closed ecosystem so no one else will ever see your data.

Q: How reliable is AI software for contract management? Will it make up answers? 

Nathan: Made-up answers are called hallucinations. In the early days of AI, it was more common to see these systems give incorrect answers. There’s been a lot of progress on this front.  

 We see hallucinations with Large Language Models sometimes because the parameters for ChatGPT are all the knowledge from training. Without firm guidance and framing of the scope of information you need an answer from, it is easy for it to go off track and “make up” answers. It’s important to note that these Large Language Models don’t understand language the way a person does. It only predicts the next sequence of characters based on patterns it’s seen in the past so sometimes it extrapolates too far.

A solution like the Ntelligent Contracts Assistant is much less likely to hallucinate. We use a technique called Retrieval Augmented Generation (RAG) to provide the LLM with exactly the context it needs to answer a question and the framing to not go outside the bounds of the provided context. These strict parameters have been created by Ncontracts’ legal, contract management, compliance, and technology experts.  

Ncontracts’ experts provide verifiable and accurate contract management content, significantly reducing the likelihood of incorrect answers. No technology is 100% accurate, but by giving our AI solution boundaries, it’s as close to 100% accuracy as you’re going to get.

Q: What else can Ntelligent Contracts Assistant do? 

Nathan: We already talked about how it can make third-party vendor contracts and agreements searchable so you can look for terms and conditions like price increases, autorenewal dates, and expiration dates. 

But it can do a lot more. It can summarize contracts – both as a whole and by section – so that hundreds of pages are condensed to just a few pages. 

My favorite feature is the contract scoring. It compares your contract with regulatory guidance and lets you know the risk of falling short of regulatory requirements. 

I’m really excited to see where it goes in the future. The possibilities are really endless!

Learn more about AI-powered contract management