Safe use of Generative AI to enhance productivity in Boards of Directors' Corporate Governance

Man typing on laptop using Generative AI

In an earlier article we discussed how corporate governance processes can safely be managed digitally via end-to-end secured sealed processes in a “digital vault” 

How can we enhance the user experience and productivity through smart and safe application of Artificial Intelligence (AI), especially Generative AI, so called Large Language Models and the latest iteration – AI Agents? 

Artificial Intelligence itself together with Machine Learning (ML) is not new - it has been around for decades especially for applications involving pattern matching and prediction.  

Early AI/ML models were used for playing chess, bank fraud detection, customer behaviour analysis and prediction, weather prediction, medical imaging, military applications and so on. More recently AI has shown up in digital cameras, smart phone Apps and even home appliances.  

AI/ML became embedded in our lives without our really noticing. We didn’t notice because the tools of AI/ML were in the hands of large organisations that had the computer power, tools and skills to build their own application-specific data models. We only needed to use the end result that was embedded in our everyday devices. 

ChatGPT from OpenAI changed everything. OpenAI used vast computer power to crawl the internet and create a “neural network” model of all that knowledge – a so-called Large Language Model. What’s more ChatGPT had the ability for a user to make queries and create new content based on the LLM. That’s where the word “Generative” in Gen AI comes from. 

ChatGPT was not the first GenAI implementation, but what OpenAI did that was revolutionary was to make ChatGPT freely available to anyone – AI for the masses!  

ChatGPT has since been followed by numerous other offerings, with one of the most recent being China’s DeepSeek. This took GenAI to a new level by using much less computer power and being much faster to market. 

Agentic AI is the latest, and perhaps most exiting capability. AI Agents are like customised AI chatbots or automated modules that we can embed in other application software. We can even create our own AI agents using everyday tools like MS365 CoPilot. 

By asking questions through the smart use of “prompts” we can use GenAI to create “new” content, or help us better understand existing content. And we can use AI Agents to automate of workflows.  

But there are risks: 

1. Content produced by AI LLMs is based on an amalgamation of multiple source documents. When we use these tools, we don’t always know what those source documents are, whether they are based on copyrighted content, or even if they are true. In fact, even if the source documents are true, the AI model may combine them in a way that is no longer true. 

2. In the process of submitting a prompt, we are sharing some information with the model. That means we must be very careful in the questions we ask, especially if we are using a publicly available model. We may be sharing confidential information with a 3^rd party without knowing. Or our questions may subtly influence the response in a may that introduces bias. 

   
   

Knowing these risks, we can come up with some simple guidelines that help us use Gen AI safely. 

1. Never share confidential or sensitive information in a prompt with publicly available models such as ChatGPT. Fortunately, it is possible to set up private instances of basically the same models such as GPT4 and, configured correctly, these can safely be used for such sensitive applications. 

2. Beware of using Gen AI to conduct research or to create significant new content. Use tools that link to the source document(s) that can then be verified and then do verify the results! 

3. Do use Gen AI to help understand existing content such as summarise emails, documents or PDF files. 

4. Do use Gen AI to rearrange, combine, cross-reference or compare your own existing content. 

5. Do use Gen AI to create simple, generic drafts of content that you can easily check for correctness, such as job descriptions, marketing plans etc. 

6. Do always do a sanity check against generated content and remember that ultimately you own it! 

With the above in mind, Trusted Services (TSV) has already established its own private LLM instance for use by TSV staff.  

Further, TSV is using the LLM APIs to build seamless additional functionality into its Board.Vision corporate governance and compliance platform. 

These features include: 

1. Summarise a document, enabling the user to quickly understand key information without reading the whole document 

2. Ask direct questions of the document and receive answers in natural language, eliminating the need to search or collate manually 

3. Translate sections of the document instantly, making content accessible in different languages 

4. Content based search across multiple documents 

5. AI Assisted creation of Surveys 

6. AI Agents for natural language self-service in various areas such as “how to”, support queries, directors’ assistant, admin assistant 

   
   

board.vision screenshot

All of this takes place within the secure, sealed, encrypted Board.Vision digital vault. Data never leaves a TSV controlled environment. 

We look forward to developing other productivity improvements, but security and safety will always be our top priority.