AI Leadership Insights: AI and Sentient Documents

Mike Vizard: Hello and welcome to the latest edition of the TechStrong.ai Leadership video series. I’m your host, Mike Vizard. Today, we’re with Dr. John Bates, who’s CEO for the SER Group, also author of a book called Thingalytics. And we’re talking about the Centene document, which of course will be enabled by AI. John, welcome to the show.

Dr. John Bates: Thanks very much, Mike.

Mike Vizard: What do we mean by a Centene document? What does that do? Does it live forever? Does it know everything about us when we invoke it? Describe for us what is the concept here.

Dr. John Bates: Well, this is the concept of being able to actually interact with a document as if it’s a living being or a bot imbued with life. So imagine you can have documents of all kinds, whether they be contracts, whether they be orders, whether they be insurance policies, or even a handwritten note. And imagine you can ask the document about itself, what kind of document are you? What are the subjects of your document? Even if the document’s written in a foreign language, you can interact with it in your language and it can tell you all about itself. And in this way, it makes documents smart and takes the era of intelligent bots into the document era.

Mike Vizard: What has to happen for that to occur? Do I need to build entirely a new infrastructure for creating documents, or can I infuse that into the way we currently create documents? What’s the road from here to there?

Dr. John Bates: If you’re like a next generation framework for how to handle content, so it’s bigger than just documents, because of course, you could be interacting with video or audio or other forms of content. And really, I see the way the world is evolving. You used to have these content services or content management platforms where you store, manage, search, or archive your documents in your other content. And then of course, there’s this desire to automate processes around content, so automatically respond to invoices or automatically respond to human resources, onboarding when you get a new employee or whatever it might be.
And then of course there’s now AI natural language, both generative AI and AI pattern recognition. But really in the next gen of use cases, you’re going to need to bring all these things together. So you’ve got a content platform that can both store, manage, search, archive, automate processes, and also understand the content throughout this. And of course, on top, once you’ve got this kind of framework, you can then start building on top of it. And this concept of sentient documents is I think the ultimate application of it, where you’re really making it super simple and you’re delivering that AI and automation to a human knowledge worker in the way they’re used to interacting with, for example, their Alexa device at home. But you can use it to apply that to your enterprise content.

Mike Vizard: Does that also mean that the document will be smart enough to, for lack of a better phrase, issue a warning or an alert that says, I exist and if you don’t act on this, your life insurance policy is about to be terminated? I mean, how proactive can it get? I mean, can I get the, danger Will Robinson button?

Dr. John Bates: Absolutely. I mean, I like it. Good. Nice to have a Lost in Space reference. Yeah, there’s a couple of aspects to this. So first of all, there’s being able to sift through and get alerts about the enormous amounts of content that are coming into your business. I think there’s due to be about 150 zettabytes of content created or captured this year, and that’s 150 billion terabytes. So a lot of content is streaming in. And to be able to intelligently recognize things that are of interest to you or some of the projects you’re working on, it’s definitely one of those alerts that you’d want to know about. But then of course there’s also the patterns over time. So to your example, it could be, yeah, a document that was registered with you, you might want to have really an intelligent content assistant that’s like your assistant that can say things like, yeah, your insurance policy’s going to expire, you should start to shop around for this.
Or remember, this contract’s going to need to be renewed, get the sales team on it. Or indeed, a more real-time one, which is let’s say you have been processing invoices and you’ve been working with certain companies maybe for 10 years. Now, an employee with institutional knowledge would know that every Tuesday, Mike Vizard sends me an invoice and it’s usually within these ranges and this is his bank account information. But in fact, in this case, I’ve received it on the last Wednesday of the month and it’s a different bank account. Say perhaps this could be fraud and that’s your danger, Will Robinson alert, I think. That’s what I call contextual AI, which is not just the pattern recognition of documents coming in, in terms of visual or natural language, but it’s more about the temporal patterns over time and looking for anomalies.

Mike Vizard: How would my documents that are “aware” interact with documents that other people have that are aware and how would they start to establish something of a relationship in a workflow?

Dr. John Bates: Well, I like it. I think documents should be able to have relationships in this modern world. But no, you’re absolutely right. I think the world has now gone to a, in the enterprise certainly and probably a lot of the upper mid market of business to a best of breed architecture. And what I mean by that is there might’ve been a day where you bought everything from Oracle or everything from SAP, but now the best of breed architecture means that you might have your ERP system from SAP, you might have salesforce.com for your CRM, you might have ServiceNow for your service automation. You might have the Microsoft Office suite and so on. Somebody else might have a completely different set, Workday, Microsoft Dynamics, whatever. And each of those systems is great because then your line of business can work in their chosen environment, but they’re also from different ecosystems and often create an information and a process silo.
So there’s documents in there, there’s process pieces in there, and they don’t necessarily interact. So one of the pieces you’ve got to deliver in an intelligent content automation framework and then enable your sentient documents to be able to access is to go across those ecosystem boundaries. And indeed, in that case, if we’re looking to get documents together, you might have an invoice coming in that a new in your world, but you might want to cross-reference that again, it’s an SAP purchase order information for example. And again, there’s different types of AI going on there. There’s the temporal relationship, there’s the sort of pattern recognition relationship. And indeed, perhaps a lot of content also in the enterprise is dark data. So consider your Team’s channels, for example, or your Zoom channels. You do all of that good stuff, you interact with your colleagues and that stuff, presentations, chats, maybe pictures, but then what the heck happens to that afterwards?
In theory, it exists, but it exists somewhere, how could you find it? So it’s even potentially that a document comes in that may be somewhat related to some of that dark content that was interesting to you at the time and say perhaps it’s an opportunity for those documents to interact and come to you together and say, “Hey, we’ve met each other.” [inaudible 00:09:16] You would give a presentation and look, there’s some more up-to-date information that’s coming from this particular company that you were interested in telling your colleagues about. We’ve got together and we’ve set up a meeting for you so we can tell you about all the latest updates. And I think that’s exactly how unlocking dark data. So you want to record all that stuff, but then you want to be actively thinking about how do I cross-reference this and documents coming in can sort of be really useful and get together.

Mike Vizard: How would I track the documents over time? Because if they’re constantly being updated, but I still have a need to go back in time and say, on this date, we agreed to X, Y, and Z and I need that record somewhere. So where does that manifest?

Dr. John Bates: Well, I think the content can live where it lives. The key is to get the metadata about that content. So you want to, whenever content’s being created or captured, you want to somehow use AI to extract the metadata. In other words, a description about what that content’s about, and it’s that metadata that you can then use to record it. And I think there’s a concept which we call a metadata kernel. So imagine that you’re automating, you’re storing, managing, searching. You’re using AI to unlock or AI to create. It’s all around metadata. And so these things should be interchanging. These technologies should be interchanging metadata. And indeed, any search you’re doing or any interaction with a sentient document should be able to get this smart information by unlocking metadata. Even if those contents are in different systems. They’re federating the content, it can live in those systems, but you’re unlocking it with that metadata and having AI bots that know how to search, create, generate metadata.

Mike Vizard: Will this streamline the number of documents we have? Because I think it’s fair to say the average laptop is a mess of documents that are often redundant with each other. So would a sentient document encounter another document and determine that it had been superseded and therefore terminated itself?

Dr. John Bates: I think definitely. I don’t think that original document will ever disappear because what you’re seeing now in this world of where we’re generating 150 zettabytes of content this year, probably be 180, 190 zettabytes next year. You’re always going to need to have records, versions, and logs about what happened with content. But what it will do is simplify it for you, because you won’t have to know about those different versions. So in your world, yes, the document will supersede and you won’t have to have all these documents hanging around. If you say I always want to have the latest version, you will have the latest version. If you always want to get a summary, you may not even want to read the document. You may want say, please, let’s say your intelligent content assistant is listening.
We have one we call Doxy who’s responsible for managing all of these sentient documents. I want to say, “Doxy, I just want to get a summary of my insurance policies and I just want to know when I need to renew them. Or actually, why don’t you renew them for me?” That’s going to be very possible. So you’re going to be like President Reagan. He said, “Look, don’t give me all the details, just give me the summary.” You’re going to be like President Reagan in your own world.

Mike Vizard: There you go. Of course, whoever creates the summary was actually in charge. But that’s another question. Let me ask you this though. Are we a little bit in danger of becoming overly dependent upon AI systems? Because we as humans, we tend to get lazy. We tend to trust a little bit too much, and if we have some sense of instant gratification, we don’t think too hard. So how do we make sure that we’re not automating the wrong thing until we discover that its way too late to fix whatever mess we created?

Dr. John Bates: Oh yeah. Well, definitely AI is not going to be a substitute for humans. It’s going to be a scaling capability of humans and a way to magnify the power of the knowledge work of I10 to 100 times. But what we have to really be careful of, I fully agree with you, is to spot the anomalies and to spot the exceptions, and we need to be able to flag those up for a human to make a decision. What we don’t want to have is AI making unqualified decisions because these AI, they’re keen… Imagine they’re like your apprentice, very keen, actually you don’t have to give them a bio break. You don’t have to give them any holidays, vacations. So that’s good. They’re very, very keen to progress, but they don’t know a huge amount and they can get things wrong.
So we have to make sure, and this may even become something at some point that’s legislated in terms of whether it be ethics or rules around AI, we have to make sure at the appropriate point, humans are in the loop, particularly in things like healthcare, because you know very well that AI is going to triage pictures of cancer and really probably just as well as humans can and spot because it’s been trained in a deep learning model what pictures of cancer. But we still need to make sure that a human is ultimately making the triage decisions. This is just a decision support technology. And the same is true of business documents. We don’t want to be making wrong decisions, so we have to make sure that the right things are flagged up to humans.

Mike Vizard: What will happen when my sentient document that is optimized to buy something at the lowest price possible for the highest quality meets up with your sentient document that is designed to sell the thing at the highest price possible and preferably at the lowest quality, will the two of them meet and cancel each other out somewhere and humans will have to intervene in that conversation? Or how ultimately will opposing sentient documents work these things out?

Dr. John Bates: Well, I think you’re painting an interesting scenario. I mean, so far, I think sentient documents are really just around helping people to understand content, to summarize content, to make decisions around content. But you’re absolutely right. The scenario you’ve just painted is possible. And we know how that plays out because that’s called algorithmic trading, which you might remember we spoke about many years ago. And to some extent, the financial markets have been, the capital markets have been ahead on this area and they’ve had bots war with each other, and they’re not quite sentient documents, but they’re certainly trading algorithms. One of them’s trying to find a pattern before the other one and trade on it, and they’re trying to essentially go to war. And so we know how that plays out. The one that makes the best decision or act quickest is going to win.
But I don’t know if we’re going yet to algorithmic trading in business. That might be for the future. I think the first step is to help deal with the complexity and all of that content out there that’s dark, that’s unstructured, that’s streaming into your business that you need to be able to handle because we just can’t bring in as many staff as we need to handle it, and the humans are super expensive, so we need to make them much more productive. That’s the first battle. I think we’ll come back to this in a few years, Mike, and see if we’ve been wiped out by sentient documents. Hopefully not.

Mike Vizard: I don’t think we’ll be wiped out as much as we may be needed more than ever. But folks, you heard it here first, sentient documents, they’re coming. It’s just a question of how we’re going to basically first create them and then B, retain some control over them and C, not get overly dependent because remember, don’t get lazy. Hey John, thanks for being on the show.

Dr. John Bates: Thank you, sir. Nice to see you.

Mike Vizard: All right. Thank you all for watching the latest episode of the TechStrong AI Leadership Series. You can find this episode and others on our website. We invite you to check them all out. Until then, we’ll see you next time.