Over the past few weeks, we’ve reviewed the history of Artificial Intelligence and AI’s role in eDiscovery and the review process. Today we will look at how you can leverage the power of today’s AI in your review process.
A quick note before we break down some of the more interesting AI models currently available. Like any emerging field fueled by investment capital, there is a lot of buzz and marketing around specific AI products. I’ve chosen to leave out product names for this article as many platforms have different versions of largely the same thing.
Instead, I’ve chosen to focus on the types of AI models out there so you can start to imagine the problems they may solve for you.
Generative AI / Large Language Models (LLM)
ChatGPT is a large language model (LLM) generative AI that draws on its universe of indexed data to answer questions in a conversational tone. It can also summarize books or articles, create PowerPoint presentations from the text of an email and even keep your inbox cleaned out. Perhaps its greatest achievement comes in making AI approachable to the common person, allowing us to be curious and ask questions.
In a legal application, these types of AI models applied to your databases will allow you to interact with your documents the same way you can interact with ChatGPT today.
In their most basic sense, these AI models allow you to interrogate your evidence. By training itself on the documents the user directs it to, these ChatGPT types of models allow the reviewer to ask questions about the data, to the data.
When properly applied, this replaces running dozens of complex searches and then reviewing hundreds or thousands of hits, with simply asking a chatbot a conversational question.
These models are extremely powerful and can handle cases of any size. Unlike the public facing ChatGPT, similar versions built on top of legal platforms also site all sources.
By asking your chatbot a simple question like: “When did engineering know there might be a design flaw with part number A5621?” You could receive a response such as, “Likely in 2005”, followed by a list of links to documents your AI model relied on to give you its answer.
That’s the exciting part. The downside with any large language model is the answers you receive are only as good as the data it analyzed. While the algorithm itself is unlikely to make a mistake or deceive you, the data you are pointing it at might.
To be clear, as of August 2023, I have not seen a version of this type of technology rolled out to clients. I have seen amazing demonstrations of these models and our team is currently putting many of these technologies through their paces with sample data. Most of our software partners are attempting to get live versions of this technology released for active case use this calendar year.
Natural Language Processing
“Natural Language” refers to any language humans use, such as English, distinguishing it from languages that computers use. “Natural Language Processing” (NLP) refers to an AI model that can analyze and understand human language.
While NLP technology emerged with Continuous Active Learning (CAL) models in the 2010s, its advancements under LLM generative AI technology have significantly enhanced its accuracy, level of detail, and effectiveness, making it far more powerful.
Types of AI models available in this category are:
- Sentiment analysis
- Linguistic analysis
- Name normalization
- Topic detection and categorization
- Text summarization
- Emotional analysis
- Behavioral analysis
Let’s explore sentiment analysis to understand the output these models offer. A fundamental aspect of sentiment analysis is classifying the polarity of a given document, sentence, or even an entire custodian. This analysis can reveal whether the opinion expressed in a document or set of documents, is positive, negative, or neutral. There are advanced versions of this that go a step further by categorizing the emotional state of an author as enjoyment, anger, distrust, sadness, fear, or surprise. There are even NLP models specifically tuned to detect the presence of fraud.
When partnered with a case theory, these AI models can go a long way toward proving or disproving theories about what occurred. By knowing more about the mindset of the author, the purpose of their words, and often their subconscious intent, you can develop a deeper knowledge of your discovery and the players involved early in your review.
Computer Vision / Photo and Video Intelligence
This newer and emerging segment of AI modeling specializes in image and video classification, object detection, and segmentation, showcasing the model’s ability to process visual data and gain insights from it, without ongoing human intervention.
While not often needed or used, these models can be helpful in specific instances. These technologies use a combination of pattern recognition and human training (who or what are we looking for) to do the preliminary work of countless people.
Anyone who has ever poured over hours of video, looking for the needle in the haystack, can appreciate this use of technology. These same systems can often programmatically transcribe the audio from electronic audio and video files, making the speech fully searchable. In some cases, you can even be directed to the exact moment in the recording the phrase you searched was said, alleviating the need to search through hours of tape.
Unlock the Potential of AI in Your Practice
AI can be a game changer in your legal practice. By harnessing AI-powered tools, you can expedite the review process, make better-informed decisions, gain a competitive edge and significantly improve overall efficiency. Now is the time to take advantage of this powerful tool, before your adversaries beat you to it. Contact us to learn more and get started.
We all know when something seems too good to be true it usually is. That’s why next week we will explore the limitations inherent in AI models like ChatGPT.