Tech On Trial

We're still on the path of talking about legal, judicial, and general law with specific AI.

Today we're going to talk a little bit about the work of having a machine understand some of this law.

Clause extraction is often just looking for a way to segment certain documents, whatever clauses are you extracting from and it has many methods of extracting those. A clause is a self-contained paragraph or a section of text or content on a page, generally, it's not incredibly hard to pull them out. The difficulty starts happening a little bit more when you're trying to understand what those clauses are about, sort of classifying them.

There are a lot of possible classifications of these paragraphs or clauses which makes it like it's that next step beyond just separating them out. If you get into the legal domain or the legal world, basically the language inherently in those clauses is obviously going to be a little more technical.

So there are more caveats and specific understanding required of your machine learning or NLP programs to actually extract.  Pulling the paragraphs out is not that hard, but once you have them out, it's again, understanding them from the legal perspective as to what they're for.

With this in mind, the FILAC Model was developed as a constraint to a specific domain of law to help structure the information. 

There are other techniques especially in neural networks, in deep learning where without supervision so an unsupervised model, you can cluster information, helping group the different clauses and the intent of those in a way that suits your workload better.

So that sort of wraps up us talking about our language models a little bit, talking about clause identification, extraction and classification, and the whole building up of legal AI.

Did you know companies can manage their legal documents using NLP and AI?

So imagine that you have all kinds of PDF docs, some of which may have been scanned, so after that, you need to understand the scan itself, which takes time as well.

There is a technology called Optical Character Recognition, OCR and that's often used to get that data in and turn it into printed documents, basically understandable computer documents.

You now have text, which is a big important piece, but you need to understand the content. Your NLP and AI program are gonna have to do some kind of whizzbang magic to basically understand what's going on in all of that content.

Most good programs will generate internal metadata about the document or the content itself, it might even go down to the bolded words being there for emphasis and the italicized words being there for another reason.

That can also lead to an adjacent to summarizing, so you might wanna summarize the document so that you don't have this a hundred-page. The idea is to get the general concept of the whole document.

Today we're gonna look to talk a bit about the life cycle of AI projects and how they're undertaken, whether that's an internal or external project because they can both go well or poorly.

 What we found was unless certain things are in place at your organization or with your people staff expertise, really experience, a lot of internal AI projects kind of go awry and don't exactly get to the outcome that's always desired or that was expected.

The quality of data relates to whether or not an internal AI project will succeed so we're kind of holding everything else aside and saying how do we isolate the data and ensure that the data is not the problem, cause that's really what we're trying to get to here.

So you wanna make sure that you have very complete data in that sense because having just a little bit of the signal and only understanding.

Missing data is going to lead to certain things weren't observed, and certain things weren't saved or cataloged, or curated, so you don't have a full picture. 

You only have a signal or two,  at this point is kind of like a needle in a haystack,  if you don't have a complete data set, and of course, just generally bad data is not gonna help you.

So you wanna make sure that you have data in a way that's going to be usable down the line,  clean data, so doesn't have all kinds of weird noise and bad signals in it, and you wanna make sure that labeled this data well.

Another problem with all the silo data is we often end up with duplicate data sets. So we may be describing the same thing in slightly different ways, but across the organization, you will duplicate the effort, wasting people's time, and having systems that we're probably paying for, for no reason, that is saving the same thing.

 Today we're gonna focus on the interplay of ERP with AI and NLP, what 1000ML does and hopefully get to a place where we can understand how you can actually benefit from each other, and at the end how organizations can get a real return of investment.

ERP, Enterprise Resource Planning, and the main objective are cataloging the types of things that are in your organization. 

The crucial things are the company resources such as people, places, or things and then also the material invoices, purchase orders, contracts, and all that stuff along with your inventory. The idea's sort of a centralized brain of all the things that make your business run.

 ERPs capture and understand complex workflows and processes with all your resources, so, AI by itself is a largely bits and bites computational engine, or I guess inference and prediction engine doesn't do well with unstructured text. You have to find a way to make those documents into something that the AI can actually understand.

So that's where the entry point of NLP sort of exists here, and NLP gives AI the ability to source the information from those documents.

Today, we figured we would talk a little bit, but the considerations of when you are accountable or responsible for these projects themselves, and you want to ensure that you have the best return on investments.

We touched last week about all the things of an executive's train of thought that would go into such a project in an NLP and AI. The executive isn't necessarily going to dig into the implementation of NLP projects and they're not really going to source data for you, the executive wants to rationalize in their mind, do we have sufficient and the correct kind of data in this organization, or can we source it?

There are ways that you can actually look to get data and look to acquire data points that you may not have. Either you, create data and create metadata as needed, or you actually buy them from some catalog logs or databases.

If we're talking about NLP projects, we're often largely talking about unstructured data, largely it's written data, it's type data obviously, but it doesn't have the structure of,  which makes it, that you have to create the structure from it. So that's often where the executive's mind needs to be, are they need to ensure that the enterprise or organization has enough NLP expertise in its staff, obviously that they can properly manipulate all this text.

It's like a mini-layer cake. It takes some work and it takes some foresight, but it's definitely doable. And it's possible, to think about this as an in-house project for sure.

Often, it also takes ingenuity and foresight, and thinking on the cultural level. So, you want to be an organization that is thinking about, innovative things and also considering new kinds of algorithms and new kinds of methods to do things and not kind of just stuck in what's working and what's paying the bills at all times.

We've talked a lot about  AI, data science, NLP generally. And we've talked a little bit about some use cases, but today let's actually hit home with the payers of NLP. Today,  we're very specifically and deliberately going to be talking about how NLP affects your business from an executive's point of view.

Regarding NLP, you get to a place where you understand that the content of language needs to be diced up and made digital in some way so that you can use computers, especially AI, to get some outcome, whatever that outcome is. So why don't we just hit home with the general idea of NLP because that probably informs what staff is doing?

So let's say you already have the content or you're putting in some documents of some sort that you've received,  the NLP programs and programmers know the kind of details you need to extract, they'll do most of that work for you. 

How would an organization go about implementing an NLP system? So if we take, you know, the decision out of the way that you're going to buy versus do it yourself, well, we may look at that later. But in, in general terms, your staff is going to look to ingest a lot of content. Now, you may already own and have that content.

Your time to deploy is usually a bit shorter and your cost may be about equal or not quite as high in the internal version. However, your support cost, your ongoing support costs. If this is not a key part of your business, can be too much to undertake as opposed to getting somebody else to support it.

So those are some of the considerations that you would have in implementing the NLP system at your organization. We talked a little bit about what NLP is to an executive, how it gets done at your organization, and the interplay between buying versus having your own staff do it.

Today we're going to talk to you guys about the myth that AI is here to kill you, and it's all going to result in this big singularity where AI gets so intelligent and takes over the world.

It's also not here to take over the world, it's taking over a few tasks. It's taken over a lot of computation actually sees it as mainstream.

Some of the earliest uses that companies and really companies and organizations started seeing with AI were really around,  intrusion detection, bot detection, spam detection, so things that the company may not have been really aware of that they even had AI in their organization. 

It's important to understand the general use patterns, of your websites, the applications of your mail, and everything, and you want to keep things around those metrics.

Another really common use has been chatbots, conversational AI, IVR so sort of like that first interaction with a company where you're kind of getting some information about them or talking a little bit about your problem with them.  So it helps in sales and it helps in customer retention and in customer service.

When a company is looking to acquire a chatbot, they usually want the implementation to cost of minimum for a super bear bones chatbot, that's just capable of answering some questions, but that brings some problems such as the configuration, where you're going back and forth, trying to really discover what your intended audience wants to talk about and how to have all the question and answers.

So there's a lot of tinkering and refining that goes on and on and on for a lot of long time weeks since chatbots have an issue with multilingual support just simply because you're the one configuring the chatbot itself, in different languages for different purposes.

The world of chatbots is kind of pervasive if you have a cell phone or almost any television service. A version of chatbots or chat technology that existed has been evolving more and more over time.

So along with that came or after that came, natural language processing, which actually gives us the ability to turn language into more machine-readable and usable data, a bot can generate speech based on some ideas. So it understands the topics or the kind of answer that you would want because maybe this is strictly a customer service bot and it's transactional.

In today's episode, you can understand more about the Chatbot evolution and how they work.

Most companies, large or small, work according to written information from written documents that contain the details necessary for different operations. This wonderful amount of data has patterns and trends. It is possible to understand them in order to make an informed decision. The ability to not be tired and mess something up is huge, NLP can basically help take the guesswork out of your hands. 

Talking about on employee productivity , if we replaced certain tasks which can be automated, a lot of what NLP really does for employees and just generally is find the needle in the haystack.

Most people prefer to do a bunch of research by themselves before they enter the sales cycle. If they ever enter the sales cycle at all, because they want to be informed and they don't wanna have buyers remorse though. This can sometimes also lead to buyers remorse if they don't go in there informed as well.