A.I., A.I., .....Oh? I Thought It Was Something Else.

The excitement is running wild for artificial intelligence, isn’t it? At long last, all of our science fiction generated fantasies are about to come to true. And look, we’ve been collecting all of this data for a while now (we’ve got BIG DATA!), and we have a data lake, and we are all set for AI to figure out all of our problems and issues and…..wait – what do you mean it’s not the right data? All I have to do is plug it in to what I have and it just works – right? HOW long is it going to take for that use case? It’s going to cost how much? And then I have to start all over again for it to solve a different problem? 

Ok, maybe this is a bit of an extreme example. But I think - I know - that there is still an awful lot of misconception and misperception about where AI is and how it can work for our businesses. There is some very exciting technology out there that offers tremendous potential for our operations. But many of us are starting without a practical understanding of what and where the technology is in its development and are letting marketing hype and our hopes that our science fiction dreams are finally coming true lead us down a difficult and costly path.  Just as many have expressed dissatisfaction with their RPA journeys (in no small part due to unrealistic expectations), many will soon see that their dreams of having their own personal HAL 9000 are not going to be fulfilled – at least not yet. 

Artificial General Intelligence vs Narrow Artificial Intelligence

Let’s start where I believe the initial problem lies. The terms “artificial intelligence” and “AI” evoke very specific images and ideas for many that have been propagated by books, movies, and television. When we hear either of those terms, what pops into most people’s heads is what is referred to as artificial general intelligence (AGI). The IEEE P2755 Guide to Terms and Definitions in Intelligent Process Automation has created a formal definition of this term* that, in essence, distills down to a technology that has the ability to describe, discover, predict, prescribe, and deduce at least as well as a human does in multiple knowledge domains. Lt. Commander Data from Star Trek: The Next Generation is an example of AGI housed within a mechanical body (some would argue that Data might represent artificial superintelligence because of his ability to process information much more quickly than a human, but for sake of the discussion, we’ll go with AGI). In a practical sense, meaning outside of academic theory or research and development labs, this form of artificial intelligence does not yet exist. Are we on a path to that? It depends on who you talk to.

In a practical sense, what exists today is referred to as narrow artificial intelligence (Narrow AI). Narrow AI can still describe, discover, predict and prescribe at least as well as a human, but only for a very specific knowledge domain. The software that has beaten chess masters and professional Go players are narrow AI because they only function for that purpose. SIRI, Cortana, Alexis, and most of the other well-known personal assistance devices work off combinations of different types of Narrow AI, like natural language processing, and then pair them with an advanced search engine to be able to answer questions. They also utilize machine learning to constantly improve the way in which they answer similar questions. When we hear talk about AI, we should be approaching it with the mindset that we are talking narrow AI that is focused on a specific function and NOT technology that can do all things. If we are talking about transactional processing, narrow AI acts as the “brain” and another technology (RPA) acts as the “hands”.

The Big Data Conundrum

So, if from a practical perspective, we can level set for ourselves what we mean when we talk about artificial intelligence, then let’s talk about the issue that will make implementation of the technology more challenging for many of us.  Artificial intelligence subsists on data. Lots of data. It’s all about the data. Sounds like a broken record, right? When over the last several years have we NOT talked about the data. But much of the conversation has been around the data’s “bigness” and not necessarily about its “correctness”. It’s felt at times like organizations have operated on the hope that the more data they collect, the greater the opportunity there is for it to tell them something. The conversation then quickly morphs into a storage and data processing conversation. Don’t we have a data lake? Should we use Hadoop or Spark as our analytics engine? Should we use Tableau or Power BI to analyze and visualize the data? Do we need data scientists on staff? 

The bigger question to be answered is do we have the RIGHT data to have a meaningful impact, not just on analytics and insights, but on transactional processes as well? In many situations, we have outcomes data that would include key characteristics to allow it to be trended, sorted and examined in ways to identify commonality. But what about the steps that it took to get to that outcome? In many organizations, that may be “dark data” that’s gathered but not made readily available - if it is even being collected at all. If we want to move AI from an analytics tool to a transactional tool, this type of data has to be part of an organization’s data strategy. If your systems are not capturing that type of data (and because of audit and compliance obligations, it would be surprising if they were not) and you’re using RPA, most of the major tools capture in detail what happens to the transactions during its life cycle. It’s a start. 

It’s Machine Learning, Not Machine Knowing

Now that we can speak to the differences in types of AI and what the most important enabler of AI is within your organization, let’s knock off one more common misperception that I think drives much of the hype and will lead to much of the disappointment in the implementations from unrealistic expectations. Machine learning technology (which is considered a subset of AI, but often grouped within the category of artificial intelligence) is one of the most sought after and desirable forms of technology being pursued by businesses. Its ability to take in information, identify patterns, and adjust as it takes in additional data to provide predictive and prescriptive analytics makes it highly attractive to organizations looking to accelerate processing. 

The important thing to remember is it’s called machine “learning” technology, not machine “knowing” technology. It must be taught just like a human associate would through ingestion of pre-existing information (by taking in a store of existing data – a person might take in exiting knowledge by reading a book) or by observation and experience. While there may be base algorithms used, the technology does not automatically know your specific history, processes, or use cases. It must learn them. Can it learn more quickly than a human? It can, but as we have already said – it’s all about the data. Insufficient access to the right kind of data for the process your automating will mean that it will take longer because you must build it up over time or allow the technology enough time in observation to be able to learn. And by the way, machine learning technology will act as the brain. You will still need people or other technologies like RPA (or both) to transact. That’s part of the reason why so many people have taken to referring to any form of AI available today as “augmented intelligence”. 

What to Do from Here?

Great question. First, we need to settle down from all of the hype and realize that, while we are on the cusp of incredible things, artificial intelligence still has a way to go to fulfill all of the expectations that are in our heads. Will we ever get to AGI? Maybe. Call me a hopeful skeptic. But that does not mean that we don’t already have access to some incredible technology to help us solve some of our business challenges regarding cost, quality and speed. We need to be more realistic about how quickly it may be able to do that for us. Unrealistic expectations are causing a good part of the frustration with automation implementations that happen today. 

Second, if you don’t already have a data strategy tied to your digitization and automation efforts, then do it now. You need to know what you have and what you need to go after to be able to make the most efficient use of the technology that’s here today and what may come tomorrow. If you want AI and machine learning to help with transactions, you have to have transaction data. It’s more than where the data is being stored and what tool you’re going to use to mine it. It’s if your capturing the right things to be useful for more than just analytics.

*For a copy of IEEE P2755 Guide to Term and Concepts in Intelligent Process Automation go to https://standards.ieee.org/findstds/standard/2755-2017.html.

Have questions about RPA or Intelligent Automation? I’m here to help with Ask AJ! Subscribe to my AskAJ You Tube channel (https://www.youtube.com/c/AskAJ) and follow me on Facebook and Twitter @AskAJHanna.