The way we create, interpret, and apply legislation, standards, and guidance is transforming faster than ever. While paper and digital proxies such as PDF and web formats continue to have their place, attention is rapidly turning to the question of how machines also consume this material.
AI, the latest disruptor, is promising sweeping benefits for all stakeholders and there is a tremendous amount of hype. Indeed, AI itself is responsible for generating some of the hype. The internet and social media are increasingly awash with textual and multimedia content speaking to the benefits of AI that was itself created with the aid of AI.
Yet, amidst all this noise surrounding AI, it is very clear that Generative AI (Gen AI), also discussed as Large Language Models (LLMs) or chatbots, could very well be as significant to the world of words as Excel was to the world of numbers. But what does this really mean for those tasked with drafting, interpreting, publishing, and implementing legislation, standards, and guidance?
Current challenges for rule-makers and rule-takers
Rule-makers and rule-takers form an interconnected ecosystem – an information supply chain. Once a law, a standard, or a guidance document is created or amended, businesses must take action to ensure that they are compliant. However, this is not as simple as it looks on paper.
Post World War II, IT innovation has skyrocketed, and we are now at the point where IT systems are an established feature of almost all rule-making and rule-taking processes. Drafts that were once handwritten, marked up in blue and red pens on sheets of paper, are now typed into a Word document and marked up with track changes.
These days, the vast majority are ‘born digital’, in other words, they do not start out as paper artifacts at all. A law, standard, or guidance document, once published as a book and read by subject matter experts (SMEs) is now published as a PDF or a web page. The contents of interest to the SMEs are then copied and pasted into knowledge management systems that range from simple documents in a folder structure to sophisticated line-of-business applications tuned to the needs of particular SMEs such as auditors, lawyers, engineers, physicians, etc.
Since the seventies in particular, these line-of-business applications have grown in sophistication and the laborious and error-prone copy/paste paradigm is now seen as a bottleneck to progress. There is no debate about the value that true machine readability can bring but there is debate about how best to achieve it. Is AI the answer? Can we just ‘throw Gen AI at it?’
AI: The good and the bad
Gen AI has captured the public imagination due to the emergence in recent years of tools such as ChatGPT, Gemini, DALL·E, and the emergence of the co-pilot paradigm for augmented intelligence whereby AI sits alongside an SME and helps them in their knowledge work, almost like having a personal assistant.
It has become clear that Gen AI’s ability to process and generate human language has numerous applications for rule-makers and rule-takers, including:
- Summarization of long, complex texts
- Generating a first draft of a law, standard of guidance document and thus tackling the common writer’s block problem
- Simplification of language, e.g. ‘explain like I’m five’
- Highlighting changes to text that are material as opposed to editorial
- Extracting requirements
- Automating tense shifting such as converting a document from future tense to past tense
There is an entirely justified and appropriate sensitivity here. Clearly, as our VP of Government Solutions captured at Propylon’s Legislative Management Technology Summit this year, Gen AI is an assistive technology and its work product needs to be reviewed just like the work of an intern should always be reviewed by an SME. For all its power, Gen AIs can exhibit bias and hallucinate.
The adage ‘garbage in, garbage out’ is often used in the context of Gen AI to highlight the dependency of LLMs on the quality of their input data. This is absolutely true but even with 100 percent perfect source data, AI’s can still hallucinate and can indeed, easily fall out of date.
Thankfully, there are techniques to address these challenges. In particular, a content organizational structure known as a knowledge graph combined with an AI technique known as Retrieval-Augmented Generation (RAG).
The volume of rules and regulations has grown exponentially in many jurisdictions over the last few decades.
The Power of Knowledge Graphs and RAG
AI innovators are indicating a preference for the term ‘knowledge graph’ to capture the concept of structured content in its most general form. Knowledge graphs encompass digital content of all forms ranging from relational databases to documents to multimedia, focusing on mapping out the interconnections between all the component pieces.
Although the term is a newly emerging one, you most likely already have the beginnings of a knowledge graph in your current method of managing content. In fact, if your field is law, you are already likely aware of the very sophisticated knowledge graphs known as ‘citation networks’ that have been in use for centuries, in caselaw citators for example.
Using RAG, the knowledge graph of your content is used to drive the core Gen AI system, feeding it accurate, up-to-date information from your business systems that then maximize the quality of the results coming from the Gen AI. Furthermore, with RAG, the AI will cross-reference your internal information in its responses allowing you to fact-check with ease. In the new terminology of the Gen AI era, this is referred to as ‘ground truth’.
This combination of knowledge graphs and RAG provides rule-makers and rule-takers with a powerful paradigm for maximizing the value of Gen AI but in order to do that, it is necessary to get your content ready for Gen AI.
Start now, become AI-ready
Emerging thought leadership and analysis suggest that overall, businesses are behind the AI adoption curve in comparison to individuals. According to a working paper published by Alexander Bick of the Federal Reserve Bank of St Louis and co-authors from Harvard Kennedy School and Vanderbilt, more than 24 percent of workers used it at least once in the week prior to being surveyed. The report defines Gen AI as a general-purpose technology, used in a wide range of occupations and job tasks at work and at home.
As Gen AI tools become a mainstay on browsers, co-pilots, and in general information-seeking behaviors, users will expect access to an AI experience in their line-of-business applications.
Getting ready for the inevitability of Gen AI requires rule-makers and rule-takers to begin to shift their thinking from documents to machine-readable data. Paper publications and their digital proxies will continue to be part of the ecosystem for the foreseeable future. However, laws, standards, and guidance must now be structured and managed at the level of the knowledge graph components that they contain.
This shift from documents to data is an essential transition as the mission critical material that makes up the fabric of our societies becomes ever more reliant on the engagement between humans and machines.
Your success with Gen AI initiatives hinges on getting your internal content into the machine-readable knowledge structure that AI needs to produce quality results.
Over the next few years, Gen AI will continue to evolve and get better and better but ultimately you need to integrate it with your highly organized, structured, ground truth to get the most out of it.
The time to start that process is now.
