AI at IBM Think 2019

hear a bell of the famous San Francisco trolley. Perhaps you'll even hear bagpipers, because we did. But I'm here in San Francisco for a specific reason. IBM invited me to fly out here and attend the Think two thousand nineteen conference and really get an up close and personal view of some of the innovations and services the company is rolling out all to their clients, their business partners, and I really wanted to share with you my own

takeaways from this event. Now, before I jump into all this, IBM is a business to business entity, meaning that if you're an average Jonathan like me, you rarely deal directly with IBM. But the company is one of those leading entities that provides the tech that other companies use in order to do their business. So while it may or may not be obvious, there's a lot of stuff that we encounter in our day to day lives that's powered

by IBM. This episode is the first of four special episodes about the conference and the technology and innovations that are at the bleeding edge of deployment, and today we're going to focus on artificial intelligence, something that you could

argue is almost synonymous with IBM. So you guys know that AI is one of my favorite topics to talk about, and it can be easy to fall into the trap of thinking about AI as some sort of nebulous intelligence living in a machine, But when you strip away the veil of mystery, you'll see that AI is just another

part of computer science. It might rely on one architecture over another, or it might require an artificial neural network approach, depending upon the application, but really it just comes down to a series of special algorithms designed to handle information in a way to allow a computer to make decisions.

It's sophisticated and it's fascinating, but it's not magic. However, you might be forgiven for thinking of it as magic if you happen to witness the exchange between IBM s AI System Project Debater and Grand Champion debater Harish Natarajan. The debate between man and machine happened a day before the official start of the conference. The two participants of the debate we're not told about their topic until fifteen minutes before the debate was to begin, and then they

were given their stance on what that topic was. They each had four minutes to establish their positions on the subject. Then after a short break, they had another four minutes to offer rebuttal of their opponent's stance, and then one short break later they had two minutes to summarize their arguments. The debate topic turned out to be preschools should be subsidized. Project Debater, who, by the way, has a gender. Project

Debater is a she. She was given the pro stance on that particular argument and Mr Natarajan got the counter stance the the the idea that preschools should not be subsidized. I am not going to go through a blow by blow of the debate. For one thing, you can actually listen to it yourself. Intelligence Squared, which is a show dedicated to civil debate on a wide array of topics

played host to this particular special debate. I urge you to seek out that podcast or a video of the debate if you want to see how it unfolded for yourself, bit by bit. I really just want to talk more about the process that was involved. So to debate, no matter what you are, whether you're human or machine, you need to have an understanding of what it is you're either arguing for or against, which is a pretty obvious statement, but I feel like I have to lay it out

that way. You need to be able to form an argument, and you have to be able to support that argument logically. You want to build your argument so that one part leads inevitably into the next part and it all supports the stance you have, whether it be for or against a particular proposal. This is a non trivial task for a human being, and it is an incredible challenge for computers.

Project debater Or has about ten billion sentences worth of data stored in its memory, So when a gets a topic, first it has to scour all of the information that is in its memory banks and look for relevant information related to that topic. Then it has to go a step further. It can't just pull up any random information about the topic. It has to understand that the information

actually supports its argument. That is, the computer has to make sure it is picking information that is aligned with its debate position and not actually against its debate position. This falls into the field of natural language processing, and I've talked a lot about this too, but in short, this describes the area of computer science in which we try to find ways for machines to suss out the

meaning from actual human language. At the basic level, computers communicate in machine code and we communicate in human languages. Machines don't natively understand human language, just as machine code would appear to be nonsense to us. The journey to creating powerful systems that use natural language processing to figure out the meaning of words, whether they're written or they're spoken, it's been a really long one, and many people have

made advancements, sometimes from completely different perspectives. We've got a lot better at this in general, but it's still a challenging problem. So think about Google Search for a second. When you search for a topic, you type your search terms into Google, and then you look at the results. You typically get a pretty wide variety of responses. Some of them are going to be more relevant than others. You might even get a few that aren't relevant at all.

Google's algorithms attempt to guess at which responses will be the most relevant based on your search and sometimes on some supplemental information like your search history. But sometimes the results aren't ordered in a way that you would prefer. You might get an okay response at the top and maybe a better one or more relevant one two or three spots down. Typically it ends up being on the first page, but sometimes it can even be buried lower

down in the search results. Project debater can't just do a simple search and return on key terms, or else it might end up spouting out gibberish. It could string together two or three sentences that contradict each other that wouldn't do anyone any good. And that leads me to another point. It's not just good enough to grab relevant information that aligns with the argument stance. That's necessary, but it's not enough. Those statements have to be ordered properly.

You have to build support for your stance. You have to have this logical progression. A good argument needs that from the opening to the closing, so you need to make sure there's a flow of information. Otherwise all you'll get is a series of relevant points, but they're in no particular order, and you have no transitions from point to point. It would be jarring and it would be ineffective. Project Debater could also support arguments with evidence, which is

kind of cool. Throughout the debate, we heard the system site various studies and quote experts in the field to provide support for her stance. This was pretty compelling stuff, and this is where Project Debater could be incredibly helpful for people who want to argue for or against well anything. Really, it's the one area I would say that Project Debater

had an enormous advantage over the human champion. Harrish. Natarajan understands how to create a logical, persuasive argument and how to find weaknesses in the arguments of opponents, but he can't research a library's worth of information in fifteen minutes in preparation for a debate, but Project Debater can. However, I wouldn't feel too badly for Harrish. He held the

advantage in lots of other ways. For example, during the debate, he brought up a criticism of Project Debater's argument, pointing out that one of her conclusions was based without first establishing the evidence needed to support were in it. In a debate between human champions, you would likely hear during the rebuttal phase a response to this, perhaps including some of the evidence that might have been left out previously.

Project Debater didn't really address that criticism. We also found out at the end of the debate that typically the Intelligence Squared format would include another round the moderator would hold around in which you would ask critical questions of each of the participants in order to test their arguments and their logic. This is done in the normal debates on Intelligence Squared, so if you listen to other examples,

you would hear that round. But Project Debater, while it's really impressive, isn't quite up to the task of handling that sort of response just yet. And since this was really a showcase for the technology, that round was not included in this debate. I was really impressed by Project Debater, and as Harish pointed out after the exchange, the technology has the potential to really help people get a deeper

understanding of complex topics. They can use it to help them support their arguments on any given stance, or and this is something I think is equally as important, they could use Project Debater to produce counter arguments to their own stances. Then they might either better learn how to argue against those who oppose them and anticipate the arguments they would put up against a specific stance, or it might actually change their own mind about the entire subject.

It might be that you have a preconceived idea of what is right, and then you get the information from Project Debater and you start to question those ideas. You might change your mind. That leads me into the next section IBM's general philosophy about artificial intelligence. But before we get into that, let's take a quick break. A common thread that was once in science fiction and now tends to be in today's headlines is the impact that automation

artificial intelligence will have on the workforce. On the most pessimistic side, there's a fear that these technologies are going to eliminate millions of jobs and that will be plunged into an economic crisis, perhaps requiring an entire overhaul of how we think about work and money. And there will no doubt be jobs that will become either completely automated or automated to the point that fewer humans will be needed to carry out that same amount of work over time.

So there's definitely some validity to that fear, but many entities, IBM among them, say that we're probably not going to see anything quite so dramatic as a job apocalypse. Instead, IBM's vision is one in which artificial intelligence acts sort of like a super smart, super efficient assistant to aid

us in our jobs. The tedious or difficult parts of jobs that humans find troubling could be handled by artificial intelligence, whereas the parts of jobs that are easy for humans but not so easy for machines would still need a person taking that position and fulfilling those parts of the job duties, and artificial intelligence will necessitate new positions in order to oversee the systems and to maintain them and

grow them over time as businesses themselves grow. I had the opportunity to sit down with Rob Thomas, who is general manager of IBM Data and AI, to talk about this. Now. We're sitting in a lounge in the w Hotel in San Francisco for this interview, So if you hear some ambient noise that's just the sound of business people being business in the background. I'm sitting here with Rob Thomas, general manager of IBM Data and AI, and today we heard the announcement of Watson Anywhere, and I have to

ask you what does that mean, Jonathan. It's an exciting day for us. Let's start with basics. I like to say there's no AI without I, A meaning information in architecture. AI is only as good as the data that you feed it. So that's a problem every company deals with and you can even see it in your consumer life. If you're using an app over and over again, it starts to know you a little bit. So your AI is only as good as your data. What we realized is companies have a lot of data, but they have

a data in a lot of different places. It might be in one office location, might be data in a different office location. There might be data on a public cloud. They might have different cloud providers. We made the decision that we were going to bring the AI to the data and enable that to happen. So Watson Anywhere is about taking the best of what we've built in Watson and saying you can have that wherever you want it,

which is normally wherever your data is. And this is going to be significant because this is what clients have been asking for and now we're making it really easy for them to consume Watson AI wherever they have data. So if if I'm understanding this correctly, you can look at this in very broad sense in two very different directions. You can look at it in the sense of I've got this big company and I have data spread out through multiple locations, and maybe I need to integrate that

in meaningful ways. That's one way, but it may also mean I have a really large company and I've got offices in other states, other countries, perhaps where that integration may not be as easy or seamless. There might be specific laws, for example, when you need b are over in the UK, where I need to be very particular with how I'm handling data in this region. I might

not be applying that somewhere else. In this way, if I'm taking the AI to where the data is, I can handle those different use case scenarios in the specific ways that are necessary. So it could be either way. It could be like integrating stuff, or it could be applying for specific implementations and I am I a lot of companies have different security policies for what they can do with their data. Like you say, some are worried about g d p R can't leave a certain country.

So we're just saying take the best of the AI, put it wherever it is however you want to do it, which makes it really easier for them to access, which kind of brings up the idea of So what is Watson? What is AI? We can talk about that for a minute. Sure. I think there's two worlds of AI right now. One is people that want to build their own AI. So Watson is a way that you can build run your AI,

manage that. We have a product called Watson Studio Watson Machine Learning that's basically how you build run manager AI. That's what the data scientists of the world do. They want to build something unique for their company. There's another world that says, I don't have those skills to build my own AI. I just want to use AI. We've

built some Watson applications. We have Watson Assistant, which is basically a customer service agent encoded in software where we automate a lot of the decision making to make current customer service representatives a lot more effective and how they can support customers. And we've got another application called Watson Discovery. Any company with a lot of data in different places, they want to discover all of their data, what's in there,

looking for the proverbial needle in a haystack. So there's kind of two worlds of a I I want to build my own and then I want to build for me. Just help me solve a problem I know I have. That's what Watson is today, what we're doing, and so

you've got the with these two approaches. I also like the idea, and it's you sort of alluded to it that this AI is really all about augmenting us, not not that you're replacing any sort of human element, but that you're augmenting what we're already doing, making us more effective, more efficient, being able to find more meaning in that data.

One of the stories we hear all the time is about just this concept of big data, this massive amount of information that we're constant only uh have at our fingertips, but it's so big of a problem that it's hard to tackle. This is another approach to doing that in a meaningful way, where you're actually able to h to create action plans based on all this information. You have to say, well, we've got all this info, what do

we do with it? To me, that's a fascinating part of this as well, because obviously one of the boogeymen in tech is this concept of of AI. It's usually a misunderstanding what AI is, and ibm S approach has been No, this is really about enhancement, not about replacement. I like to say AI is not going to replace managers, but managers that use AI are going to replace managers that don't. And it actually it's a wave. It gives

you a superpower if you're willing to use it. Give you an example, Royal Bank of Scotland, big retail commercial bank. They're trying to serve all of their ret hell banking customers. They're using Watson Assistant for customer service. They're getting faster answers. It's still going through their agents, but their agents are using the Watson Assistant to say do I understand that question? What they're really looking for? So they're getting faster answers,

they're getting better answers, you get more satisfied clients. So that's why I go back to what I said, Managers that use AI have superpowers, and so I encourage everybody to be open to that because it makes you more effective. Means you have to do less of the boring work. We all have boring work we have to do. You can automate a lot of the boring work. So I think it actually makes jobs a lot more interesting, which

is exciting. And as an average user and average person, the one of the results is that you get you just get better results when you're using these different services that are incorporating the AI, because you're getting the right answer, You're getting the right answer faster. Uh, you don't have to worry about as much follow up. So it's takes a lot of the frustration out of those interactions between customer and say, uh, you know a customer representative. I

think I've been on both sides of that. I've been on the side as the customer who's frustrated, trying very hard not to let my frustration spill out with my interaction representative. And I've I've married to a woman who was a customer representative who would come home and I would I would hold her for an hour because she

had been nailed at eight hours straight. So uh, that is something I think that a lot of people lose in this too, because when we have these discussions, we're talking about the enterprise level frequently, and the average person says, well, how does that affect me? It affects them because this ends up being incorporated into applications that are forward facing for customers in some cases. So I'm really excited by this.

I think Watson is an incredible platform. I've had various interactions on various levels with Watson throughout the years, ray raging from seeing how it could be used in a customer service aspect to Watson Chef, which was my favorite implementation I've ever seen, just to get weird fun recipes generated by Watson Um. Last night I saw the debater presentation, which was fascinating to see this platform be able to actually put together a cohesive, coherent argument from lots of

different points of data. That is a phenomenal achievement. People don't realize how incredibly difficult that is. What excites you most about where we are with AI and where you see us going in the future. Let me give you an example that I think hopefully everybody can relate to, which will kind of bring it to life for how AI is not just impacting businesses but individuals. AMC Networks is a big client of ours, so I'm sure some of your listeners have seen some of their TV shows.

Breaking Bad was a popular one, but they've got many and AMC networks challenge was, how do we understand what viewers are responding to, what are they liking? And can we adjust plot lines based on what we're hearing and what they're liking. They also, you know, their businesses also around advertising, so can we give advertisers an idea of when to engage with the users, how to engage with

the users. They're using our technology on cloud where they federate data coming from step top boxes companies like Nielsen third party data. They bring that behind their firewall and the company to manage that data. And then they're using things like Watson Studio to build models and say hey, you should go do this kind of thing to an advertiser, or even reaching out direct to a consumer saying we

thought this show might interest you. So AI is behind the scenes actually everywhere, and I think sometimes you only notice that when you kind of have an aha moment where you're like, wow, that felt magical. It's it's actually AI is not magic, it's just computer science. But it is impacting every individual, whether they know it or not. Today. That's incredible. I only have one last question for you, which is are there any questions I should have asked you?

I wish you would ask me why IBM and how is how is IBM relevant to this space going forward? I'm sorry I did have one more question. Okay, why IBM and why is IBM relevant in this space going forward? Great question. IBM has an amazing history over a hundred years old, and I think we've always been a steward of responsibility and integrity. And when you work with IBM, you know what you're going to get, which is you're going to be satisfied. This whole area of data AI

make some people a little squeamish. They're worried about lack of transparency. They're worried about is my data being shared? The best part of working with IBM and being part of him is that you know your data safe, you know your models are safe. You know IBM not sharing this with anybody else. You know that we will be the stewards of responsibility and AI. That's why last year we came out with explainability and bias detection for AI.

I think we're the first company to do that, because there's a lot of things that can go wrong in the world of machine learning or AI unless you're thinking about societal impacts, human impacts, and we spend a lot of time on that at IBM. So that's why I'm very optimistic. Absolutely, and yes, you definitely don't want something like artificial intelligence to become a black box technology where you have no idea how it's making its decisions behind

the scenes, because obviously that breeds mistrust and unease. So I'm very happy to hear that as well. I remember at the Think Conference last year, I was at those those presentations and I took away I was really impressed

by the discussions about bias and transparency as well. It's something that a lot of people have been arguing for, and to see a leader in a space take that very seriously is uh is a great relief and it gives me a lot of optimism about the future that Mr Thomas, Thank you so much for taking time to talk with me and my listeners. I really appreciate it

great being here. Thank you appreciate it. Now up to this point, I haven't really mentioned Watson, but it's a good time to remind ourselves that Watson is an artificial intelligence platform. IBAM made a really big splashback in two thousand eleven when Watson competed on Jeopardy against two humans who were former long reigning champions, and you can see in Watson some of the same concepts that emerged more

evolved in Project Debater. The system can parce language, including more subtle stuff like wordplay, figure out what is meant by that language, and then evaluate what the proper response should be. If the evaluation meets a certain threshold of confidence, then Watson will submit it. Otherwise it kind of keeps its electronic trap shut. Now, in real world deployments, Watson rarely has quite so difficult a task to perform as

to compete in Jeopardy, which is all general knowledge. Typically, Watson is working within a fairly well defined set of parameters for its implementation. For example, a car insurance company using Watson to help with customer interactions wouldn't have to worry about someone asking what the capital of Belgium is or what's the best barbecue restaurant in Atlanta, So by showing off Watson's potential on the grand stage of Jeopardy, IBM was able to lay the foundation for a pretty

convincing sales pitch. Yes, Watson could do all these amazing things, but imagine what it can do when it focuses on a very particular industry such as healthcare. That's really where IBM's focus has been. I have more to say, but first let's take another quick break. Another interesting thing that Mr Thomas brought up was the idea of taking artificial intelligence to the data as opposed to the other way around,

and I think that's a pretty smart move. There have been so many high profile, high impact data breaches over the last few years that I imagine most companies are pretty reluctant to move mission critical information if they don't

have to. The potential for something to go wrong, for some bad actor to find a vulnerability and exploit it and thus get access to private information, or perhaps worse, for the process itself to go wrong and to accidentally dump information into the public sphere without any need for

outside interference. That's enough to make any company decline incorporating AI if it means porting data over to where the AI is so by making Watson available to companies to run on their own private cloud or on premises or on prem as they say here at Think two thousand nineteen, or in the public cloud. This is a huge deal.

It removes that barrier of entry. Now companies that are interested in using AI with their services can do it without the worry of having to move their own data around and her keynote speech kicking off the IBM Think two thousand nineteen conference, Jenny Rometti, the CEO, president and chair of the Board for IBM, spoke about this. She outlined two general approaches to incorporating AI into services. One

is what she would call the outside in approach. This is where companies would take their pre existing applications and then they would add a layer of artificial intelligence on top of those applications in order to make them work better and more efficiently. This is an approach companies might take if they lack the expertise or time to build out new apps entirely. But some companies might opt to

do the reverse, the inside out approach. In other words, that's where they create all new applications and processes that incorporate AI into them from the beginning to try and maximize the value of having the artificial intelligence involved. So what the heck does all that mean? How does that impact us as average people? Well, largely, it means the service as we use, such as mobile apps or computer software, will work better, become more sophisticated, and they will incorporate

more features. And this will become more important as companies continue to grow and place data in different data centers and clouds. As that happens, it becomes increasingly challenging to manage all that information and to coordinate between those centers of data and pull together meaningful results. It helps if I give you an example. So let's imagine that you have downloaded a travel app and it's all to help you plan and book a trip you want to take.

Maybe you're traveling to another country in six months, so you're planning well in advance. The app helps you by consulting many different sources of information. It prices out flights through various airlines to help you find one that fits your budget and your schedule. It gives you information on hotels at your destination to have availability. It provides a list of possible activities you might want to do while

you're there. Maybe the includes a calendar that's populated with cultural events that are going on at the location while you're there on vacation. There are options for restaurant recommendations, information on average weather during that time of year, so you know what sort of clothes you're gonna need to pack, and maybe more pieces. Maybe there are things that will populate over time so that each time you go back to consult your plan, it's update with the latest information.

These various pieces of information don't all live on one server somewhere connected to that app. There's no business out there that has all of this information stored on some magical computer. Instead, the information is coming from numerous sources and organized in a meaningful way for your mobile device interface.

The goal of an AI approach is to have an automated system in place that's able to do this kind of stuff quickly and without error, so that the end user you, in other words, ends up with a seamless and helpful experience, so that you get the information you need, when you need it and where you need it. And

this is harder than it sounds. And the general message at think two thousand nineteen that I've been hearing is that it's not a question of if companies should start employing these sorts of AI approaches in their processes, but rather when they should. That if you don't do this, it means you're not going to be able to keep

up with the demands of business and growth. So it sounds like we've got a future of artificial intelligent assistance ahead of us, and I think that's pretty fascinating, especially when we think of it in the context of augmenting what we humans can already do, not replacing what we can do. That's a pretty cool message and one that I really found inspiring while I was at Think two

thousand nineteen. I'm going to have a lot more episodes coming out in the near future about some of the other things that I am looking into while I'm at the conference, including some more interviews with some really interesting people, So make sure you stay tuned and check out those

when they published. They'll be coming out very soon. If you guys have any suggestions for future episodes of tech Stuff, whether it's about a specific company, a specific technology, maybe you have follow up questions about some of the stuff I'm talking about this week, make sure you reach out let me know about those The email address for the show is tech stuff at how stuff works dot com. You can also visit our website at tech stuff podcast

dot com. There you're going to find an archive of all of our past shows, as well as links to how to connect to us on social media and to our merchandise store. And that's it for now, but I'll be back again in just a short while to talk more about the incredible stuff I'm seeing here at Thanks two thousand nineteen. Thank you very much, IBM, and I will talk to you again. Release soon for more on this and bathands of other topics, because it has to works dot com