What is the economic value of AI and AIOps?
In this article, Will Cappelli explains why any business that does not deploy AI will either be afflicted with making really terrible decisions or being drowned by the fixed costs of making good decisions.
AIOps is nothing but the application of AI.
So to understand the economic value of AIOps, we first need to define AI and understand what elements of AI are relevant to IT Operations.
However, AI is a relatively ill-defined term and means many things to many people. Within AIOps circles there has emerged a conception of AI that is different to many proposed definitions and it does shape AI and makes it particularly relevant to AIOps.
What is AI?
So what is AI all about? In the AIOps world, we look at AI as the reverse engineering of algorithms that underly the brain’s cognitive processes which have evolved over time to cope with, and to manipulate, complexed rapidly changing environments. Let’s look at traditional computer algorithms – these were designed to repeat basic operations multiple times and process large, difficult to work with, repetitive structures that don’t change during the course of computation.
Since its inception as a research programme, AI looked at the way the brain worked and instead of building algorithms from the bottom up, to deal with those large, static, repetitive structures, it was designed to imitate the human brain and develop algorithms to deal with a changing environment. But what does AI actually yield? How can we make money from it?
The five definitions of AI
Fundamentally, the yield is five distinct types of algorithm. The first type of algorithm takes a stream of data that’s coming at you from all sides and selects which part of that stream, and data subsets, are of most interest and require further investigation. This is the data set selection.
Once data has been selected, the second type of algorithm works on discovering patterns in the data – how does the data make any sense, how are they relevant to one another? You are ultimately looking for data insights. When people talk about machine learning or deep learning this is the type of AI they are describing, the one that is causing most business excitement. You start by feeding the machine data and out of the machine pops some kind of pattern which structures the data.
Now onto the third type of algorithm…imagine you’ve got patterns which are describing datasets, and you want to draw out all the information which is implicit in these patterns. These are automated inferencing algorithms. This type of AI caused much excitement for business, military, and medicine back in the late 1980s. It was the heyday of expert systems and automated inference engines. Known as knowledge engineering, the assumption was that you started with basic patterns (not created by algorithms) and you would then run a further set of algorithms to drive inferences from the patterns, which were then fed into a machine. This third type of algorithm is commercially very important and far more pervasive than many people realize. It just occurred many years ago and is now embedded into our computer systems and society, so today it’s not even classified as AI.
The fourth type of algorithm takes the results from data selection, pattern discovery and automated inferencing and communicates them, whether it be to human users or other applications, and wraps them up in such a way that they can be presented, communicated and discussed. Here is where previous work conducted in language processing application bears fruit – these algorithms start with some kind of content that has been sourced via inference or pattern discovery and then converts that content into sentences which eventually, via the system, is communicated to a human user.
Finally, the fifth kind of algorithm is the transition of AI into robotics. You take the first three algorithms I mentioned and instead of communicating it to a human being you communicate it to a machine, which can then act on what has been communicated and respond to that contact accordingly. This algorithm is the transition from communication to action.
It all comes back to the brain
In terms of all the research and commercial work being done in AI today, you could argue that it all fits into one or more of those five algorithmic AI categories. But let’s take a step back, let’s go back to AI being a research programme, a reverse piece of engineering of cognitive processes. Well, if you look at our cognitive processes, they pretty much break down into those five categories – there is perception, there is pattern discovery, there is reasoning in terms of thinking about those patterns, there is communication, and then there is action. This is true for any kind of cognitive model, be it logic or mathematical based.
When we look at modern computer systems, we could describe them as being highly complex, highly modular, highly dynamic, all in a fluid environment which is changing all of the time. In order to apply IT Operations to modern IT systems (to do the monitoring, to do incident management, etc.) the algorithms need to understand what is going on ’under the hood’. Today, you need algorithms which are specially designed to deal with complex, rapidly changing environments which have been reverse engineered from cognitive processes. This is why businesses need AI – this is what AI does.
What about bad data?
How does data quality impact on all of this? It’s a huge issue. However, the first type of AI algorithm I’ve described deals with the data quality issue and this brings up a very important point. The idea that you need to purify data before it’s fed into an AI system is really a non-starter in the modern world. You no longer have the time or capacity to take all of the junk and redundant data and have a human being or a methodology process it before it gets fed into the AI system. You actually need AI to do the data clean-up.
Almost all vendors of AI technology don’t recognize the first layer of AI I have outlined. And this is one of the reasons why this data quality issue comes up. The general assumption is that somebody needs to clean the data before it’s ready for AI processing, and that’s just wrong. The AI has to be involved immediately, and that’s why the Moogsoft platform takes all the junk and processes it. Ultimately, I think many vendors don’t appreciate what is needed to deliver practical AI.
How do we make money from AI?
There are a number of different ways to look at the economic potential of AI and AIOps. I like to look at where these technologies are going to impact the business because that’s where the true economic value is most prevalent. In all honesty, most enterprises see AI as a means of reducing cost, or at least enhancing efficiencies – not necessarily getting rid of people but making them more effective. Therefore, most new technologies are initially seen as a labor-saving device. That’s the simple way of looking at it.
But I think there’s more of a subtle way of looking at it, which differentiates AI and AIOps from other kinds of technology…When we look at what has happened to business over the last ten years, one of the major things is the pace at which business operates and one key consequence of this acceleration has been the need for business agility. Because things change so rapidly, the decision-maker needs to be able to make new decisions in line with the pace of the business. I would dare to say, that the number of decisions that now have to be made during a quarter has increased phenomenally over the last ten years. Formerly, you’d lay out a quarterly strategy, and the rest of the quarter was really about execution. In the world of IT, the timescales were even longer – more like an 18-month cycle. The velocity of business no longer tolerates this approach.
The cost of making decisions – good and bad
There’s an underlying economic fact which impacts all of this. The cost of making a decision, the cost of gathering the data, the cost of identifying patterns in the data, the cost of making the correct inferences, the cost of communicating the decision, and then the cost of enacting the decision…yes, we’re back to those five categories again…in effect, we’re looking at a fixed cost of decision-making, and the cost does not decline as the number of decisions increases. Why? Well, you still have to gather the data, you still have to see the patterns, you still have to make the inferences…almost by definition if you corrupt the quality of your decisions you reduce your agility.
You don’t just want to multiply the number of decisions, you want to multiply the number of good decisions. However, your cost of decision-making still remains fixed. So what happens to the business? It’s almost a vicious cycle, business velocity demands increased agility, increased agility demands multiple decisions, multiple decisions increase the velocity of the business, and the cycle goes around and around. This is all wonderful, but if I’m right, an increasing cycle of decisions drastically increases costs, even if the decisions you’re making are good ones.
So in effect, you’re either making bad decisions and damaging your business, or you’re making good decisions and incurring more and more costs. Eventually, those costs will cause your business to fail. Enter AI and its economic value.
The economic value of AI and AIOps
If you look at the five components of what a decision involves, it correlates to the five algorithms I’ve used to define AI. By automating those five aspects of the decision-making process you stand a chance of reducing that fixed cost and hence enabling a virtuous cycle rather than a vicious one. This leads to increased agility and increased number of decision points – but the decision points are now less costly and the quality off the decision isn’t detrimentally affected.
To tie this to AIOps, when you look at a modern business, one of the key things that has driven this increase in velocity and has enabled agility and the multiplication of decisions has been the transformation of digital business processes. So you now have business processes which are infused top to bottom with IT. What does this mean?
SEE ALSO: New technology rises: AIOps aims to facilitate, unify and modernize existing Ops processes
It means that many of those decisions are ultimately decisions about IT systems – how to fix problems, how to adjust systems, how to enhance performance…Ultimately, the decisions you make circles back to the five categories of AI outlined earlier, but this time focused on IT systems. So, although the general economic value of AI is about reducing the cost of decisions, the majority of decisions being made concern IT systems. Therefore, a big chunk of the very real economic value of AI to the economy derives from decisions being made about IT. This is how the entire picture of AI and AIOps combines into a single coherent narrative.
Any business that does not deploy AI will either be afflicted with making really terrible decisions or being drowned by the fixed costs of making good decisions.
What about the skeptics?
Is there skepticism about the economic value of AI? Of course. There is a school of thought that it’s all just rubbish, and then there is a school of thought that believes it works but what it delivers is not of sufficiently high value to justify the investment. Frequently these two kinds of skepticism fade into one another, and the skeptic does not know where he or she stands on the spectrum of skepticism. Naturally, over the last couple of years, there has been a business level excitement of the potential of AI. However, when you try and press the opinion makers who have stirred up the enthusiasm, you tend to get vague and mystical answers.
This is due to the fact that the technology itself has not been defined very well. If you say, “here I’ve got a bag of unicorns”, people who don’t get to see the unicorns are going to be skeptical about what’s in the bag. And let’s be honest, AI has largely been presented in business magazines and in the analyst community as a bag of unicorns. Therefore, from my perspective, skepticism is based upon a lack of definition as what AI actually is.
We need to demythologize the technology and ascertain how AI can deliver true value. Ultimately, if you look at it in terms of where it delivers value, i.e. in reducing the cost of decision-making, then I think there is a very strong business case. Remove the mystery and be very clear where it is impacting the costs that businesses incur. If we can do this, then AI becomes a commercial certainty.