Thinking of using AI tools for legal help and information? Read this first.
Earlier this year, CPLEA worked with Alberta Law Libraries (ALL) to gather frontline insights about how people are using AI tools to find legal help and information. We used our findings to develop a new public resource, Using AI for Legal Help. This article is part of a two-part series that shares key insights and tips from the project.
It’s easy to be flooded with AI news and tools everywhere these days. Eventually, though, there’s a real risk that everything AI starts to sound eerily familiar. Remember those late-night infomercials peddling variations of common household products under a catchy, made-up name like wowRAG?*
You know, that incredible, multi-purpose wowRAG with super absorbent powers that will clean up any kitchen mess in mere seconds – better, smarter, and faster than your mere human hands with any other cloth can do! Want a super clean, sparkly finish? You got it, just watch the magic happen! wowRAG also functions like a security blanket – especially when you are tired, short of time or need extra support! You won’t be able to live without it!
Let’s admit it, wowRAG can do amazing things. But before you take that wowRAG and use it to tackle something serious like a raging grease fire, have you thought about whether it can reliably do what you think it can do? You probably should not put out that fire with just a wowRAG – after all, it is a synthetic rag made of unknown substances. Maybe one day, industry innovations will be so advanced that wowRAG can help solve our everyday problems. But for now, it probably can’t put out that grease fire for you – or at least in the way that’s been promoted to you.
The point is that we can be both optimistic about the potential of AI with legal help and aware of its current limitations – especially as they relate to the use of readily available generative AI (GenAI) tools. We know from our everyday work and the Nielsen Norman Group’s recent user experience research that users are increasingly turning to these GenAI tools for information-seeking tasks. This includes starting with vague ideas, juggling multiple requirements, or combining information from multiple sources. Users then turn to traditional search methods, such as Google, for accuracy and trusted sources.
So how do we make sense of AI for legal help and information – especially as we hear more about cases of AI misuse ending up in court? It’s helpful to start by understanding the tools themselves.
How do AI tools work?
GenAI tools like AI chatbots, assistants, co-pilots, and agents can produce content in many formats. Large Language Models (LLMs) power these tools. AI developers train LLMs on massive amounts of data that include websites, books, and articles. The data can come from all sorts of sources, places, and jurisdictions. When trained, these LLMs put words in order of statistical probabilities in a way that sounds human-like. However, they can hallucinate sometimes and make things up to please you – the user.
Not all AI tools are the same: non-grounded vs grounded AI tools
There are two main types of AI tools: non-grounded and grounded. This distinction is not well known, but important to know. Non-grounded tools rely on their training data and do not access real-time information, like the live web. On the other hand, grounded AI tools draw on external or real-time information sources, including the live web, and back up their output with references.
Grounding an AI tool ensures that the outputs are not just plausible but backed up with verified data sources or a closed knowledge system. This reduces the likelihood of hallucinations or fabricated outputs. Grounding can also involve fine-tuning an AI tool using specific instructions and parameters as input. If it doesn’t do what it’s supposed to do, you can correct it with proper inputs so that it generates more accurate outputs.
You can think of it this way – the process of grounding a tool is like training a puppy. Let’s suppose you train the puppy to fetch specific objects in your home, not a bunch of random ones. “Go fetch me slippers, not that random ball, bone or leash over there!” If it’s a good and reliable puppy, it should be fetching what you’ve trained it to fetch within the home. The puppy should not bring you an imaginary object or run rampant in the home. It should not break out of the home to bark up the wrong tree, roam around freely, or, worse, break into someone else’s home to steal slippers for you. While using grounded AI tools tends to be less risky than non-grounded ones, they are still not completely free of hallucinations or errors.
Grounded, law-specific AI tools and where to find them
Unlike readily available AI tools, grounded, law-specific AI tools are not well known to the public. These law-specific AI tools are grounded in their own proprietary databases, and the legal industry typically uses them. While these tools are expensive and require subscriptions, some are available at local law libraries for free. For example, in Alberta, AI-enhanced or powered legal research tools such as LexisNexis+ and Westlaw Edge are available for free at Alberta Law Libraries.
In Canada, there are a few free, publicly available, and grounded, law-specific AI tools. CanLII has a new AI-powered research system, CanLII Search+, that draws on its own database of case law, legislation and commentary.
In British Columbia, public legal education organizations offer grounded, law-specific AI tools for people seeking legal information. For example, the People’s Law School has a Beagle Chatbot, and the Justice Education Society of BC has a Legal Help Chatbot on their FamilyLaw website to help guide people to legal information for British Columbia.
Learn more about using AI for legal help and information in Part II: Tips and Best Practices of this article series.
* NOTE: wowRAG is a made-up name and not a real product. Any association with any product is coincidental. However, the term “RAG” (Retrieval-Augmented Generation) refers to a real process for grounding an AI tool. RAG supplements an AI tool or system’s training data by providing verified data from specified or limited sources. RAG can reduce AI training data gaps, as well as prevent hallucinations – but it may not result in entirely error-free outputs.
The information in this two-part article series, AI Quick Start Guide for Legal Help, is based on CPLEA’s current understanding of how AI tools work today (which is ever evolving). The analogies and metaphors used to explain these concepts reflect the author’s perspective.
Looking for more information?
Looking for articles like this one to be delivered right to your inbox? SUBSCRIBE NOW!
DISCLAIMER The information in this article was correct at time of publishing. The law may have changed since then. The views expressed in this article are those of the author and do not necessarily reflect the views of LawNow or the Centre for Public Legal Education Alberta.
About Judy Feng
