Demystifying AI – Critical Questions for Business Leaders

Demystifying AI – Critical Questions for Business Leaders

Presented By Glen Hilford

Julia:

Hi, everyone. Happy Wednesday and welcome to, Demystifying AI, critical Questions for Business Leaders. I’m Julia, and I’ll be one of your hosts along with Angela. So if you’re here today, we assume you’re interested in learning more about AI, are considering an AI investment or have been part of an AI initiative. So on the internet, you’ll find lots of information about AI’s potential benefits, promising products, mind-numbing technologies, and endless amounts of hype. But you won’t find an in-depth objective, old eyed examination of AI in a business context and its potential effects on organization. So that’s why we’re doing this webinar to provide concrete, actionable information so that you, as a business leader can make informed decisions about AI. So today’s webinar is the first in a series. Today we’ll spend time examining AI at a high level and then the remainder of the presentation we’ll spend our time probing how AI interacts with and affects businesses. So, to start, I’ll hand it off to Angela, to launch a poll and introduce our speaker for today.

Angela:

Hi everyone, welcome. Our first poll of the day says, “Where is your organization on its AI journey?” So I will launch this now, please take a minute or two to select which option is most relevant for you. Okay. Let’s say about 30 more seconds. So Glen, it looks like we have some interesting numbers today. About 53% are in the researching phase of the journey, and 23% are actually in proof of concept with projects underway. So just about everyone has voted. I will close the poll now and can proceed with introducing our speaker. Glen Hilford is vice president of Corporate Development for Access Sciences. He’s been with the company for just over 14 years and is our resident AI expert. Feel free to scan the QR code to connect with Glen on LinkedIn and check out some more of his content. Take it away, Glen.

Glen:

Thank you, Angela. Well, why don’t we get started with a real world success story? Our example focuses on pipeline system that transports natural gas from production fields in the south to the Chicago market area. The red lines on the map represent the pipeline. It takes about three days for the gas to reach the market area. What if we could forecast demand three days in advance? It looks like we could save millions. We’ll expand on this example throughout today’s webinar. So what is AI? Well for one thing in today’s world, it’s not this guy. Let’s spend a few minutes taking a tour of AI, what it is, what it isn’t and where it’s most commonly used in mainstream business. First, is it new? Well, the original AI program, a game of checkers debuted 70 years ago this year. The term artificial intelligence was first used in 1956. So it’s clearly not new. Is it a fad?

Can you imagine life without your funds weather forecast? Or your bank’s facial recognition function? Global demand for AI services was 90 billion last year. And in five years, that forecast be just under a trillion dollars globally. It’s definitely not a fad. And finally, is it monolithic? No. AI is an umbrella term that covers many technologies and approaches. We’ll look at this in more detail in just a moment. How intelligent is AI? Today, AI can be defined as artificial narrow intelligence, a machine ability to perform a single function extremely well, often better than a human. Good examples are weather forecasting and sports analytics. AI should eventually evolve into what’s known as artificial general intelligence. Where a machine can be made to think and function like a human or a human mind.

C-3PO, the gold robot in Star Wars is one example, preferably without the quirkiness. Today, artificial general intelligence is not a reality, it’s not even close. One estimate is 2040. You may recognize Alan Turing, as the lead character in the movie Imitation Game. Turing is generally recognized as the father in computer science and artificial intelligence. In 1950, he created what we now know as the Turing test, when he observed that a computer could deserve to be called intelligent. If it could deceive a human into believing it was human. You may recall in 1997, IBM’s Deep Blue program beat the reigning world champion, Gary Casper of chess. Did that pass the Turing test?

Can Alexa pass the Turing test? The final stage in AI’s evolution would be super intelligence and self-awareness. That’s Stephen Hawking and Elon Musk, have expressed concern that this could result in the extinction of humans. Thankfully self-awareness remains science fiction, at least for now. What can AI do? Well, there are many academic definitions. I prefer this one that recognizes AI’s attempt to mimic and improve on human functionality. Some humanlike things that AI can do, include learning from experience, making complex destructions and decisions, recognizing and acting on visual information, interacting with human language, coordinating inter relationships and automating physical activity and processes. That’s a pretty tall ogre.

Diving a little deeper, few businesses need complex solutions like self-driving cars. We should focus on the AI techniques that address our most pressing business issues. What can AI do for us? Let’s look at some real world examples of these business challenges. The classic example of predicting the future is weather forecasting. Today’s amazing predictions are on only possible due to AI. We all use spam filtering every day, where emails classified into one of two categories, spam and not spam. Medical research and marketing have embraced AI for its ability to discover hidden patterns in data. A prominent example of visual object recognition is your phone spatial recognition function.

For better or worse, we all deal with intelligent process automation and I’ll put intelligent in quotes, when we interact with financial institutions. Language translators ability to convert unstructured text into meaningful data have certainly made my life easier. In connecting and extracting meaningful data through semantic relationships, knowledge graphs are a new concept for many business people, but are increasingly important as we need to seamlessly connect systems and data silos. We’ll look at this more in our next webinar. While we won’t focus on specific AI techniques today, I did want to show you the ones that align with these business challenges. We’ll do a deeper dive on these next time. I hope you can join us.

One observation that we’ll also talk about next time is that these foundational techniques, are frequently combined to solve complex problems. A component oriented approach, commonly known as composite AI. Returning to our example, we can see that our business opportunity could be a forecasting or prediction problem. Something that AI is good at. And if we’re successful, we can use that forecast to benefit our business. Why has AI exploded on the scene? Why now? AI techniques and tools have been democratized. And this is an important concept, as we look at our techniques next time. These make what used to be the domain of academics and deep pocketed corporations accessible to mainstream businesses like ours, but make no mistake. These aren’t simple, but they are accessible and achievable. Computing power has increased to make using AI techniques practical. 20 years ago, AI could predict tomorrow’s weather with a high degree of accuracy. Unfortunately, the calculations took more than a day to complete.

As AI techniques have evolved and become more powerful and accessible, new and novel use cases have been identified. Uber doesn’t exist without a brilliant idea and AI to make it a reality. Most importantly, today’s digital environments generate vast amounts of data, that in turn fuel and enable AI. As we end our overview of AI, let’s do a recap. Today, it attempts to mimic human functionality, can perform a single task extremely well, addresses many business problems and is fueled by enormous amounts of data. The invitation to this webinar promised to explore critical questions before investing in AI. That’s what we’ll focus on for the rest of our time today. These questions are intended to be provocative, to help you identify and avoid AI pitfalls within your organization. While some of these challenges we explore may be daunting, none of them are impossible. So before we begin, it’s worthwhile to reflect on the enormous benefit that AI can bring. Every day it enriches people’s lives, helps businesses increase revenue, lower cost and mitigate risk and makes the world safer and healthier.

So, are we prepared? Is my organization ready for AI? First, we should ask ourselves, can we identify valuable AI opportunities? Can we determine if AI is a viable approach to solving those opportunities? Can we successfully implement and deploy solutions, and then defend their results? Can we address the change that AI inevitably brings to an organization? And can we operate, maintain and govern an AI program? We’ll explore these questions in more detail during the rest of this webinar. One concept that’s new, even to those knowledgeable about AI, is an AI program. As business people, we tend to focus on individual projects and for good reason. Project investments are intended to deliver tangible and measurable benefits to our organizations, operating and maintaining one or two silo AI solutions is possible for a while, but this model is sustainable and it definitely doesn’t scale. Many organizations find themselves in this situation.

What’s missing, is what we refer to as an AI program that provides the framework to confirm business value and viability for a proposed solution to a business problem. To provision sustainable and robust data infrastructure that fuel the solutions, to curate or govern the interrelated data models and outputs or results that come out of the solution, to ensure that as business conditions evolve our solutions evolve with them, to drive repeatability, transparency and explainability, increasingly and concepts and challenges as AI continues to terminate the business environment, to support our users as they face change. And it will change. We’ll look at that in a moment. And finally expanding strategic opportunities. As business leaders, it’s important that we look beyond individual projects to see a AI is a strategic part of our business.

Are we at a disadvantage? How is our competition leveraging AI? Are we falling behind our peers? How do we measure ourselves against them? And where should we focus our efforts? Building AI capability requires investment, but how much is enough? Global financial institutions need to be very mature to safeguard assets and maintain a competitive advantage. As a result, they invest enormous sums to be self-sufficient and cutting edge in AI. But most organizations don’t need this level of sophistication. Is there a way to gauge our maturity with AI? How mature do we really need to be? And how do we identify the gaps where we require improvement? One familiar way to gauge maturity and identify areas for growth is by using a maturity model. These help measure maturity across key factors, ranging from strategy to data, to tools, to change management. They’re also used to on the level of maturity, that makes sense for a specific organization. Not all organizations are the same. The gaps between these two levels identify growth areas that require attention.

AI requires specialized skills, that span opportunity discovery, business analysis, data science, model identification, training, and testing, governance, change management and program management. Do we have these? Do we have all of these? Are our business analyst, AI savvy? Do we have the data science and AI expertise to implement and deploy AI solutions? Can we sustain and govern them? And can we successfully communicate, train and support our workforce when they face the change that’s going to enter their business lives? Returning back to AI savvy us, this extends traditional business analysis with the ability to look at a problem and see how AI could be applied or not. We’ll explore AI skills and organizational structures, in more detail later in this webinar series. Returning to our example, our AI savvy business analyst, restates the opportunity statement using AI terminology. Here the verb predict, is a clue that machine learning prediction, might be an option.

Software marketers can be very effective, especially when it comes to AI. If you’ve never visited the Chicago Bean, it’s amazing. And an excellent illustration for shiny object syndrome. We’ve all experienced it. It’s not new or limited to AI. As business leaders, we all know that using business requirements to drive M and A decisions is a key to success. Similarly, we should always define requirements before embarking on an AI project. This is doubly important for AI solutions and separating fact from fiction. In a caution here, many vendors use the term AI to hype a variety of products and services that are in no way intelligent, artificially or otherwise. Simply slapping AI on a label, doesn’t make the product an AI product. Buyer beware. If we aren’t careful, enthusiasm for AI can become a hammer looking for a nail. We should and overlook alternatives.

Here we’re going to detour for a minute, and talk about the difference between deterministic systems and probabilistic systems. Understanding this dear friends, is critical to our understanding of AI. Before the advent of AI, virtually all traditional IAT systems, were deterministic. We’re all familiar with these at the to top of the gold diagram, rules are coded into the system and then data is input where the rules are applied. One of a calculator’s rules is addition. If we input two and three and select the plus the result is five. Exactly five. All every time. In flexibility and precision, are the hallmarks that are deterministic systems.

Probabilistic, AI systems, especially machine learning are different. They learn rules by being trained using examples. The result is like training a parrot. It can learn to imitate human sound, but it’s not speech and it’s never quite right. Let’s look at an example. We’ll start by training the AI system or model. In order to train the model, we present it with both input data and corresponding results. A different model than what we’ve seen in the past. If we give it enough examples, it should learn, in quotes, the rule or the code. This is shown at the top of the green diagram. At the bottom of the diagram, you’ll see training data that are examples of addition. Two plus three equals five, seven plus one equals eight and so on. Using this training data, the model, once again in quotes learns edition. At this point, we have a train model. We’ll show how it’s used in production with the gold diagram. Data is input into the train model, and the model generates an approximate result.

The word approximate is important, and is the reason these systems are called probabilistic. Looking at the bottom of the gold diagram, we can see that using two and three as inputs, the result is approximately five. Hopefully very, very close to five, but rarely exactly five. The ability to learn and approximate are the hallmarks of probabilistic AI. This doesn’t sound very important. Why does it matter? Well, for three reasons. First, the ability for a machine to learn from past experience is the foundation of AI’s intelligence. In my opinion, without this ability, AI doesn’t exist. We’ll learn a lot more about this in our next webinar. Next, because input data, the trained model and the model’s results are interrelated, they must be treated as a unit. This concept is critical as we consider AI governance. We will return to this in a couple of moments. In returning to our hammer and nail analogy, sometimes AI might not be the optimal approach.

There are classes of business problems that are generally better to non-AI approaches. Process oriented systems, such as ERP or content management, transactional systems, such as finance and accounting. Have we developed requirements that are completely focused on our business needs? Are there non-AI alternatives for addressing a problem? And what are the pros cons of each? With some imagination, we can identify plenty of opportunities where AI can bring value without trying to force a square peg into a round hole. Sometimes, seemingly viable AI solutions don’t pan out. This can happen for any number of reasons, incorrectly identified model inputs and adequate historical data, an unidentified external factor that pops up, models that don’t produce accurate results. The list goes on. In other cases, even technically successful projects don’t deliver anticipated value. In one famous example, IBM invested 62 million on a cancer diagnosis and treatment solution. It failed miserably. With adequate due diligence, could they have cut their losses earlier?

Sometimes AI might not be viable at all. Have we identified an AI approach that directly addresses our business problem? Have we performed our due diligence to determine if the is valid and that it can reasonably expect it to produce the needed results? And can we make this determination early enough in the process to avoid prolonging an investment in a dead end? In other words, get out of jail free card or button. An investment costs pile up, it’s easy to throw a good money after bad. What’s known as the sunk cost fallacy. To illustrate these concepts., let’s revisit our pipeline example. We’ve successfully stated our business challenge using AI terminology, and clearly identified the output that we want. Behind the scenes, we’ve identified machine learning prediction as a promising solution and identified the input data that would potentially be the solution. A side note here. This is an art, not a science. Equip with this information, we can make an informed decision about AI’s viability for this opportunity, and we can do it early enough in the project to avoid the sunk cost fallacy.

As we compete for limited budgets, we have to be able to demonstrate value, in terms of revenue and margin, reduced cost or reduced risk. ROI and value should always drive investment decisions. In addition to solution, viability is a concern. AI initiatives can be expensive in terms of monetary investment, high value resource time, opportunity costs, organizational turmoil, and personal professional and organizational reputation. Developing a clear value proposition, business case and ROI upfront can mitigate these risks, and provide a tangible way to measure a project’s value. Returning to our pipeline example, we were able to calculate that a successful implementation would yield several million dollars of operational savings annually. These are bottom line dollars based on better insights.

Before attempting to calculate a solution’s return, we first need to identify its real costs. And the technical solution is only the first consideration. Once appropriate data inputs have been identified, they must be sourced, either provisioned internally or purchased and a data infrastructure or plumbing must be created to ensure data availability, cleanliness, consistency, timeliness, and governance. Poor data always equals poor results, especially in AI. It’s unusual for a business system to function in isolation and AI solutions are no different. We should anticipate that an AI solution will be integrated with other business systems, adding to the overall cost.

As we noted earlier, as business conditions evolve, supporting AI solutions must evolve with them. The cost for ongoing retraining and maintenance of the solution has to be factored in. While we haven’t talked about governance yet, there is a non-trivial cost for curating AI solutions. Recall the data model and results sets that we talked about just a moment ago. And finally end user communications training and ongoing support is critical for solution adoption. Now equipped with this information, we should be able to calculate a solutions real ROI and make an informed investment decision. Study after study, shows that lack of adoption is a primary cause of AI failure. The premise and promise of AI mean that some people’s jobs are going to change. Change is hard. Are we equipped to anticipate, understand and act on it?

Are our employees prepared to shift familiar tasks to AI and work alongside the machine? Are we prepared to address the resistance that this will foster? Employees jobs often change with the introduction of AI, frequently for the better. Humans can glean insights from data that humans can’t and these insights help employees make more informed business decisions. AI can also automate an employee’s road activities, freeing them to apply that time to higher value work. A benefit to the organization and the worker. The challenge is that employees will be working alongside machines or computers, a change that’s often poorly communicated, and they can foster resistance. When we implemented the demand for the pipeline, it replaced the seat of the pants judgment of pipeline system operators, experts with 30 plus years of field experience resistance. Resistance isn’t as strong enough word to describe their reaction. If an organization isn’t aware of and prepared for these issues, they will have an impact.

Recall our discussion about the probabilistic systems? The probabilistic nature of AI solutions means that familiar governance structures must change to be effective. Are we equipped to govern and curate an interrelated set of data models and results in a way that supports repeatability and explainability? Those questions will arise. AI can introduce… Pardon me. AI can introduce new ethical questions. Do we have the experience and expertise to address these issues up front? And the auditors always get a say, “how do we respond to their audits?” Our final topic isn’t a question or a caution, but a challenge. We’ve all experienced it, that moment when the light bulb comes on. AI delivers value, sometimes incredible value to organizations like ours. As leaders, do we have the ability to recognize opportunities to leverage it? Can we recognize opportunities that for predict the future, classify information, discover hidden patterns in data, recognize objects visually and act on them intelligently automate processes, convert text into meaningful data and drive connectivity through semantic relationships? In light of our examination is forth question. Angela, I believe you’ve got another poll for us.

Angela:

I sure do. Okay. Poll number two says, “How confident do you feel implementing AI in your business?” I’m going to launch that now. Please take a minute or two to complete. It’s looking like we have a clear winner. Give it about 30 more seconds. Okay. Glen, it looks like the majority of the audience feels somewhat confident. Thank you for participating in our second role.

Glen:

Thank you, Angela. And that’s an interesting metric. We’ll have to look at that in more detail. As a parting observation, AI is here to stay and it’s transforming the business world. As business leaders, it’s imperative that we understand it’s capabilities, pitfalls, and unique characteristics. And to take advantage of the enormous benefit that AI presents to our organizations. The late great Vince Lombardi said the best.

Julia:

Great. Thank you, Glen. So, before we get to questions, if y’all are ready for more AI related content, Glen has a supporting podcast and blog series diving deep into more AI topics. We also encourage you guys to sign up for Glen’s next webinar on Wednesday, April 20th, it’s called deconstructing AI, A Deeper Dive Into Common AI Solutions. And Glen will cover the most common AI techniques for business along with how they’re constructed, where they have succeeded and failed, their pros and cons. And he will dispel a few myths along the way. You can access the supporting content and register for the webinar with the QR code on the next slide and it’s also linked in the downloadable slide deck. Now onto questions.

Angela:

Don’t forget to submit your questions through the feature, the panel feature on the right hand side, what’s on the right hand side of my screen. Glen, we do have questions from the audience already submitted. The first one says, can AI help stop vulnerabilities like ransomware?

Glen:

No. It’s an interesting question. I think it can certainly help. Ransomware… There are aspects of AI that can certainly help with security issues, but in and of itself, AI is probably not the specific solution that you would apply.

Angela:

Okay. Question number two says, can Glen point us to some leading sources for ethics and data as it relates to AI?

Glen:

What was the first thing? Something in data?

Angela:

Ethics and data.

Glen:

Ah, sure. Not in the moment, but I’d be happy to respond to that. If you’ll let us know who you are.

Angela:

Yes, I have that name. So I’ll make sure that we get that answer. Can AI help with inventory management?

Glen:

Hmm. If you think back to the discussion about deterministic and probabilistic solutions, inventory management tends to fall into the first category deterministic, which is not an AI type solution. I would anticipate that the options available really don’t play in the AI space. There may be some aspects in those products. And this sounds more like a product question that could use a AI for some learning capability, but I don’t think it’s the core solution for that type of opportunity.

Angela:

Okay. What team size and timeframe are we looking at for medium sized first projects?

Glen:

Oh, medium size could mean many things to many people. And it really depends somewhat on the type of opportunity and the type of AI technique you’re going to apply. I don’t have a specific answer for that. We are going to look in the future in one of our future episodes, about the skill sets that are needed in more detail and how that organization could be constructed. Whoever submitted that I’d be happy to talk to them offline.

Angela:

Okay. I will connect the two of you after. And next question, what are some of Glen’s favorite models that give accurate forecasting, mathematical algorithms?

Glen:

I’ve used two in the past. The pipeline example that we’ve used is, was a genetic algorithm. Mind you, this was over 15 years ago when we implement at this, for this pipeline company. And the technologies that were available then have been surpassed. What we ended up using, because there wasn’t a lot of software available at the time, was adapted from day trading. And there’s an interesting correlation between day trading and other forms of prediction, to think about that at some point. The specific technology was called a genetic algorithm, which attempts to mimic the way human evolution happens. Once again, I’ll be happy to have that conversation with a little more detail, it can get very wonky, very fast.

Angela:

Okay, next question. We have lots coming in, so this is great. Maturity models are extremely important in evaluating and implementing technologies or practices, who generated the maturity model that you presented?

Glen:

Our organization did. Access Sciences did.

Angela:

Okay. Any case studies available for planning or forecasting that can be shared?

Glen:

Planning or forecasting? It would probably be better if I pointed whoever asked that question to the literature that’s available on the internet. Part of it depends on the level of academic content they’re interested in looking at, since this is a business oriented webinar. And I think that’s what our audience is focused on as well. Some of that may be a little too technical. Not that people can’t absorb it, but it may be more technical than is used for. Once again, I’d be happy to chat with whoever submitted that and we can figure out what the right material would be.

Angela:

Great. Can you elaborate on the foundational components for governing the use of AI in an enterprise?

Glen:

Oh, my. Sure. And I’m going to focus on machine learning for a moment. I’m jumping ahead to our next webinar, but one of the primary components and in my mind, the most important component in true artificial intelligence is machine learning. And you’ve probably all heard that firm before. In prediction, the example where you looked at is a form of machine learning. The challenge there is that when you train a model, and we saw an example of that, the training is dependent on an input data set, which is typically historical information. And a set of results that correspond to that input data. So if you think of a weather forecast, we would look at several decades of historical forecast and the output, perhaps its temperature or humidity, pick your metric. And then use that information to train the model. And then we would present the model with new conditions.

I built a little system internally for doing just this, and it will predict what the heat index is. So the three of those things, the historical data used to train the model, the trained model itself, the results, historical results used to train the model, have to be managed or curated at as a unit in order that we can then repeat how that thing was built. How we can prove how it was built. And that’s going to be increasing… That is important and is going to be increasingly important. As the questions about ethics and repeatability drive themselves into corporate America.

Angela:

Okay. Glen, do you see a clear solution where AI helps a nonprofit with compliance and fundraising?

Glen:

Oh boy, one area that could be applied is the ability to discern or identify five patterns in day. So for example, if you have a donor pool and it’s too large for a person to process, let’s say you have several thousands or tens of thousands of contact names, and you’ve got some history about their giving pattern, then you could potentially use that to predict who the most likely candidates for regiving or more giving are. And by the way, I know who submitted that question.

Angela:

Okay. How are AI system different from normal requirements?

Glen:

System requirements? I think I know what the person’s asking and there’s really two questions based on their intent. So I’m going to answer both. System requirements and I’ll infer that they mean business requirements, are very similar between non-AI systems and AI systems. Between deterministic systems and probabilistic systems. Business requirement is a business. We need to predict gas demand, three days in advance. Is a business requirement. Regardless of how you solve it. How you then take that requirement and discern if AI could be applied to that problem, is really the art. And that takes some experience and skill. That’s what we were talking about when we refer to AI savvy business analysts. Their ability to understand not only the business world, but also the AI techniques, how to build and construct solutions especially from components, and how those might be applied to business problems. That’s the art that makes business requirement development different for AI opportunities.

Angela:

Last question. You mentioned needing AI systems to evolve as business conditions change. Do you have an example?

Glen:

Oh, yeah. Yeah. The pipeline problem once again. So This gets a little wonky, but I think it’s what we’re describing. When we first designed that system, we knew that the inputs were primarily weather based. So we could get a weather prediction, probably not generated by AI at that point by a meteorologist. Understand some other factors about time of year, day of week and so forth. And some traditional demand numbers, because it was historical data that was used to train the bottle and build a model that was very accurate. It was shockingly accurate. What happened in the mid to late two thousands is that electricity deregulation, occurred and companies that generate built water called peaking plants. And a peaking plant is a natural gas fed electricity generation plant.

And they can be turned on and off. I’m going to use the analogy like a light switch. It’s not quite that simple, but they can be turned on and off very quickly. And their demand was not predictable. And since they were connected to the pipeline, they would pull off some pretty significant amount of gas, which would mess up our forecast. That was a bit of a challenge. First, we had to identify what the problem was, take a little while. Then we had to figure out what the root cause was, which was actually commodity prices. And finally, we had to bake in commodity prices, both current and futures into our model, as data inputs. And getting the historical information for that was pretty challenging as well. So that’s a long story to say when they brought those plants online and started pulling gas off the pipeline, we had business conditions evolved and we had to evolve our solution with it.

Angela:

Great. Thank you so much Glen, for your presentation today. Everyone attending the webinar will receive a survey. Please take a minute or two to complete the survey, after this presentation concludes. You will also receive an email with a link to this recording and a link to the slide deck presentation and also a link to register for webinar number two on April 20th. Thank you all. Have a great day.

Glen:

Thank you.