Sentient AI: A Q&A with IPSoft’s Chetan Dube

AI and machine learning adoption continue to be hot topics in the world of technology. Recently, Dan Woods had the chance to speak with IPSoft’s Chetan Dube about AI adoption and productization and IPSoft’s unique approach to these topics. IPSoft’s portfolio represents an interesting way to productize AI with different frameworks that all work together. This allows a company to have the automation it wants inside the framework and then increase the results from AI without having to play above its skill level and try to craft AI directly.

This is an edited Q&A of their conversation. The full podcast can be heard here.

Woods: It’s likely many businesses are going to mess up adoption of AI. AI has come of age in an era where open source is common and so there’s a temptation for companies to try to do it themselves. My view is that people in this first wave of adoption will do much better if they adopt AI products and focus on gathering as much data to deliver to AI systems as possible. What are you seeing in terms of the adoption of IPsoft and AI in general with respect to productized and non-productized adoption?

Chetan Dube: The biggest impediment to adoption of AI, outside of organizational change and organizational and stakeholder buy-ins, is the manually intensive overhead associated with adopting AI solutions. If you talk to just about anyone, they will tell you that even with an RPA tool, 45% are dissatisfied because it is manually cumbersome to install, configure, and manage.

Are you saying that these AI systems come as airplane dashboards that you then have to modify to get them right?

That’s exactly right. If you just take an out of the box AI solution and try to adapt it, you have to configure it and that cycle typically takes three to nine months. And it’s fairly intense – you need data experts, and the ability to take all the knowledge that is particular to a corporation and imbue it into these AI systems. An AI system is supposed to automate. Isn’t manual automation an oxymoron? Should we be manually building agents to automate and streamline routine tasks? And that’s one of the things where IPsoft has taken an interesting approach. We believe AI must be sentient. AI has to self-actualize itself. AI has to get to a point where it can actually be dropped as a platform and absorb information just the way humans do through empirical and observational learning. Sentient AI is the answer to the biggest challenge that is impeding adoption.

People are going to do better adopting products, but just because you’re adopting a product doesn’t mean you’re going to have a satisfying experience. The product has to have a proper scope and level of maturity. The system has to recognize the anomalies, do the automation, have some role in responding, and understanding the effectiveness of the response. And if it can do that, then it can be sentient in the way that it can be self-improving and active learning. But IPSoft goes beyond that as you can then observe what the person did to solve the problem and learn from that. The highest level of sentience is that not only does it learn from a feedback loop, but it also learns from what is outside of the feedback loop, correct?

You are absolutely spot on. Learning is the key to continuous improvement in outcomes that the AI system is able to deliver. What we call “machine learning” is mostly about identifying an utterance that you couldn’t understand and then retraining your vector machines or your deep neural networks to be able to recognize that utterance. But that can make you stuck. Humans learn with empirical observational learning, not just intent and utterance, synthesis or deconstruction to be able to say, “That’s what the intent was.” And that’s the difference in true AI systems.

Your product is improving the semantic model, not just doing pattern recognition. Normal productized AI seems to only solve part of what a customer could do if he used IPsoft. Other than automation and supplying data to the AI, what are the other aspects you think people should focus on to be the most effective users of productized AI?

Dan, the best analogy would be, seven out of the top ten banks use IPsoft and its cognitive agent. Six out of the top ten insurance companies use IPsoft and its cognitive agent. Are any two of them using it in a very similar way? No. What we have discovered is that you can bring a space shuttle to seven different entities, and each one of them will apply it very differently. And sometimes, it doesn’t leave the hangar too well. Sometimes, it’s taxiing on the runway. But most of the people are flying it and flying differently.

An example is to look at credit card volumes. At the largest banks, you’ve got a credit card incoming volume of 1.2 billion calls coming per year. Think about the cost efficiencies and also the efficiencies in customer satisfaction that you can drive by being able to transform that 1.2 billion credit card calls that are going to an offshore location to be done with onshore quality done by cognitive agents.

So IPSoft will compete based on your automation architecture. For a new emerging bank, you’re going to create a new architecture to serve the customer with automation of different types of things and then get people behind that automation?

Yes. I always ask, why should AI be any different than humans? You’re basically hiring people in your organization. You could say every organization is the same. No. It’s about how organizations use those people. AI is a platform, just the same way that humans are. And it is how organizations are using and leveraging this platform to be able to create differentiation, not just in their core, but also in adjacencies is what’s making a difference.

How companies leverage AI and AI technologies is differentiating their business and their ability to adapt and define new business models.

I want to go through the IPsoft product portfolio using a framework that I developed initially for analytics. Analytics offerings have different levels of productization. Level one is the custom kitchen, the raw platform. Then the next level up we call “dinner in a box.” This is like Blue Apron where you get things to assemble. Then, on top of that, you have an artisanal brew. Now this seems to be like the IPsoft layer, which is, “I want my half-caf soy latte extra hot.” There’s lots of freedom, but you’re safe. The final level is the value meal, which is the in-process step that is there for someone to use. You may not even know AI or analytics are happening. You just get a good suggestion. And the value meal isn’t usually provided by the vendor; it’s usually the productization of the lower levels inside a company. Given this framework, how would you explain the IPsoft product portfolio?

You capture it very well.  You find the exact same parallel of your architecture in the cloud also. You start at the base infrastructure layer, you have EC2 and S3. You move up a layer and then you’ve got a platform. You move up a layer and then you’ve got a middleware as a service platform. You move up another layer and now you’ve got the applications. And you move up another layer and then you’ve got business process services. And I think the same thing exists where you could start at the base level and you’ve got a whole bunch of tools at the infrastructure layer of AI, theTensorFlows and the different open APIs.

You’ve got a ton of these building blocks of AI. And you’ve got an incredible number of libraries for every kind of deep neural network and bidirectional LSTMs to recurrent neural networks. Just name it and there are technologies being made available for it.

But that’s like giving a person bricks and saying, “Go build your wall now.” Now you build on top of a model and where you see IPsoft graduating into is the artisanal level which is providing a person a smart Mensa kid.

What’s the definition of artificial intelligence? The purpose of artificial intelligence is to mimic human intelligence. Now you can go ahead and allow this kid to be able to learn your business processes. It will rapidly assimilate that. Now, what you see in your top stack, going into the value meal, is it graduated into something where people are saying, “Give me a banking AI.” Banks are saying, “My differentiation is not going to lie in how I do accounts and how I do credit cards and how I do wealth management and how I do basic processing of short term liability — these things are basic.” They’re the core that a bank will do. My differentiation lies at the edge. How do I provide my customers a differentiated experience? Should banks still continue to focus on the core? Or should they start to graduate into the adjacencies where the differentiation lies? Then they can add more value to their end customer experience.

And gain a deeper understanding of what the customers’ desires are.

Exactly. And that’s where we find your last layer, the value layer. Our product, Amelia, is graduating from the artisan where she is basically able to say, “I’ve graduated from the University of Michigan and I have now become a banker. I have now become a healthcare specialist. I’ve now become a retail specialist.”

IPSoft has Amelia, IPcenter, and OneDesk. IPcenter is where you start to understand all the APIs of the world. IPcenter seems to be where you learned how to connect in a sophisticated way to the IT systems, but now you’ve abstracted it away so IPcenter is the API umbrella that you use to connect everything.

That’s right.

Then OneDesk, on top of that, is where you learn the semantics of, and the automation of, the tasks. So you’re adding automation tasks, but you’re also, at some level, building a semantic model. And then Amelia is where you handle the customer interaction. And you now use all of that foundation that you have below. Is that right?

Yes. Our mission statement has always been that through the leverage of AI, we can create a more efficient planet. We started with autonomics. We felt that IT systems could be managed as self-governing, self-healing, and self-correcting. A manifestation of that is IPcenter. The fact that these network, system, and database devices could be self-governing and self-healing is a good thing. But there’s a big human element in the front office that is left. We need to be able to make sure that, in the end, we are in the business of seeing if we can mimic human intelligence. And that’s the cognitive layer, Amelia. Autonomic, IPcenter. Cognitive, Amelia. Front office, Amelia. Back office, IPcenter.

We are trying to find a way to converge the front office cognitive capabilities and the back office autonomic capabilities. That was the manifestation of OneDesk. The convergence of both the cognitive front and the autonomic backend to be able to provide end-to-end automation. Technology has come a long way to cognitive fusion with the autonomic backbone.

Implied in what you’re saying is that some aspects of an organization are more susceptible to automation. Have you seen a refactoring of companies away from less automatable infrastructure toward more automatable infrastructure?

Absolutely. Companies drift more towards standardization. They drift more towards data center consolidation. But then you need a digital catalyst—which is the AI accelerant—to be able to get to the next S-curve, which is an exponential increase in productivity. And then this curve gives you the launching pad of having data consolidated and having processes standardized. But after that, if I throw in an AI accelerant and a catalyst, I move onto who’s running my digital processes now? Is it still being run by people? Or are artificial agents starting to run it? I’ve digitized my processes. I’ve standardized my processes to such an extent that I can have AI agents run it. Now that’s an exponential curve. We are finding organizations rapidly growing on this digital Darwinistic curve, as they move up.

But, if you need something today in any organization, you call a front office. And you will say, “My laptop has got a problem. I need somebody to come in and see why this is not printing.” The front office is going to open a ticket. Then a back office resolver group, who’s expert at Mac, will pick this ticket up, and say, “Well, Mr. Woods, does it happen in your New York office or is it in your London office?” You send another message back saying, “It’s my New York office. I’m currently close to Wall Street.” Now there’s another response, “Have you tried updating printing drivers?”

That’s how any enterprise works today. Could you disintermediate the vast amount of ticketing systems in the middle? It could be a keeper of records in the back. Could you disintermediate the layers and layers of human middleware? What is required is a cognitive layer – a real cognitive layer, not a cute chatbot. AI needs a real cognitive agent to be able to understand and translate what you said in natural language to ITSM speak. And then my autonomic backbone could pick up that task and execute it for you, providing you on-demand, self-empowered service. That’s where the world will go to in 2019.

In the cryptocurrency world, there’s been some interesting hacks, attacks, and raids that have been successful not based on breaking into the systems, but using the business rules of the systems themselves to allow somebody to withdraw more money than was actually deposited. It wasn’t a problem with blockchain itself. It was a problem with the business rules on top of it. I’m sure you wouldn’t claim that your cognitive layers are going to be perfect. So what kind of governance have people put in to make sure that cognitive layers don’t do something stupid?

You are absolutely right. But we have not had a single customer security incident caused by our autonomic and cognitive technologies. And why is it? Because you have to have control gates of security. AI can learn by itself. It’s sentient, as you said. But we are going to have the control gate so that from midnight to 2:00 a.m., we’ll see all the learning. We’ll have a control gate over the Continual Service Improvement department that reviews all those learnings and says, “Ratified.  Ratified. Ratified. Ratified. No, this one needs to be improved. No, this one needs to be discarded.” So all those mindmaps that Amelia has built during the course of operation by observing experts and servicing them, they’re all going through a control gate that has said, “Yes; all this learning is good. You can now go ahead and engage in that.” If you really are paranoid about security, as most of our financial institutions are, you must have a gate that just says, “We will ratify.” Now 99% of the time, they do say, “Amelia’s learned.” Because in the end, Amelia’s learned only by empirical learning. Some human did something that allowed Amelia to learn.