Artificial intelligence (AI) has been changing our lives for decades but never has it felt more ubiquitous than now. From Siri to self-driving cars, AI is progressing rapidly and is becoming embedded in our everyday lives.
Susie Sheldrick works with technology consultants Silverpond, who are empowering 7‑Eleven, Powercor, and Australia Post with AI. Susie reveals that we don’t know how AI makes decisions and that’s what makes it such an exciting field to be working in.
Mark Chatterton is co-founder of inGenious AI, who are using Natural Language Programming (NLP) to create chatbots with a human feel. Mark’s goal is to eliminate interfaces with AI so that we can talk to them in the same way we talk with friends and family.
Disclaimer: Transcripts may contain a few typos. Similar sounding words can lead to them being deciphered wrongly and hence transcribed likewise.
Interviewing Public: I think from my perspective I’m looking at it and how to use AI in data science and data analytics so that I can customize and personalize customer experience better hopefully get insights groups customer into segment of one.
Interviewing Public: Well the particular business that I’m employed by uses it for vision systems, drone vision systems for the defense industry so basically are in involves object detection object tracking in video from a drone.
Serpil Senelmis: For WeTeachMe, this is the Masters Series where industry professionals share their secrets to success. I’m Serpil Senelmis from Written and Recorded. Watch out the robots are coming or are they here already? And should we be worried? If you stop to think about Artificial Intelligence for a moment, the reality isn’t anything like we see in the movies. The fact is, AI has seeped into almost every aspect of our everyday lives. And in a major way. Can anybody say, Alexa or Siri? So it’s not surprising then that machine learning and task automation are widely used in business these days. But how will AI continue to shape the future of business? And what do businesses need to do to keep up with AI in order to gain the most out of their marketing efforts? Mark Chatterton is co-founder of inGenious AI, which designs and builds chatbots using conversational interfaces and artificial intelligence. Mark says high accessibility is making conversational chatbots a game-changer.
Mark Chatterton: Just like today we’ve gotten up and started talking the closest thing to an interface is I need to speak English. That’s it. So there hasn’t been an interface you’d need to learn you don’t need to start to understand the buttons on my face or anything like that. It just starts talking.
Serpil Senelmis: We’ll hear from Mark soon, but first, Susie Sheldrick. She’s the marketing and community manager at Silverpond, a technology consulting company. Silverpond has been working with startups and organizations such as 711 and Australia Post to help them navigate major technological trends, including AI. Susie outlines the different types of AI that are available, and the scary thing, apparently, we don’t know why AI thinks the way it does.
Susie Sheldrick: Yeah, I’m from Silverpond, where a machine learning company where a team of data scientists, technologists, marketers, and software engineers, and our aim is to accelerate the adoption of AI. So just because we have that range, we’re just going to set a bit of the scene of what what is AI and machine learning and deep learning. So we’ve got artificial intelligence which is like computers mimicking human intelligence. From there we’ve got machine learning which as it sounds is where we teach machines how to learn, instead of programming the rules and it finding out the answers, we’re giving the machine’s problems that we don’t quite know how to create the rules for. So the popular internet option is cats, for example, setting the rules of how to give a computer a new image of a cat and define that is actually really hard, giving the computer an image of a cat and giving them thousands and thousands of that it will learn its own way of defining that is a cat. So that’s sort of the benefits of machine learning. And deep learning, which is taking its inspiration from the way the human brain works, and that’s why you’ll see a lot of brains been associated with deep learning and AI because it’s using those neural networks. So we’re going to talk about a few of the applications that we at Silverpond have been using deep learning predominantly, but especially a lot of machine learning. Sports analytics, traditionally, things like Hawkeye have been huge in Grand Slams and professional tennis. But that requires eight cameras in a very complex computer network. What we’ve been doing is working out how to use object detection. And instead of having these eight cameras, it can be reduced to the cameras are already being used by the television networks and a computer model. Road analytics, traditionally used are these rubber tubes. But the rubber tubes can’t tell you what type of cars are going over. And that’s really important for the road authorities to know like Alamo commercial vehicles being used, the sizes and that kind of thing. Asset detection to do these sorts of visual inspections requires a person either going up in a helicopter or on a crane with lots of costs and our h&s considerations, we’ve been able to use machines to detect not only the number of assets but also are they in good neck. Time series data for this is about the measure of smart meters over the course of time. And time series data is used a lot for prediction like stock market predictions, if you ever see machine learning and the stock exchange, they’re using time series data. And this is a bit of a scary thought for people the first time they hear it, but we don’t fully know why AI thinks the way it does. And makes the decisions that it does by giving it the opportunity to learn its own rules. We’re still trying to find out what those rules are. And so trying to understand how AI comes up with the decisions it does, is really important, and that’s called explainability, which goes into like, the transparency where machine learning gets really cool and interesting is when it’s doing things better than people. So previously, this case had a detection rate of about 40% based on people analyzing the images that went up to 70 to 80%. within the first few weeks using the Silverpond model. Ai isn’t perfect, it can still be tricked and this is an important case for when it was going with autonomous vehicles, where the model was able to be tricked with just stickers being placed over the stop sign. So these are things that we need to overcome before we have the autonomous vehicles running on the road. Back to interpretability. So we’ve got a lot of really great ideas, but understanding why AI makes the decisions that it does is something that does need a lot of research when it’s taking cats from the internet. Not so important when it’s being used in justice systems like it is in a number of countries and especially the states, they’re using AI models to sentence people and you know, decide on their parole, their bail, things like that. And so understanding how it is that the AI came up with that decision is incredibly important in situations like that. Bias, so it is from a story where this lovely young gentleman didn’t fit the passport criteria because the AI model said that his eyes were close which clearly they’re not the problem there was there wasn’t a wide enough data set used to train the model. Yeah, you really need to have a really good data set and a diverse data set, depending on your needs, of course. But if you’re detecting people, you need to have all different types of people in your data set. If it’s bad data going in, you’re gonna have bad models coming out. If it’s not diverse going in, it’s not going to be very effective coming out.
South Korea has got the greatest density of robots in the world. They’re really at the cutting edge with robotics. And so holding such a world event, such as the Winter Olympics, they wanted to really make the most of that. So they had about 85 different robots on the grounds doing a number of things. There was the robotic fish in the fish tank. I’m not sure what the purpose was, but they were there. Then you had greeters at the airport that could answer basic questions and direct people towards another human that could get them in the right way. They had them doing planning. So essentially that the robo-backs on steroids as well as drink dispenses, and things like that. And then we also had the robot skiing championships, which wasn’t an olympic event. But there were plenty of winners, and quite funnily cash prizes for the robots, but not for the people competing in the game. But what was really interesting about this, so we’ve got Korea, which is probably the most advanced country with robotics, and certainly right at the cutting edge, and this great big world event. And yet each of these robots had a person by them the whole time, they weren’t confident enough to have the robots roam free, or they weren’t confident enough with the people. So I think that says a lot with where we actually are with maturity and AI and robotics is that they’re doing some really great things, but it’s not quite at the stage of letting them sort of run free and take over as many people are worried about. Another thing to be mindful of with AI is the ethics and regulation and thankfully we’re seeing a lot more interest happening in this. The days of the engineer going, I just build the thing someone else can deal with it as heartfully becoming behind us where people are actually being mindful of what they’re creating now, and where that’s going to be in a few years time. To that as well governments are really getting on board with making sure they’re not making the laws 20 years too late, which is traditionally happened with technology the big government has done in Australia first with creating the all-party parliamentary AI Committee, which is fantastic that they’re getting this together and making sure the right regulations are in place, as well as the right support for the industry in Victoria. And then we’ve got things like the AI Now Institute, which is a leading ethics body. And then universities such as Harvard, MIT, and Stanford are really leading the way with offering ethics courses within their computer science degrees so that the people that are going to be graduating within years to come, are going to have that background of thinking what are the implications of what I’m creating. So if you’re wanting to get started really start small and experiment. Play around with some great tools online and data sets you can play with tools you can start to play with. So start small experiment and work out what’s right for you. And that’s all we’ve got for today. Thank you.
Serpil Senelmis: So well AI is often better than humans at detecting detail, it’s still not perfect and can be tricked. In a moment we’ll meet Mark Chatterton, watching large corporations cut corners in customer service drove Mark to want to deliver a better customer service experience. He found the solution in chatbots. You’ll hear more on that right after this.
Ad Guy: If you’ve ever wanted to upskill learn something new or just follow your passion, WeTeachMe can help you get there. WeTeachMe is like the eBay of lessons. We can help you become a barista, a baker, or a blockchain expert. You will find all the best workshops, classes, and courses happening around the country at the click of a button. Find classes near you at weteachme.com. And every week WeTeachMe’s Masters Series podcast connects you with industry professionals to help your startup to take off. The Masters Series podcast is brought to you by Written and Recorded. Journalists for hire, Written and Recorded create memorable content that helps your business rise above the noise. From blogs to podcasts, social media to feature articles, Written and Recorded will shape your content to bolster your brand. Find out how content can amplify your message at writtenandrecorded.com. And now, back to the robots.
Serpil Senelmis: Mark Chatterton is co-founder of inGenious AI. It’s a company that creates conversational interfaces, which enable customers to communicate with companies in the same way they would with friends. Mark says by 2021, more than 50% of enterprises will spend more per annum on bots and chatbot creation than traditional mobile app development.
Mark Chatterton: So what we found is every 10 years has been a generational shift in interfaces. So what we mean by that is, basically, in the late 90s, the websites with the cool thing to do so we’re all building websites. So that’s been going for quite a while now. In 2008, Apple released their ecosystem for building apps on the iPhone or whatnot. And then the app started to explode and people were getting lots and lots of apps on there. And finally, in 2018, we’re seeing conversational being the new interface, and what we mean by conversational is just like today we’ve gotten up, we’ve started talking, the closest thing to an interface is I need to speak English. That’s it. So there hasn’t been an interface you’d need to learn, you don’t need to start to understand the buttons on my face anything like that, it’s just start talking. And we can actually start to have a two-way conversation like that. So it’s that very high accessibility, that enables conversational to be a bit of a game-changer. So what is a chatbot? The chatbot is basically a robot that understands that wage and can process it and start to do actions off it. So you know, screaming at your Alexa to turn the lights off and things like that it can actually do it in action. So the first thing is to understand the difference between conversation and search is one of the big challenges. Search is all about asking a question into something like Google. But the problem then is you kind of have to know the right question to ask. And then you have to do all the work. So Google, you’ll fire off a question like, Hey, I’m looking to purchase an SUV. And Google sort of taunt you and say, here’s 9 million results in point five, seven seconds. Like, that’s great. Now we got to read through all of them to try to figure out what car is right for me, it’s great that Google can sort of give you all that information very quickly. And it tries its best to try and put the most relevant stuff to the top. But still, it’s pushing all the work back onto you. But think about how a conversation goes. So if you call up someone or go into a dealer, and you’re not sure about what type of car you want, there’s a conversation. There’s an interaction that goes, well, what are you looking forward to? Do you have a family? Do you want to tow a boat? Looking for a sports car? You trying to compensate for being bored? That type of stuff. So it’s all these bits of information that come in through that interaction, and you end up narrowing things down. So it’s this ping pong of conversation that goes through. And one of the important things is that as a human, we’re kind of pretty crap, basically, the computer can do a lot more than us in the fact that we can maybe remember in the one sort of conversation three to five things at a time, but going beyond that, it starts to really stress our cognitive load, as they call it. So keeping that cognitive load low enables us to do a lot more, a lot more efficiently. So asking one question at a time and bouncing back is nice and simple, asking me to think about all the different parameters in like, one blurb straight out is quite challenging. And also, it’s from a momentum point of view with conversation. It’s not so much the number of terms, but the amount of thinking that required to progress. So as a quick example, when you go into like a barista or something like that, they’ll ask you, hey, hey, going, what would you like? Not what do you want, right? It’s not about shortening the experience to the minimum thing, like, you know, if they really want to be rude to be what? So, but if that’s a hell how she interacted with you, there’s no way you go back in like you rude burger. Right? I’m not doing that. So understanding that conversation is a lot more nuance to it, and it’s a lot more about the experience.
Mark Chatterton: So what’s the technology that drives a lot of the chatbots? Well as the silver home to humans sort of mentioned the natural language processing or NLP is a big part of it. That’s all about understanding not just the words, but your intent. So for example, take the word hot, so hot can mean spicy can mean temperature can mean sexy can mean stolen. So there’s all these different meanings for the same pronunciation and the same spelling, how the hell is a computer gonna understand what we’re saying? So if you boil that down, it comes down to context. So if I’m talking about a pizza, probably not talking about a sexy pizza or a stolen pizza, it’s going to be a spicy pizza. So it’s understanding that context. Now while the NLP is getting good, it’s nowhere near foolproof today. But they’re improving it. The other main technology is speech recognition. So just speech to text recognition, so being able to translate what we say so that the computer can have a chance at understanding that feeding into NLP. So the way it works is the speech recognition, understands our words, converted into text, fix it through an NLP to understand the intent of what we’re doing, and then starts to do an action. So for example, what’s the weather Nolan, translate the text sends it off to NLP, yep, Melbourne, Australia, send it back. And then finally natural speech generation. And that’s all about making sure that we can understand the computer back to us so they can generate the speech in a natural way. So you’ve probably heard of Siri and things like that talking to you, they’re slowly getting better on the way that that voice is coming back to you so that it does sound as natural as possible and not as disjointed as the old school robotic sort of wording that used to do so the whole sentence links together, it’s not just saying the words after, after each other. So one of the types of chatbots. So just from a very high-level technical view, there’s kind of rule-based ones these are fairly simplistic in the fact that it’s a big If This, Then, That. So if I’m looking at home loans, then this is the content that you want to look at. So it’s fairly simple to build from a flow point of view, but it doesn’t suit all scenarios, things like that. The other way is what I call AI-based or NLP based and this is typically Siri engagements or Alexa at the moment is short, sharp engagements where it’s question answer, question, answer question, answer. There’s not too much of a flow to it. So what’s the weather in Melbourne? It’s raining as usual, not a lot of depth to it. And finally, the best model that we’ve sort of found at the moon is actually a hybrid approach. So, as an example, if you’re a bank, I can start the conversation of, hey, what are the fees? The bots going to go? Well, like a human would? Well, this, we’ve got lots of stuff with fees or products that you’re interested in. So then they can go home loans. All right, cool. So let’s talk about home loan fees. But if I was already talking to you about home loans, and I asked what are the fees? Don’t come back to me and say, what products do you want? Like I’m obviously been talking to you about home loan. So tell me the fees about home loan. So understand that context. And the context is one of the big challenges with chatbots and making sure that it’s managed correctly and set correctly. So go some quick examples. This is a very simple messaging chatbot to gift bot called giddy, and they’ve got a really good engine at the back end that’s done. I guess. quirky gift recommendations. So as just some very quick questions about your friend that you’re trying to buy the gift for, and then it’ll start to provide a recommendation. So it’s just basically a bit of a q&a that it does and then provide some smart recommendation. And I found this with the combination between the channel being the Facebook Messenger, and the really smart AI engineer the back end doing the recommendations. We’ve got about a 30 to 40% conversion rate on doing things like that. So the combination can be quite powerful. Why Facebook Messenger, especially for Australia, the main one is that there’s 12 point 4 million monthly active users on messenger. And to give you an idea of Australia’s population, if you remove under 13 and over 75 because they’re still running letters that is about 70 to 75% of Australia’s population that monthly active users on that channel, so it’s a huge chunk. It’s also a persistent channel. Just like with your friends and family. If you start chatting, it’s a channel that you can jump back into at any point and continue the conversation. You’ve got a history and things like that. So the same thing can be with the chatbots. And to give an example, say for example, you’ve been talking with Australia Post, you’ve identified yourself and things like that being able to jump back in and go, where’s my parcel, as opposed to busting out the tracking ID and all the other bits and pieces just with my parcel? Because that’s directly getting to the answer you’re looking for. That’s pretty feature-rich. So you can do location sharing, you can share images, short video clips, it’s interactive. And you can do things like account linking and stuff like that as well. So you can match it with your Australia Post account. Obviously, importantly, no app downloading, which is a big thing at the moment, like getting people to download apps is pretty tough. Like, just think about how many apps you’ve downloaded recently, I think there’s a statin that’s coming out of the US that the majority of people haven’t downloaded an app in over 12 months. So it’s pretty tough now to get people to download new apps. So chatbots are a very quick and easy way because it’s already in an ecosystem that people are playing with. And messaging apps are becoming the number one place where people are spending their time the moment as well. So it’s taken over from Instagram and Facebook and other social media platforms. Now account set up. So there’s a fair bit of information that’s out there that’s provided as part of that initial conversation, like their name and things like that. And finally, notifications are enabled. So because their friends and family are already talking to them on this on this channel, most people have enabled notifications on by default so they can get their messages from their friends and family. So as a business, you can reach out to them and say hello as well. But obviously be conscious of the fact that if you abuse it, you will be using a very sensitive channel for two people. So example of voice so Alexa’s just come out in February in Australia, but it’s been out in the US for going on three or four years now. So it’s got a quite a head start. So obviously, you can just ask it to pay your credit card bill and things like that. So once you’ve done the account linking side of it, which is a little process you go through, you can then start to ask it to do very simple tasks like what’s my balance, please pay my credit card bills, things like that. And it goes through a very simple flow from that point of view to say, alright, well which account do I want me to transfer the money from to pay card, that type of stuff.
Mark Chatterton: So as an example, large bank that was part of the launch, for Alexa, they’re only getting about 200 users a month thru they bought at the moment. So voice within Australia is still a very niche thing but it will start to grow very rapidly once the devices get out there. And so Alexa is kind of cool. So you can play with the voice of Alexa a little bit. So you can whisper you can pause and add basic emotions in there. So you can play with the actual way that Alexa talks back to you and things like that. And notifications are enabled. So on the voice platforms, you can actually do push notification, so they just have the for Alexa, the ring goes yellow. So when you walk in after work or something like that, the yellow ring, you can kind of go like so what are my notifications and it will start telling you. So when should you consider a chatbot? There’s a lot of use cases that people think would be great, but these are some of the basic rules that we use for clients, big payoff little effort. So asking for an Uber to the airport and saying you know, Alexa, book me over to the airport is a lot quicker than opening your phone, unlocking your phone and going into Uber typing in airport, that type of stuff. So it’s probably like a minute to go through the app versus three-second soundbite, from that point of view. It’s more convenient, so this whole process here is what they call the raw chicken moment. So you’re in the kitchen, you cook, and you’ve got raw chicken all over your hands, and you have to do something. So you can just using your voice, tell it to do something. So you don’t need to put the chicken down to deal with whatever problem you’re trying to solve at that point in time. On top of mine answers, so this is one thing to remember. So the difference between voice and text does have a big impact depending on the use case. So for example, if I’m on a voice bot, and I asked you, hey, what’s your textphone number, there’s very rarely many people in the room would actually know that off the top of their head, but you could recite to me your mobile number or your address, things like that. So understand the different mediums so on voice, it’s very time-sensitive, whereas on text-based things you’ve got a bit more time I can sort of put that down, go get the details and punch it in. Gardeners trying this stuff out, they were just always liked, because it’s what we’re doing. But the gist of it is, but in 2021 more than 50% of enterprises will spend more per annum on bots and chatbot creation than traditional mobile app development. And the reason for that is actually becoming quite expensive to actually build out and maintain like, Apple keeps throwing out a new damn interface and updates and whatnot, you need to get on top of it, make sure it’s fresh, the UX designers sort of changed the way things are working. So you have to keep current with the new trends. Whereas the chatbot it’s just the content, there’s no interface, so I can build an Alexa skill. And all I need to do is make sure that you can talk to it. So that interface isn’t really going to change for quite a while. So from the development point of view, it’s nice and simplistic. It’s a lot cheaper to build and enhance these quickly.
Mark Chatterton: So the chatbot content, can we just get the AI to create the content gonna destroy my frequently asked questions, website at the AI and go dip, just learn it and then start telling our customers what to do. Well to the earlier point around the Harry Potter one, some of the weird and wonderful stuff that really come out of it. So the tall death era was wearing a shirt that said Hermoine has forgotten how to dance, so Hermoine dipped his face in mud. Right. And the scary part about this one is there was some humans just giving, when AI got stuck, it sort of showed them like two or three options, and the human had to pick the best one. So even though there was like still some corrections from humans in this, it’s still basically come up with this weird and wonderful story that made no very little sense. What we’re doing at the moment is manually creating that and the reason for that is a number of things is one that I really do suck at it, but also the experience you’re trying to craft. So if you think about the way that call centers are trying to die, like they get given scripts, and I get told that I had a great people and a lot of the stuff is very, very artificial. And you know why they’re kind of turning humans into robots. So we kind of go well, let’s turn the robots in the robots, but take that same thinking about the conversation content and the way you you’re constructing it and teach the robots how to do that. So we do that very simply by creating a persona. So what’s the conversational tone? Do you use emojis? Do you drop a GIF in there every now and then to spice things up? So how do you want to do it? Having a doctor be really funny and cheeky towards you probably isn’t going to be important in trust with you so being understanding of the right time for the right purpose and things like that, it’s very, very important. And we actually see that there’s going to be a kind of a new role, kind of a conversational design going forward. And the cool part about that is the roles are coming from weird and wonderful places. So while you know everyone’s looking at stem, and we need more people to sort of jump into that ecosystem, we actually see there’s going to be a role now for the more of the arts and the fluffy side of the education side of things. So some of our best conversational designers. A lawyer, he wrote some great stuff and your lawyer’s conversational design, but they used to writing very structured documents and explaining it to people is something that’s very common in an easy way. And another one was actually HR. So you know, human resources, they’re always trying to take these legal documents and get people to follow the process and the procedures within these organizations, giving managers tips on how to performance, manage people and things like that. So it’s weird and wonderful where their skill sets are coming from for that side of things. The other thing we’re finding is there’s sort of golden errors, as we say. So we had an insurance client, and he sells business insurance for some 1200 occupations. So he had a bunch of people coming in, and I started to ask do you sell Uber driver insurance? He didn’t sell it at that point in time, but he was getting like, a bunch of people a month asking that one question. He’s like, Well, obviously, this is a product that people want. So I’ll just tweak my existing cab, sort of policy to be Uber driver one, and then I’ll sell that. So if he wasn’t looking at those the errors in the insights, then he wouldn’t be getting that. So what we sort of say is, don’t give the errors to the devs to fix so they go yeah, sorry, we don’t sell that product. Okay. Give it to the marketing and the product teams to get the insights that have What their customers are saying to them because chatbots are very cool in the fact that they’re instantly available. They never going to judge the person for stupid questions. But it’s also important to remember that texts and screens aren’t going anywhere. But pitcher will always say 1000 words, NAB found that the perfect number of transactions is 20 displayed on a screen. But no one listened to 20 transactions over voice, it was just too many, no one got past three. So understanding the differences between the two is very important. Plus voice is not very private, if you’re in a public space, like I’m not going to go reschedule my colonoscopy appointment over voice on the trank, but I’ll type it in. So there still will be use cases for that. So for us, we see conversation is being our first and last UI. You know, 100,000 years ago, we began this journey as a species to speak to each other and have conversations. And now we’ve finally got the technology to be able to do it to the point where it understands us again, with that. Thank you very much.
Serpil Senelmis: I just love how chatbots have a low level of judgment compared to us humans. Thanks, Mark. And thanks, Susie. I’m so glad that the robots are our friends. Next week when you will startup isn’t a startup anymore. You’ll learn to take off your training wheels, scale-up, and propel your business to the next phase. Until then, I’m Serpil Senelmis from Written and Recorded and for WeTeachMe, this is the Masters Series.
About Masters Series by WeTeachMe
Masters Series is a show about inspiring entrepreneurs, creative thinkers, and visionary dreamers, and the stories behind how they built their companies.
Subscribe to podcast
Podcast brought to you by
Thank you to Jahzzar for the music.
Masters Series is presented by WeTeachMe.
The Masters Series podcast is produced by Written & Recorded.
The views expressed by the contributors on this podcast and linked websites are not necessarily endorsed by the publisher.
Question of the day
What was your favourite quote or lesson from this episode? Please let me know in the comments.