7 Things Every Business Leader Should Know About Artificial Intelligence
![7 Things Every Business Leader Should Know About Artificial Intelligence 7 Things Every Business Leader Should Know About Artificial Intelligence](/pic/7_things_every_business_leader_should_know_about_artificial_intelligence/MDZ2OUxNWEhneUk_.jpeg)
Good afternoon, everyone. Thank you very much for tuning in today for our very first free webinar of 2021, we are confident it's going to be a very good year so thanks so much for joining us today for today's presentation, seven things every business leader should know about artificial intelligence presented today by Smith business insight and by queens executive education. My name is Meredith Dalton I'm a business journalist at Smith School of Business, which is in Queens University in Kingston, Ontario. Before I introduce today's speaker I want to remind you, as I always do that this conversation will be recorded, you'll be getting a link to the recording in your email once. Give it a day or two, but you'll be getting a recording to that link You're very welcome to share that with anyone you think might be interested in watching it.
If you want to ask a question and I encourage you to do so, please use the q amp a button at the bottom of the screen will be getting two questions in the latter half of the presentation so our guest speaker will present, and then we'll get to questions, but fire your questions in as he speaks, and then we'll try to get to those as many as we can, at the end. It's now my pleasure to introduce our guest speaker, Dr. Steven Thomas, Dr. Thomas is executive director of Smith analytics and AI ecosystem, as well as an assistant professor at Smith School of Business, a recognized expert on AI and analytics, Dr. Thomas teaches on natural language processing, machine learning database design, big data, and mathematical analysis. His research has, has been published in prestigious technology journals.
He holds a doctorate in computer science, and consults with companies in the areas of AI, big data, and text analytics, Steve Thomas thank you so much for being here today. I'll invite you to turn on your camera and join us on screen and then I will disappear and let you take over. And you Meredith. Okay, Good afternoon, everyone.
Welcome to seven things every leader should know about AI. Number one, and first and foremost, Ai, is finally here. Everyone's been hearing about it for quite some time, but now is the. Now it is safe to say that most organizations are using AI in one way or another. Some of the more popular AI success stories that you might have seen are for example self driving cars, Tesla Google, Apple they're all in a race, and making great progress on self driving cars expert game players so AI algorithms are now able to beat the very best human players, and not only games like checkers and chess, but infinitely more complex games like, go and Starcraft. Most of us are using AI without even realizing it on a day to day basis when we use our personal assistance on our Apple iPhone or Google Home. There's an incredible amount of technology underneath the hood. That can translate what you're saying go fetch the right answer and get it back to you, all within a few milliseconds. And then of course, my favorite AI success story is deep fakes, which is a, which allows allows you to put one person's face on another
person's body and make it look seamless and therefore you can have Nicolas Cage star in whatever movie you want. So we are really living in an age of enlightenment. But what is ai, ai is one of those terms that gets thrown around. It can mean anything it could mean nothing It can mean everything at Smith and in our masters programs we like to use the following definition of AI. It's a computer's ability to make decisions, or to predict something and kind of higher level. It's whenever a computer is trying to do things that humans do like learning or seeing or speaking or moving or the four big examples. And there's a lot of as a big field, and that's why it gets talked about so much and you can focus on several different aspects of AI. For example, a lot of, lot of people that are researchers a lot of practitioners, they're interested in the engineering piece behind AI, like how to develop new algorithms that do the learning and seeing, or how to collect more data that fuel the algorithms or building new hardware, like GPUs and GPUs to actually make the algorithms, much, much faster or building new software platforms to make, make it easier for data scientists to build an API into existing products or building new apps that take advantage of this. So there's a lot of activity on the engineering the computer science the math side.
But there's also a lot of work on the application side. So, what do we do with this cool technology, how do we change our business practices. What can we do in terms of new tech products. What can we do in healthcare and medicine, and that kind of stuff. How do we apply it. And then finally there's an, there's a lot of work going on on the ethics of AI. And this is a fast moving and very interesting topic, but people are asking the hard questions like what is fair is a prediction that the AI make Is it fair to all subgroups of society or, or is it bias in some way.
And there's the issue of data privacy people's personal data is usually used to fuel AI models, but what about people's rights to individual privacy. This is something we haven't got quite right. And then the hard questions of accountability. So if a self driving car makes a mistake and has an accident, who's at fault is it the driver. Is it the AI manufacturer, is it the software developer who wrote the AI. Is it the insurance company, you know, these are all unsolved problems at the moment. So AI, when people say I they might be focusing on maybe one of these pieces or the whole thing it's hard to know and that's why it's such a kind of slippery term.
But in general, AI is it the reason it's gotten very popular in the last few years, is because, first of all, most companies are storing more and more data about their transactions and their customers. Second, there's been some advances and CPU power CPUs are cheaper, faster than they ever have been before. And then second of all, a third of all with with the advent of cloud computing. It has gotten much cheaper for organizations to to access huge arrays and networks of compute power. So when you add those things together. Companies are getting faster results on bigger data sets, which has been leading to models that can create high value predictions, they can, they can, For example, very precisely predict which customers are about to leave, or which offer to give a customer or which product to recommend the customer, and these are very accurate, and therefore the customer will use them, and therefore companies are making money, as I alluded to earlier.
Ai really has four separate main abilities, and it's helpful to break them down so we can start understanding the pieces. The main one, and this is probably the biggest in business today is machine learning, this is these are a class of algorithms that can learn rules and patterns automatically from historical data set. Then there's computer vision. These are a set of algorithms that can analyze images and videos. And, you know, try to see the world and detect objects and try to figure out what's going on.
There's natural language processing, which are a class of algorithms that can analyze text and speech, and do, and break, break texts down into sentences and words and try to extract meaning and try to figure out what the human wants to do. And then finally there's robotics. And this is all about building hardware robots that can move around the world. Now work in the factory work in the farm, that kind of thing. So will be kind of diving into a few of these today.
One point of confusion. A lot of people have is, what's the difference between analytics and AI, these two words are sometimes used interchangeably. or some people think that the same thing. In my mind, analytics is is a process that a human does. It's a process to make decision, makes decisions with data. And when you're doing analytics, you have a toolbox of tools that you can use. Sometimes you just need to create some simple visualizations and create some reports and dashboards. Sometimes you need to just run some simple statistics. Sometimes you can you make some reports.
So sometimes when you do analytics, you don't need a at all. But you can also lean on some of the AI capabilities in your analytics process, and in particular machine learning and natural language processing can be very helpful. as you're trying to make a decision using data. So, in my mind, ai provides a suite of tools that go into our analytics toolbox, so we don't have to use AI when we're doing analytics, but we can now most companies, they focus on three main objectives when they're using AI.
The first one is to reduce costs. So some examples of reducing costs, is you can automate existing business processes. So JD calm and China automatically fills orders using robots, or in provision in China. Also, they can automatically detect cancer from x ray scans Domino's has a tool to automatically check pizzas, as they come out of the oven to make sure that all the ingredients that the customer ordered are in fact on the pizza. You can also reduce costs by just having better operations and logistics and a great example of this is ups, Use an AI algorithm to optimize the driving roots of their drivers, and they saved. I think is about 2 million miles a day in, in time and gas.
And you can also reduce costs by decreasing customer churn. So, The United Health Center in the US. They monitored customer complaints to be proactive about customer resolution, and therefore, decreased churn. You can also increase revenue. And the best way to do this using AI is to have a really good recommender system to recommend products to your customers, and everyone's everyone's seen Amazon's recommendations you know people who bought this also bought that is a great way to expose your customers to new products that they didn't explicitly search for and in fact Amazon makes about 40% of their revenue from their recommendation system.
You can also increase revenue by just having better pricing. So companies like Macy's will use advanced AI algorithms to optimize the price of 73, different unique items 73 million unique items per day to to maximize revenue. You can also, you know, optimize your, your shelf layout in your physical store. If Backman physical stores were a thing. But, but again by analyzing shopping patterns of your customers, you can find a better ordering to increase revenue. And then finally, a lot of companies been using AI to develop new offerings.
For example, Apple provides new features to their existing products like their iPhone with an AI agent, like Siri, Tesla, you can buy a Tesla car, and then you can spend 10,000 more to buy Tesla's autopilot, which is all based on AI Disney uses the magic band which is full of AI to track customers and help them optimize, not only optimize their parks, but also recommend things for the customers. And then finally some whole some whole products were are only enabled by AI, for example, Netflix. A lot of people think Netflix product is the, the TV shows, and the movies, but really Netflix product is a recommender system, which is all enabled on AI. In Canada, there's a lot of things going on in AI, and this kind of drives the point home that Ai, is finally here. And this is a this is a nice chart.
That's updated about once a year I'm still looking for the 2020 version, but it's showing that all the companies and all the activities going on in the major cities in Canada, Toronto, Montreal, Waterloo. So that's this outer ring is entering his academic labs that have started in the space for example the vector Institute in Toronto is a big one for Mila and Montreal, and other government. Government and Public nonprofit, basically, there's a lot. So for a smaller country, Canada is punching way above its weight in terms of AI progress and technology.
And in fact, I think one of the reasons that's true, is because Canada was the first country to have an official national AI strategy. So back in 2017, Justin Trudeau and his government, and now they launched the pan Canadian AI strategy which puts a few hundred $9 on the table for academic research startup funding, things like this. Since then, almost every other major country has also launched a national AI strategy, and they all have their unique flavors and niche markets, but it is interesting to note that Canada was the first one to do so. Here's a nice McKinsey study from about a year and a half ago, that shows where AI is being used across many different industries. And this was done from a survey of about 3000 executives in the US. But basically, industries that are higher up on the list, use AI more. So telcos use it the most high tech, and then financial services, these are the ones.
These are the sectors that use AI the most. And then the business units within those sectors, the ones that are more to the left are the ones that use AI more. So as use most right now in service, and then product development, and then marketing. So overtime we're going to see. Basically, all of these boxes slowly but surely creep over to the right as AI gets adopted more and more and more. Lesson number two. ai is narrow focused and hungry. So this is kind of the.
This is a reality check of AI. The AI that's available to businesses today is called narrow AI, or sometimes called week AI. And these are AI algorithms that are very specific very focused, and they need human guidance, so they can only do one thing they can do that one thing very well, but that's it. For example, a chess algorithm, a chess chess algorithm can beat the world champion Magnus Carlsen easily every time. But that same algorithm as smart as it is a chess, it can do anything else, it can't. It can't scramble eggs, it can't predict customer churn it can't do anything else, it's only trained to do that one thing, and even Amazon's recommender system really good at recommending a nice book, but that's literally all it can do. It can't. It can't type choose it can't read a book, nothing like this.
Most people like my mom when they hear the word AI, they think bigger they think generally I, and generally I is the notion of human level intelligence, or even beyond human level intelligence. These are systems that can solve broad tasks, you can give it very vague instructions and it will know what to do, because it has general reasoning capabilities, it can teach itself. And the reason a lot of people are like my mom, and they have this notion is because of Hollywood, like the Terminator, or, or Did anyone see the movie her with Joaquin Phoenix, you know, or even RTD to from Star Wars, these are general a human level intelligence that can do anything.
Now unfortunately these systems do not exist. And the best estimate from some of the top AI researchers, is that they might be available in 50 years from now. I'm not gonna hold my breath, because in the 1940s, the best estimate for general AI is that it was 10 years away. Here we are 70 years later and we're thinking it's 50 years away. So, that's the bad news. The good news is narrow AI is good enough. It's good enough to get a lot of business value. So we don't we don't have to wait for general AI. When I say AI is hungry. What I mean is AI needs data, it needs a lot of label, what we call labeled training data.
So let's say for example you wanted to build a credit risk model to predict whether a loan application is risky, or safe. What you need to give that give your AI algorithm is a lot of past loans. So for each loan you might know some attributes of the person who applied, their age, they have a job and so on. And then you need to know what happened. What happened to that loan with this person pay back this person did not this patient paid back this person did not. This is what he needs, because the algorithms are going to use this data to basically figure out who who ends up paying back, and what kinds of people don't.
Without this data. Ai can't do anything. And this is a common misconception. People think that AI just works, somehow, know it needs to. It needs to be guided by this, this human labeled training data. Another example is for. For Computer Vision type AI. So, computer vision algorithms are really good at detecting objects in an image, but that's only because human beings have painstakingly manually labeled a lot of images beforehand. For example, this is a tool that a human might use to label objects in an image so that a human has drawn a box here and said okay this is a tree.
This is a person. This is a chair, really boring work. But as a human. If a human does this enough times and I'm talking 10s of thousands or even millions of times, that's when the algorithm can start to learn for itself. What a tree looks like, what a person looks like.
So in general, what do we know what do we know about AI. The main point I want to make here is as exciting as AI is. It's still early days, we're still early in this journey. We know that AI is good at providing predictions, like it'll predict who might churn, which customers might turn, but it's not good at providing decisions, that's still a human level activity. Okay now that I know that Bob is going to churn What should I do, should I give him a discount. Should I just let him go too expensive, what do we do, that's still human
He is good at automating tasks, but not yet at automating complete jobs. So I'm of the, of the opinion that there's not a lot of jobs that are truly at risk of being completely automated by AI. What it will do instead is a I will automate the simple tasks that are that are very redundant and very data driven, and frankly very boring for human, and that that's actually a good thing, it will allow the human to focus on more human level things like being creative and talking to other humans, and you're making that social connection. Instead of, you know, doing monotonous tasks over and over again. And as I just said AI is, it's very narrow very focused. So as long as it has clear parameters, and its job is very well defined, it can it can be good.
So what AI is not good at. It's not good at solving novel problems. So if you don't have a lot of that labeled historical training data, to give the algorithm. It won't know what to do. Like if you ask the AI algorithm. Hey, should we merge with this other company. And, but you've never merged with any other companies and you don't have any data to give it, it won't know what to do it well, its guests will be no better than yours.
Similarly, it can't solve a problem if you don't have any data. If there's no. If you haven't saved a bunch of data from the past or you can't buy data or download it somehow. the data is not going to help AI is not good at having common sense like human humans are somehow. It's very clear, you know, common sense. There are a lot of decisions that humans would never make because it just doesn't make any sense where ai, ai has the ability to make spectacularly bad decisions, sometimes
AI is not that good at creating something completely new. Like there's been some cool work on AI creating music, classical music, but that's only after AI was, was able to listen to, you know, millions of hours of existing classical music, you can learn to kind of mimic the patterns. But if you couldn't just say create classical music without telling it what classical music was it can't really do that. So I needs a lot of data predictions of one other consideration is AI predictions have very low interpret. Some of the best AI algorithms. Humans don't truly understand how or why they work. We know that they work, we know that they're pretty accurate, but we don't know why. So for an in an industry where we want to know why.
ai is not a good solution. Also, AI is not always the answer. Sometimes, simple as good. Sometimes you just need to print out a piece of paper and have a simple lookup table. It's not always the case that you need to invest millions of dollars to build a fancy AI system.
Yeah. And so, just because AI can solve a problem doesn't mean that should. This is a trap that some early tech companies have fallen into where they tried to use AI for everything. and they realized, well this was a little bit overkill. Okay, so just, just a warning. Okay listen number three. AI is it provides prediction on steroids, and by this I, I'm talking about that machine learning component of of AI.
This is where this is the part of AI, this is the class of algorithms that can make really really good predictions on the behalf of a human. And the way it works is the AI algorithm will create this model. This model it's like a. It's like a file on your computer that you can ask questions to. And you can give it, let's say this model was trained to predict which loans, loan applications are good, which loan applications or bad. We can do it once we have this model is we can give it a new loan application. And we can say hey model. What do you think, is it going to be a risky loan or safe low. Excuse me.
And the model is going to say, it's going to give you a prediction was gonna say I predict that this is going to be a good loan. Based on what I've seen in the past. And that prediction is going to be very very accurate.
Now where did that model come from that model came from what we call a machine learning algorithm. So that machine learning algorithm. It looks at a bunch of historical loans that we as humans have given the algorithm. And the algorithm will crunch and do a bunch of numbers bunch of math, which is statistics and create this model. And this model is what will use in the future.
But if you give, if you have enough historical training data. These algorithms can create really really accurate models, much more accurate than any human could do on their own. So that's what I mean by prediction on steroids. Now of course there's a lot of pieces in involved and still a lot of research going on lot of work going on on all of these boxes. I for example on the data side, you know, what's the best way to get it. What's the best way to clean it and prepare it, and how can we detect if there's any errors in it or biases, things like that. Very interesting topics on the algorithm side, and this is a kind of a computer science problem. There's a lot of algorithms that have been proposed, and that are used. You might have heard it sound like a decision tree random forest neural network. These are all algorithms that can learn and model. They all have their pros and cons.
So we need to know as business leaders and data scientist, which one to use when. And then once the model is built, we have to ask questions like how do we tell exactly how good the model is, you know, is it 90% accurate. 9598, how do we test it. Also, how can we explain the predictions that it made some of the algorithms are very easy to understand why it made a certain prediction. Others are impossible to understand.
And then finally on the production side. Once you have predictions. What do you do with him. This is kind of an often overlooked question. But, you know, let's say you do get really high, high quality predictions about your loan, you know, about loan applications. What are you gonna do. Are you just going to use the predictions as is or are you going to have a human, take a look at them. or you're going to run them through some other system, you know, there's a lot of possibilities, all with pros and cons. So that's another kind of active area of research.
Machine learning is, I would say that the most used type of AI in business today. Here's an eye chart of all the types of things that machine learning has been used very successfully for things like what I've been talking about predicting credit risk predicting customer churn, but they can also be used to predict device maintenance or it's using health more and more, to predict, you know what kind of ailments or diseases, somebody might have. It can predict whether a transaction is fraudulent. It can predict how much someone is going to spend in their lifetime. It can predict even which employees might be at risk of leaving the company. It can provide quality assurance automatically, and so on and so on. We could spend a year, talking about all the machine learning applications, very exciting stuff.
Lesson number four. AI can see. And this is the computer vision piece of AI. So here's a nice video let's see if I can figure out how to play this. Okay, here we go. So this is what a computer, a modern computer vision system that's mounted on a self driving car does. So this in fact there's a lot of little algorithms that are each looking for very specific things in the road somewhere looking for cars. Some are looking for stop signs, some are looking for lines on the road. Some are looking for pedestrians and in concert, they're able to basically have a really good understanding of the outside world.
This is an example by the way of what Tesla autopilot can do. And so the, this is a great example of AI it's very mature it's getting better and better and better companies are pouring tons of money into it, to try to make self driving cars reality. But self driving cars aren't the only way that computer vision can be used in business. Another example is Amazon has launched a series of stores that don't have any checkout lines, because they have cameras all over the store and they will watch you as you walk around and place things in your, in your cart. It will notice that you placed in your card and what the product was, and basically it'll build up an inventory as you go. And then you just walk right out. It does facial recognition knows who you are as your credit card on file. There you go.
It can be using and QA processes. So you can have cameras and on your factory assembly line. And as products go through certain stages, the algorithms can automatically detects if your product has the right size, shape, color, you know any other kind of defects so you might need to be watching out for it can do facial recognition and this is a hot topic, whether facial recognition is is good or bad, but there's a lot of cases where facial recognition might might be good. And so these are very mature technologies. It's used a lot in medical imaging more and more, so can help doctors automatically analyze x rays CT scans, you know, any kind of medical imaging and automatically pinpoint. You know cancers bone breaks, you know, disease, indicators, things like that. So we can help automate this and takes the load off of of humans. It's been used in security cameras to automate watching security footage, and it can alert when anything suspicious happens. For example, agriculture is using CV more and more. So for example, there's drones and other robots that can selectively seed and water plants, only if they look like they needed at this time. So this will help reduce wasted water and seeds on plants that don't really need it right now.
And even banking we've all used, you know, simple version of computer vision when whenever you take a picture of a check with your cell phone. The bank is using some computer vision on the back end to, you know, verify the amount and your signature and things like that. So lots of great uses uses uses. Number five, AI can talk, and this is the natural language processing piece so natural language processing is all about trying to, you know, break down sentences into smaller parts and understand how the words relate to each other.
And ultimately, Try to understand what the human is saying. Excuse me. So for example the sentence, Steve was born in New Mexico, you know, a natural language processing algorithm will first of all break these into individual words, and then it will know things like okay born, was it is a verb, and New Mexico is a noun phrase, and it'll kind of relate the words together, and kind of build up an understanding of of the sentence from there. And with these kind of techniques, there's a lot of applications and op is another really exciting area.
For example, what you can do what's called information extraction. This is where you can the algorithm can scan a document and pull out names places, events, or even relationships between how things are related to each other, like for example, Steve was born in Mexico that's relationship between Steve and New Mexico. So it can help you provide structure to your documents. It can help you make predictions on documents. So for example, if a loan application included a written description of what the loan would be used for NLP can help you know predict whether this is a good use for the loan or bad news for the loan sentiment analysis many probably heard of this, but this is these are algorithms to automatically read a document and determine if it's up you know positive negative if the right writer is happy, sad or disgusted.
And this is great uses uses for monitoring social media are monitoring your brand's reputation, or monitoring the news machine translation, things like Google translate to these are great tools to translate one human language to another human language chat bots are becoming bigger and bigger. So these are tools that can automatically respond to your customers and answer kind of basic questions, take orders, things like that. And then finally there's some interesting new tools to generate language from scratch, so they can write songs from scratch, like generate words that actually makes sense and sound like a real self, or they can generate product descriptions, given some description, or some data about their products, or they can generate new stories based on some bullet points, or they can generate bullet points of have a meeting to kind of summarize the meeting and then send it out to all the participants. So a lot of really cool stuff and natural language processing. Finally, AI can move. This is the fourth main capability of AI. And by this I'm talking about robotics.
So, again, robotics is all about building hardware robots to physically move around the world. Here's a nice animation from Boston Dynamics, which was just bought by Hyundai. But they're one of the leaders of building these humanoid robots that can move around the world and are very agile very nimble have pretty good balance. And, basically, they're, they're made to do things that humans don't want to do. It's too dangerous or just too difficult.
But here's some examples of how robot robotics can be used in business. One is and this is the ultimate, and a lot of companies are pushing for this is drone delivery. So imagine you know ordering a pizza, and it skips all the traffic and arrives on your doorstep 10 minutes later, nice and hot or Amazon you click Amazon. Add to cart and then 10 minutes later it's at your doorstep robotics is used more and more in business, I'm sorry in agriculture.
So, you know, picking the crop washing the crop slicing the crop, sorting the crop, you know all this is being done more and more by advanced robotics self driving cars has a lot of physical robotics in there. So it's not only the computer vision part, but it's also the steering and the braking and a lot of the physical parts as well. Probably the biggest use of robotics is in manufacturing. So, you know, think of a auto assembly line is almost entirely these big fancy robots at this point. warehouse is so big companies like JD calm, like I mentioned, an Amazon, they're relying more and more on robots in their warehouse fulfillment centers to actually put you know fulfill an order very quickly, completely automated farming. So, you know, nowadays, there's robots to kind of scrub the cow massage the cow clean up after the cow milk the cow, you know, monitor the cow. So it's being done almost completely automated these days.
They're self cleaning robots are cleaning robots. And it's not just the Roomba that your grandma has, but it's, you know, industrial cleaning robots that can wipe down the factory floor after a hard day's work. Even robots to provide companionship, and this is a big thing in Japan right now, in some eldercare homes where, you know, a dog is the perfect companion for for lonely person, but for an elderly person. Real Life dog is can be a pain in the butt, because they need to be walked and fed and cleaned up after were robot dog perfect solution. Finally last lesson I want to leave you with today.
And this is a common quote to make amongst AI practitioners, is this stuff is still hard. So AI is still just hard, like I said it's early days. There's a great promise, and there's already a lot of great solutions and applications, but there's still a long way to go. For example, it, we are learning more and more that AI systems can have bias unintentional bias in their inner predictions. For example, they might. There's a lot of famous examples of this but Amazon got caught red handed. Using a system to screen resumes. And it turned out that this system, always rejected female candidates.
And so that was a big problem. Turns out the reason they did this is not because AI is inherently biased towards females. But really, because the training data that was given to the algorithm was biased. Why was the training data biased. Well, turns out that the managers at Amazon, that the data was based on were biased. So basically, that the AI will perpetuate any bias that already exists in the training data.
So that's an active area of research to watch out for. The other another challenge is interoperability like I said earlier, a lot of these models, we, we don't really understand how they're working. We know that they do work really well, and we can make money with them. That's the good thing but the bad thing is we don't know what they're doing exactly. And that's a little bit unsettling and sometimes it's illegal to use to make a decision that you can't explain outliers and data in your training data can sometimes throw off a model pretty, pretty badly.
So that's an active area of research. Like I said earlier, these algorithms are very data hungry. So you just that's just the the cost of entry, you just have to have a lot of data. If you don't, Then you just can't use AI.
And also, like I said earlier is, these, these algorithms have no common sense. They'll sometimes make spectacularly bad predictions, so bad that any human no human would ever do that, but an AI doesn't really have a baseline of, you know, common sense. Another set of challenges is on the people side and these are actually harder than the technology side.
For one, managers don't understand AI, and when humans don't understand things. They're usually afraid of them. And so they don't want to use it. I don't want to use AI, I don't understand it, it might break everything. So let's just skip it. So that's a big challenge for companies to truly adopt a new challenges, a lot of times it's hard to integrate these models with your existing legacy software. You know, if you're using a piece of software that's been around for 20 years, and nobody remembers how to build it or it's very expensive to make any changes might be super hard to, you know, upgrade it with AI models. Change Management is always an issue, you're trying to convince an organization to change the way that their businesses running to use AI, It's an uphill battle. And a lot of people.
You know, a lot of people along the chain will again they won't understand AI or they'll think that it's too risky so they'll say no. And along those lines. A lot of managers will say, okay, it's too risky, there might be problems I don't want to deal with it. So let's not do it on the kind of the bottom inside frontline workers might push back as well thinking, oh my job. I'm going to lose my job, and so they'll find ways to argue against adopting AI.
Like I said, I personally don't think many jobs are actually on the, on the chopping board but some people believe that they are. And then finally, ethics, there's a whole can of worms with ethics like what what should be done. Who's in charge what happens when it goes wrong. What about the privacy piece. There's a lot of unsolved problems here that sometimes trip people up. Okay, so, I know this is a whirlwind of information. If you want to dive deeper, here's four really good sources of inspiration some for good books that you that I'd like to point you to. And I'll just quickly summarize I drew a line between these four books because they kind of give conflicting advice to managers, the ones on the left, they say, if you want to start with aI don't do this big moonshot project start with something small a small proof of concept, and then kind of build up from there. That's the best better approach, where the other side of the line they say, No, don't waste time with proof of concepts, take on a big project with big impact.
You know, shoot for the fences. They also say, and this is good advice, I would say, no matter what is you want you want to look back, backwards from strategy, not forward from AI techniques. What that means is, don't go looking around for problems you can solve with AI.
Instead, go to your strategy and say okay what problems do we have. And some of those might be solvable with AI, but don't try to force, force it in this book also says prioritize revenue growth over cost savings. So, take it for what it's worth.
And for more details you can take a look at these books. Okay. So with that I'll conclude our seven main things that every business leader should know about AI summarized here. So it's here. That's the good news bad news is it's, it's pretty narrow still it's very focused on one thing and data hungry, the four main capabilities that it can do is it can make predictions, you can see, it can talk, it can move. And as excited as you should be about AI.
I do want to be realistic and that it's still very difficult to to get it right. Companies are getting it right. But it's harder than it looks. Alright. So with that, I will conclude, and open it up for any questions. Wonderful Thanks Steve oh my goodness, we've had so many questions because there's so much. There's so much we don't know.
I feel like you put all these possibilities out there and now we're all like, Whoa, we have questions. Here's a question from the audience. Can you elaborate on the role of AI and capital in the capital market sector in Canada, so something you can touch on. Yeah, I mean there's a lot of work it's still early in the capital market, and investing. But there's a lot of auto traders Autobots.
There's a lot of risk assessment models. There's a lot of models to automatically monitor, like news and try to figure out what the trends are in today's, you know, what are the trends of today and what does that mean for for stock prices. There, in fact, we, we just launched a program at Smith called in fits, Master of financial technology, and they touched on this a lot is how to use analytics and AI and all there is a finance but including capital markets, so it's a deep product. It's a deep topic, I should say, with lots going on so short answer is yes, there, there is activity in there, in that space. Thank you. Um, this is a very specific question in response to something you said earlier about the notion of narrow AI our self driving cars using a narrow AI.
Good question. Um, yeah so a self driving car would have, I would say, hundreds of individual algorithms that are each very narrow, like one algorithm, literally all it will do is look at a picture and say, is this picture of a stop sign, yes or no. And so that algorithm and that model is very narrow. One cool thing though about self driving car is the biggest challenge of self driving cars get all of those models to work seamlessly together and have like a super like a brain almost that takes all these predictions. All these probabilities from the individual models and combine them to make a decision, and the decisions are. Should I continue going forward. Should I hit the brakes. Should I turn left. Should I turn right, you know these kind of things. So, I guess, definitely, each individual model is narrow AI.
Maybe you're making the argument that when you combine them, it's, it's, leaving narrow AI and heading towards general AI, I guess I guess that would be true, I guess that would be true. I would, I would still say though that self driving cars, even if they perfectly integrate all these individual models. can scramble eggs, it can still can tell a bedtime story, but that's going to be amazing when it happens, the egg scrambling self driving for your breakfast while you're going to work.
So related to cars is the high use of AI by automotive companies in the manufacturing sector due to robotics, are there other reasons to explain the 49 value. What was that on the McKinsey study. Yeah, it's definitely on the robotics. So, on the manufacturing process. In fact, I was fortunate enough to go to a auto manufacturing plant in Germany. A few years ago, and I was just blown away. It was absolutely amazing what their factory floor looks like it's like, first of all super clean, super efficient, like pumping cars out, but almost completely automated. And, you know, the road, there were humans there definitely but a lot of the work was done by you know these really big kind of scary looking robot robot arms would do the welding and move things around quickly and. And they have inside of them, like computer vision and machine learning and robotics. So yeah, that's what's driving the high value there for automotive.
This one's complicated, because so many of these questions are complicated. Can you comment on the ethics behind people, I end users, knowing whether something is AI versus whether something's been made a decision has been made by human. Yeah, um, My only comment there is that it's still an open question, and there's no resolution, and nobody knows the right answer. I went to a conference on AI ethics in Toronto, about a year ago. And I remember feeling.
When I left the conference I felt more confused than when I went and I felt like less problems have been solved in a room when we started. But yeah, that's a good question. I mean, like, can, can organizations, say when products are be a win, win recommendations are automated Yes. Some companies do. But some people don't care. It's like, I just want, like when for example when you're using Google Maps, you just want to know the right right way to go. You don't care how that happened, or an Amazon, if it shows you a book that you're interested in. Great. Do you care whether it was collaborative filtering, or deep learning.
Not really. But on the other hand, when it's kind of more personal life changing decision, like being approved for a loan. I think that's where things get a little fuzzier, because if you're if you want a mortgage, you want to move into a house with your family and get more space so your kids are out of your hair during coronavirus. And, and the bank says, Oh, I'm sorry the machine learning models said no you're not approved. And you're like, Well why you know I got a nice job. I've been paying my bills on time but I don't know deep learning model set so and that's you know that's that's our master now.
So in that those cases, I think it's responsible to tell the customer, how the decision was arrived at. That's a good question. So long story short, there, there are some times when nobody really cares if you know where the decision came from. And other times where it's crucially important. Right. Yikes, so complicated this stuff. Again, this is a big question and, maybe, I don't know if you can answer it but is there a methodical approach available to scaling AI solutions across an enterprise, thoughtful.
Yes. It's hard to say you know quickly in a webinar q amp a, but there's a lot of great guidelines and you know like playbooks. And a lot of this will be based on using a cloud provider like Azure AWS, those, you know this is their bread and butter, this is what they try to do is they try to help organizations, you know, ingesting data into the cloud and then scale the operations, you know, piece by piece and eventually across the whole organization. So, yes, there are is a short answer, and there are some great resources out there to help you do so, and we have programs at Smith to help you do so as well. Thank you, um, gosh, we have so many amazing questions coming in, we're going to need another hour, sorry audience.
I'm gonna try to sift through some of them quickly. Um, this person for the session and says, For someone interested in business applications of AI, what depth of knowledge is required, also any learning are ways to gain or build knowledge about the same that you suggest. So, yeah, there's kind of different levels you can dive in and, you know, we've been we've been thinking about this at Smith for a few years now. Like the first thing you can do is join a one hour webinar with with Steve Thomas. So good job for doing that
kind of the next level of getting your feet wet is doing an executive course like a one or two day executive education course which we happen to offer. It's called a jumpstart for managers and that dives a little bit deeper than I did. We talked about the different algorithms how they work. The next level is to get like a master's degree in artificial intelligence management, which we also happen to offer. It's called the AI degree. So I mean, the more you learn, the better the more effective you'll be like for example if you spend a year getting a master's degree in this, you'll be an extremely effective leader of AI, you can lead a data science team, you can speak their language, you know the the gotchas you know which algorithm to use you've done it yourself a few times. And so you'll just be super effective and you'll know what to do. Whereas if you if you just do the executive course you'll be aware of like the landscape and you'll have learned some of the terms. But since you haven't done it yourself, so many times you'll still be kind of, you know, it'll be harder to get going.
So, I know that's a long answer but the more you invest the better you'll be just like anything in life. But there are options. There are also a lot of non Smith options are some Coursera courses, and other online training courses for free which will also kind of whet your appetite and give you an overview of what's available. Thank you and now from the individual to the company someone asks how do we evaluate if our if our company is a good fit for a good candidate for a great question, um, the way to do that. And there are some. There are some assessment tools that you can google around for basically where you can answer some questions. They'll say like, they'll ask some very specific questions like, Where is your data and your customers, and it will give you three choices and you'll choose one. And like, who makes the decision, level. And once you answer all that it'll put you on a scale of one to five, where one, and they're very nice with their wording it's like new to AI, which really means you you're. You've done nothing and you have, there's a lot of room to grow, all the way to like AI advanced. So, that can be a first way to kind of see where you sit on the scale compared to your, your peers and your same industry.
But then, really what you need to do is you need to set aside half a day or a day with some of your key team members, write down the problems you have, what are your major obstacles. And then, and then think through. Okay, for each one is this is this something we can automate using AI, or is this something we can solve using AI. Do we have training data, our algorithm do our rooms exist.
This is hard to do if you've never don't have any education in AI, but you can you know have a consultant come in and help you with that for a day or something like that. And that will give you a sense, sometimes you'll have a bunch of problems but none of them can be solved with AI. At least you know that, or maybe there are some low hanging fruit that you can start to move forward on. And then what does moving forward, look like.
So moving forward would be collecting some data prototyping a model and a lot of these models are actually, you don't need to do any programming or anything like that. The cloud tools that I mentioned earlier, as your AWS. They have kind of pre built, ready to go point and click models that you can you can deploy, which is really nice. So it's really just having a semi technical resource on your team who can connect all the dots, find the data, upload it, build the model, get the predictions out and then put the predictions into whatever format you want. And for a simple problem this could take as little as a day. If you have the training data. That's it.
That's how easy it is these days from complicated problems it could take, you know, six months or a year. It all depends on the nature of the problem and how much data you have. Okay and then someone asks about the issue of them convincing people that you should be using as if someone doesn't want to understand if you said there either challenges with managers not understanding how do you communicate the value of AI and hopefully bridge the gap so that people can get on board. Yeah, that's the hardest challenge of all of them all, you know, if all the calculus math stats that you can talk about with AI. The hardest part is the human convincing.
It's like convincing your wife that it's a good idea to, you know, buy a new gaming chair, you know, it's a. You need to in. And the answer I'm about to give is actually not unique to AI it's just general change management, you need to put yourself in their shoes and say what is, let's say Meredith is the person trying to convince convince, what, what is Meredith view of the world what is her biggest problem right now if her biggest problem is is it takes her too long to respond to emails. Well you then frame it like that you say okay, Meredith, I have a solution to your email problem. Would you like to automatically categorize, you know, 75% of emails away as you don't need to respond. Yeah, of course. Well, I can do that.
And it turns out, you're going to use AI to do that but Meredith doesn't really care Meredith has a problem. So frame it in terms of their problem, not in terms of your AI solution. Because at the end of the day, people don't care how you solve problems, they just want their problem solved. My wife is marketingprofs and she what she tells her students is. Nobody wants to drill. No one wants to go to the store and buy a drill. People want to hold on their wall. That's what you actually need. And so you need to talk about all the holes in the wall that you'll drill that you'll create for them. By the way, the way I'm getting these holes as with the drill.
So it's kind of like that, with same with aI don't start talking about neural networks and all this stuff, talking about the problems and and what it means once they'll be solved. Okay, we have two more minutes to go on this webinar and I want to ask you a bit of a heavy one, we'll see what we can get at around the question of bias which I know is complicated but how do we handle biases in the training data like really how do we do that so visible minorities, for example, have low credit data in general that AI for example might learn not to Linda visible minorities I mean these are real problems. What are we supposed to do about those things, as a society, so I have a PhD student who is focusing on exactly that problem. So, financial credit to minorities. It is a big problem number one. So the good news is, a lot of organizations, and even Ottawa, is realizing that this is a problem. So that's step one.
Now step two on the technical side, we've identified kind of three classes of ways to solve the problem. One is you can actually change the training data itself, and. That's one way to do it. Another way is to change the machine learning algorithm itself. And basically, reward the machine learning algorithm for not being biased or for basically say, you should be giving the same amount of loans, roughly to minorities
and non minorities. And if you don't penalize yourself, and basically by doing that, it'll learn a fair model. And then the third way is you don't change the training data you don't change the model but you look at the predictions, and then you can change the prediction based you can flip some of them say, look, the proportion is out of sync, you know it's out of line here something's wrong. If we were to change the threshold a little bit. Things will get better. I know that's kind of complicated probably don't understand. I actually did a talk on this. I think it's on Smith's YouTube. If you want to take a look. A few months ago. Okay. I mean, at least it gives us a sense that there is hope, like there's things that we that can be done. We're not about Kimbo is is forever. You know, there's about 12 different groups of researchers around the world who are throwing you know going in trying to solve this problem, and there's good good progress being made. So there is hope. That is really fantastic Steve, thank you so much you've been wonderful we have so many questions and to our audience and very very sorry that we don't have
to get all of them but we'll, we'll pass them on to Dr. Thomas, he can review them all and just maybe fine we'll find another venue hopefully to tackle them all. And before I want to wrap up before we end today I want to wrap up by letting the audience know Steve you did put in a plug for did put in a plug for executive education. We do have lots of programs online. They're all online programs, and we do have several coming up on analytics and AI, they'll be launching later this year and so there may not be info quite yet about them but I'll invite you to visit Smith queens. COMM slash exec Ed on a regular basis bookmark the site, go back periodically and just see if there's something that aligns with what it is that you want to learn and Smith exec Ed can hook you up with the training you need.
Our next free webinar we're moving to a monthly delivery model so our next free webinars taking place next month. So Thursday, February, the 18th at 1pm. And this lecture will feature professor, Laura Reese on the topic of emotional intelligence in the workplace. She will be challenging you to question how rational and bias read your own decisions and actions, really are, you'll be getting more information
2021-01-27 00:36