You Asked, We Answered: 5 Top Questions About the Future of Enterprise AI

Every day, there’s another story in the news about a spectacular new artificial intelligence development. But, as exciting as mind-reading clothing and playing first-person shooters with your brain may sound, these gadgets will hardly affect your day-to-day routine anytime soon. What will imminently impact you and your business are all the cutting-edge advancements happening in enterprise AI.

But, for those of us who don’t have a degree in data science, terms like “Shared Learning” or “Composable AI” tend to seem abstract and difficult to absorb in terms of our own business needs. That’s why WorkFusion’s Co-Founder and CTO Andrew Volkov and VP, Data Science Abby Levenberg held the “Tech Radar: Forecasting the Future of Enterprise AI” webinar. They discussed the Tech Radar (which you can see and zoom in on above), breaking down all the fancy AI terms and mapping out what technology advancements will directly benefit your Intelligent Automation or RPA 2.0 programs. As you can imagine, such a fascinating topic elicited many questions, here are the top five:

As a layman I find it very hard to discern what RPA products have AI capabilities, how do I know for sure?

The reason we created this Tech Radar and held the webinar is to help you assess these things. You’re asking how to differentiate which robotic tools really have AI capabilities from the ones that claim that they do, but don’t really. That’s why we have mapped out the AI capabilities that are available now and that we consider important. If an RPA tool has machine learning, these are the capabilities you should look for. We will also soon release a very comprehensive guide that will not only explain what the capabilities are, but also tell you why you need to have certain things and what they would do for you and your business.

When a system keeps learning as new data comes in, does that mean that the bot is smart enough to validate the underlying algorithm or does a developer decide which algorithms to use?

Usually, algorithms don’t change very often. Depending on the domain and size of the data that comes through the system, it can be important to have those capabilities. When you talk about what type of algorithms that produce the best results for different types of problems, it’s quite prescriptive. That’s why WorkFusion’s product comes packaged with different kinds of algorithms. The ones that you see on the chart above are packaged against the type of use cases that we talked about. They go from what we call data extraction, where we use one set of algorithms, to classification, which is a different set of algorithms, then to ranking which is a decision routing, and then chatbots where you’ll use something else. To answer your question about developers choosing algorithms: WorkFusion chooses the algorithm, not a developer, because there is no developer. The cognitive bots are created by observing how people do their work day-to-day and the technology is able to choose a strategy of how to use an algorithm or a technique, depending on the problem. Every WorkFusion product release adds additional use cases, categories and algorithms. And even in a single package, if you’re talking about information extraction, which is data capture, one algorithm can be used for 3,000 examples and a totally different algorithm can be used on 100, 000 examples. That’s why they come pre-packaged and we add them as we go.

Now that we have GDPR and other compliances that are coming in, an individual has the right to request that an organization deletes all their personal data. In that case, how do we create data for machine learning?

That’s when we use something called Shared Learning, which is an area that we’re already testing with some of our clients. There’s a part of it called Differential Privacy, which addresses some of these new regulations. What it means is that there’s no private information recoverable from the data, but the data is sufficiently valuable to create an outcome that is just as good as the outcome that would have been created using the private data.

When regulations change, do I have to redo and relearn the machine learning process again? Or would differential changes be accounted for in the existing framework?

First of all, having machine learning doesn’t mean that you should discount rules when you have rules. Imagine if you have a self-driving car that isn’t taught to stop at a red light, that would be pretty stupid, right? So, we have rules that are guardrails for how the system needs to work. But to answer the question, the very first thing you do is that you would add that rule in front, because you have to use common sense as you transfer the process that exists today into the world of automation. Secondly, the software has the capabilities to relearn using statistical quality control, which is a small sampling of the work that the cognitive bots produce that is examined and allows you to detect a major deviation. So, the new rule can be relearned in a longer period of time, when a problem is caught and the system switches to relearning mode so the existing cognitive bot or model can adapt to the change.

Where do you see Intelligent Automation in 10 years?

We like to think of this technology as the next Industrial Revolution. So, I think there will just be more types of bots that are doing things that we don’t see them do today, but what they do is based on the same principles. We don’t believe we’ll deviate from that too much. There will be different types of bots doing more complex tasks, which are being done by people today, perhaps tasks that require that bots can talk to each other in bot-to-bot interactions that also can be learned. In addition, this technology will establish something called System of lntelligence, which is in addition to System of Records [an information storage and retrieval system that provides a centralized, authoritative source of data elements in an IT environment], which is already here, and System of Engagement [decentralized IT components that incorporate technologies such as social media and the cloud], which is almost here. This technology will enable enterprises to mind their assets and create a System of Intelligence that would essentially have all the business intelligence in one place and not only keep it there, but also deploy it to productive use to launch cognitive bots off of that knowledge.