Host: Marco Oliver, Client Success Director at We Build Bots
Guest: Julian Harris, Head of Technology Research at CognitionX
Marco Oliver: Julian has a really extensive work history. He spent eight years at Google, he’s got experience in the government and he’s a really interesting guy to talk to. Let’s dive right in and hear what Julian had to say.
Hi Julian. Thank you for joining us on The Botcast today and being one of our first ever guests. I’m looking at your work history, which is in front of me and it’s mightily impressive I must say. Would you like to give the listeners a rundown on your journey to becoming the head of technology research at CognitionX?
Julian Harris: Yes, sure, I’ve been very lucky to have been creating internet software since the early ‘90s when the Internet really became consumerized. And so really I’ve become one of the first digital natives where I’ve been using Internet technologies for probably 25 years now. And in that time I’ve also spent about seven/eight years at Google, it was very interesting to be in a company at that stage where it was leading in so many areas.
I then spent some time in government where I was tasked with creating an emerging technology function for the Department for Work and Pensions which was really fulfilling work; working with another part of society and another target audience actually led me to create the UK Government’s first Alexa skill for pension age calculation - that’s open source, you can see it on GitHub.
This was interesting because their journey was coming through the early stages of conversational computing, and looking at the world of the Department for Work and Pensions, the beneficiaries of that department typically have struggles with digital inclusion; that is using computers. And so, when I saw the Echo one that came out in the UK in 2016, I thought well, what we have here is a hands free voice only computer that costs £50. And so, that’s pretty extraordinary and that could be a really new avenue for making available government services in a very effective way. In fact, what’s really exciting is HMRC has actually launched a skill recently too taking advantage of that insight.
So, that was a very interesting journey and being a part of emerging technology, I looked at the landscape in different areas and thought well, where is the next really interesting space? You know, there’s Blockchain, there’s VR and there’s some AI and other things, and I think that I’ve really put my finger on artificial intelligence and machine learning as an exciting area because it seemed to be another wave that was starting to disrupt pretty much everything. That’s exciting to be part of, and that’s what I like doing most; connecting people with technology in useful ways.
And so, to that extent, when I got invited by Charlie Muirhead, the co-founder of CognitionX to join them as a senior leader on the team, I definitely put my hand up because CognitionX is doing some very interesting things around helping people navigate the space, navigate the impact and disruption that AI is making, and help accelerate that in a way that’s also ethical and sound for people to do.
MO: Yes, that’s a really good point and I think there’s always an ethical dispute when you talk about AI to people. And I think some people believe that AI is just trying to replace jobs but that’s not necessarily the case. I think people who work in AI know that it’s about improving customer experience. It’s about making people’s lives just generally easier and I think there’s opportunity there for people who do repetitive tasks on a day-to-day basis. It will remove all the mundane aspects of their job and then that allows them to focus on and cross train on the more exciting aspects.
So, there’s several sides to it and we can cover that in more detail but let’s just dive into CognitionX. What is the business model for CognitionX? What’s the primary focus?
JH: Yes, well our tagline is ‘Accelerating the Adoption of AI Responsibly’ and so we have multiple different offerings that are all really trained to help people engage with the AI community. With understanding and making sense of the disruption and providing different ways in which people can adapt and harness the technology and the consequences of that.
We have a number of different things. First we have the work with the research programmes; I’m leading one on chatbots and the natural language processing. We also have HR and the impact of AI on HR; this actually launched first with my counterpart, Ian Bailey who was a senior executive at Cisco for many years and had some experience with us.
We have this workstream of research streams where we help people understand a space and keep them up to date with that space. And so, the chatbot primer that you’ve read, is a good example of understanding the landscape. So, the landscape says this is what the reality is, this is how it is in business terms, this is the impact and this is the future.
We have others that we are building as well. As part of our offering we’re also wanting to build a directory of resources for people keeping up to date. This directory is products, news and other resources - all things AI for each of the topics. So, it’s a place where people can go to to learn about everything related to the impact of HR and AI in terms of news and products.
We’re going to be doing that with chatbots too. There will be a place where we’ll have a curated source of information about chatbots, and from a business perspective largely. So, there’s a lot of technical resources, and we see there’s a big space where people are trying to say well, how do we make sense of this? What’s important? And so, we have several hundred newsfeeds that we read every day and we produce summaries of these and insights. That’s going to be launched sometime this year for chatbots.
So, that’s the help that’s going to be largely free, if you head to directory.cognitionx.com you can find that list of HR products there. It’s a catalogue of products rather than a comprehensive review source and that’s because we want to address the primary objective of scope, e.g what is out there? What are the things that are there? So, the categorisation and cataloguing is our first and foremost objective. There’s lots of resources out there to look at the exact capabilities of each individual one of these products and ways of helping that develop over time.
The first and foremost objective is, for instance, when we were commissioned by the Mayor of London to create a report on the AI census of London, we came up with 768 companies in London who provide AI capability and with that number we have them listed in the directory as well. So, if you’re wanting to see which companies in London do AI, then that is one of another facet that’s provided.
MO: Yes, that’s a large number of businesses just for London. I’m amazed by that actually. we’ll talk about the primer shortly but even in the primer you mention the Gartner Hype Cycle and we’re kind of at the pinnacle of that point at the moment I think - before things start to level off and the hype dies down a little bit. But even given that, that’s a large number of businesses just focusing on AI and chatbots. Is that something you were expecting to see?
JH: I think everyone’s first reaction is ‘wow there are a lot’. And, you know, the number of companies in London with that profile is more than Berlin and Paris combined, so it’s definitely putting London as the AI capital of Europe and that’s a deep insight that’s really helping a lot of people. I don’t think anyone was expecting that number. All the engagement of those 768 or thereabouts companies, we got well over half of those responding to a detailed questionnaire. So, that free report you can find here.
A report was created on the AI census of London and we came up with 768 companies in London who provide AI capability. This is more than Berlin and Paris combined putting London as the AI Capital of Europe.
That’s a free download and that was provided by the Mayor of London. That report has a lot of insight that can help anyone put forward a case for why they should be investing in London which is really exciting. And, you know, to us that’s great because that’s where we’re based, and it’s really nice to be helping foster the adoption of AI responsibly, and so, we feel like this really supports the overall objective we have.
MO: Yes, you’re primer was fascinating and we briefly discussed it before we had a chat and not only have I read the primer, I was lucky enough to watch the webinar from a beach in Italy where I was sunbathing, so that was very nice! There were some really fascinating points in there. Obviously at We Build Bots we’re focusing primarily on task bots and I hadn’t really considered the other side of that, that is that the social bots primary focus is keeping people from feeling lonely, which is amazing! One of the ones that you actually mention in there is the Microsoft Xiaoice, which I believe has a hundred million users and 25% of those users have said I love you at one point or other. That is unbelievable.
JH: It’s jaw dropping. The primer has a link to the original paper that Microsoft published on this. It was definitely one of the most interesting papers I’ve come across because what they’ve been doing is extraordinary and they’ve been doing it for some time. I mean, in chatbot years they’ve been doing it for four years. Obviously, we’ve had ELIZA from 1966 and we’ve had Mitsuku that’s been around for many years but this stuff that they did with Xiaoice, Rinna, Ruuh, Zo and the little blip for 16 hours with Tay, this is extraordinary and it’s created good feedback because I’d had a lot of people saying that the slide about task bots and social bots was probably the most interesting.
I think in terms of future, which is always exciting because there’s no real commitment to anything, it’s really fascinating because we’ve got the Alexa Skills Challenge too and you see what’s going on there. That’s been very interesting in terms of the objective of just being able to have a chat with Alexa instead of a task. So, basically another way of seeing the Alexa Skills Challenge is that it’s the Alexa Social Bot Challenge. Basically, it’s asking all the universities and other institutions that were invited, can you help us create technology that makes Alexa a social bot? And why is that important?
That was one of the light bulb moments for me, social bots are really the next operating systems that people are vying for. That’s what Microsoft, Google, Apple and Amazon are all doing, they want this to be the next Windows, you know, the place that you go to by default to do anything, to access any information, products or services. And this is something that I didn’t really appreciate but imagine that every one of those vendors has got that bullseye right in front of them. But it is an insanely difficult challenge because you need a chatbot, a genuine chatbot, that you can chat to so that you can say, for instance, ‘order Chinese for us after my last meeting today’. That’s actually a task but you can only really do that when you have a view of all of your points.
That was one of the light bulb moments for me, social bots are really the next operating systems that people are vying for. That’s what Microsoft, Google, Apple and Amazon are all doing, they want this to be the next Windows
MO: It’s becoming pretty integrated that point isn’t it because you’re pulling data from various different sources and having to link it all up together and then pass that information back via the chatbot?
JH: Yes, and that’s actually just the first level. That personal assistant – and I really like a description someone made, apologies I can’t remember who it was. They said if you want to think of the ultimate virtual assistant, think of a really good human assistant. That’s really what you’re aspiring to; to be a personal assistant. Part of being really good is two things, having enough information and enough context to be able to understand what is meant but equally enough intelligence to take initiative. So, if you think of a great assistant, I’d say ‘look this invite came through, it’s this really important pitch that you’re working on so I’ve rescheduled these other events and by the way, because it’s back to back, I’ve ordered some sushi to arrive at the meeting room when you come in.’ You know, this kind of stuff.
We have great assistants in our company and that’s what they do. They’re proactive and they have the context. But it’s immensely difficult and that’s why the first level of being able to connect services together in that way where the context is not reasonably available to typical services. For example, the calendaring system won’t know what your favourite food is nor will your favourite food supplier know your calendar, so that’s just level one.
Then when you go from there the next thing is, and the Microsoft paper illustrates some of this, when there’s a conversation in Japanese where that person says ‘I’m hungry’ and they just talk about what do they like to eat. And it says, ‘what about cookies?’, ‘Oh, yeah, they sound good.’, ‘What flavours are nice?’, ‘Oh, well, coconut’s good and I can give you a coupon’. It’s a conversation. It’s a chat not a task because they didn’t even know what they wanted. They said I’m hungry, but it was very vague. So, this is why the other component of the chatbot quality framework we use, I think was still holding water in it a little bit which is the ace framework, the ability of a chatbot to handle ambiguity, handle conversation and handle emotion as a way of conversation.
MO: Yes. So, I think we need to get that implemented ASAP because Saturday evenings in my household will be far easier if I can send my wife to a chatbot for a takeaway rather than spending hours deciding. That’s a really interesting use case actually. It’s a step, as we said earlier, in the right direction in terms of just making life easier with chatbots.
JH: I think you’re right, and I think that making things easier is fundamentally what is driving society development if you look at it. Growth and ultimately making things easier. So, there’s not really anything new in that except that this is a new frontier where there’s only so much more ease you can gain from using a desktop computer. We do need to move to this new frontier of being able to walk around and talk to a computer and explore ideas and have a conversation
This is a new frontier, we do need to move to this new frontier of being able to walk around and talk to a computer and explore ideas and have a conversation.
My ideal scenario is actually around travel. So, we eventually did go on holiday last week but the conversation that was involved in coming up with a place is complicated and the natural language conversation is really good when you do have that ambiguity, that uncertainty, and there’s a lack of clarity and you need to have an exchange to iterate and eventually hone down what you really want.
In the case of travel, you say, ‘I want to go on holiday’, and immediately I would like it to know that one of my kids is on school holidays so there’s a pretty clear defined period that it’s connected to. And that in fact, with young kids probably no more than a couple of hours flight, so prioritise on flight time. Temperature: you probably don’t want to go somewhere that’s over 35 and so there’s a bunch of context there but also I don’t really know and in fact I think a lot of websites could pretty much implement this today. Say if you were to say that we’d probably like to take two weekends but some time in the next six weeks. And it will come up with ideas from there, I don’t think any website currently does that well but they will start doing this. And when they have a voice conversation, the working memory of what’s going on, that’s one of the most difficult tasks with voice design because you have no guide rails by which you have a conversation.
The best way is for the system to deeply respect the most likely things the person’s able to remember at any one point. You and I can have a conversation that will actually have a very complex shared understanding or context but that’s because we know and have some expectations around, as human beings, what is a reasonable thing for the other person to remember and understand. This is the biggest design challenge that voice designers have today, is that you may start with the veneer of a voice conversation but the moment it stops relating to you as a human being, all bets are off and you suddenly don’t really know how to engage or what it can reasonably be expected to understand and to have memory on.
There’s a really good essay on this actually in a book that was published recently called ‘Ubiquitous Conversations’, and one of the essays is about this challenge of the personification and how that changes your relationship. You can find this book here.
The voice space is very interesting and there’s a lot of excitement in the VC world. Voice is hot and there are actually a lot of product companies who have dumped text and use voice because it’s hotter in some way. It’s also immensely difficult though.
MO: That’s really interesting what you’ve just said there. Voice is definitely the direction that things are going and I think with voice we’re a little bit off where it needs to be at the moment. It’s not quite as capable as the text element of it but voice gives for example, the elderly the accessibility to these kinds of services. They might not be as familiar with texting or being on social media or using messaging services, where millennials would use that service quite heavily. So, do you think there is room for both voice and text going forward or do you think it will maybe move towards the direction of only voice?
JH: So, voice is the natural interface for people. It’s the thing that kids use from even as young as 18 months so it’s the most natural thing and the best computer experience is one that marries our abilities and adapts to that and takes advantage and supports that. So, I think you’re right, voice is natural for when you want to communicate certain things, so, when you’re talking about an idea or you want to have a conversation around travel dates or winter, do something in the evening with friends. So, this is a very natural way for computers developing what’s going to make things easier for people in that way and make it more effective, so in theory freeing them up to have more time.
Voice is the natural interface for people. It’s the thing that kids use from even as young as 18 months so it’s the most natural thing and the best computer experience is one that marries our abilities and adapts to that and takes advantage and supports that.
There are other aspects of course. We can’t forget our eyes. Our eyes are actually pretty good things. They’re pretty good for absorbing vast quantities of information instantly. And so, I think the real sweet spot, which is a term kicking around around the future of chatbots and any emerging tech, is where you have voice as your starting point to initiate and when there is supporting information that’s best communicated back to people visually.
You’ve got Amazon Show and Amazon Look as two leaders in this space, or examples of this, and that’s going to continue. This mixed mode model will be par for the course and so people have their phones and other devices will eventually get connected more and more. I think that the ability for you to have a conversation with your phone and have the phone show visual images to support the answers and then be able to press buttons as a complimentary aid is going to be incredibly natural. There’s an interesting development around text, the number of people we’ve estimated who are sending instant messages around the world is around about a hundred billion a day.
There’s an interesting development around text, the number of people we’ve estimated who are sending instant messages around the world is around about a hundred billion a day.
MO: Wow. That is massive.
JH: That’s a lot and it’s not going away, it’s done by most of the people on the planet. There’s a number of reasons why this is attractive. One of them is that you can do it in private, you can send a text and you don’t need to worry about largely people understanding or hearing or discovering what you say. Text is a private channel. It’s also persistent, that you can have a visual conversation. Now, obviously you can talk and it can be transcribed, no problem but there’s an element of discretion and correspondingly even if you’re not worried about people hearing what you say, what you’re saying is naturally going to be distracting for other people.
MO: Yeah, that’s right. I think that’s why maybe chatbots won’t ever go completely voice operated. But it’s a really exciting prospect for how easy it could be for people just to get what they want, whether that be customer service or just placing orders. It’s very exciting. So again, coming back to the primer, you mentioned there was a bank in South Korea which I think is 80 to 90% automated now online via chatbots and they’re looking to get it to 100%. That’s quite interesting; a world where we have 100% chatbot automated services. What do you feel about that?
JH: There are chatbots today that are 100% already. You’ve got a lot of bots that there is no human escalation component, like Just Eat’s chatbot - it’s like a website. You need to bear in mind that when you go to Amazon, it’s 100% completely automated. When they have escalation, that’s where it gets really interesting. I think that the idea of 100% it’s being sold in different ways.
You’ve got a conventional bot with no human backing at all. You’ve also got systems where users become a fall back. If you look at Giffgaff in the UK, they’re proposition is that we’re cheap because our support is mostly from the community. They’ve created a service that is essentially zero maintenance. I’ve had Giffgaff for some years now and I just basically don’t have any problems. I had one issue and I went to the forums and I think I had one issue with another provider in another service where there was ultimately an escalation but it was extremely rare compared to something like a bank, like First Direct. They do an excellent service by having amazingly quick pickup times on a call when you call First Direct. They say hello within one/two seconds and their model is a personal touch and such like.
There’s always going to be that spectrum and so there’s going to be essentially commodity services where the business model is so well - or the operational model is such a well oiled machine that you just don’t really need people. All of the exceptions have been codified. It’s all just dealt with and a glimpse of how this could be completely 100% is in the Blockchain world of digitally autonomous organisations, DAOs they’re called, and this is where basically they say code is law and every single policy, every single business process is coded, it’s written. And so, that’s an interesting world where people want to create an entirely autonomous service that has no people involvement roles. So, they design the entire business. And the thing is, you can’t do that straightaway, you need to get through trying it out first with people and things.
So the 100% - it’s almost like you get this atom tope where, is there really any point in going 100%? And I’ll say in some situations that will absolutely be the case - that you can but most times people will find that getting around 80 to 90% is pretty good. You talked about the figure of 80 to 90% - Autodesk measurement in terms of the amount of customers they’re dealing with, they’re getting very good satisfaction rates too, so it’s not just that it’s automated but also they’re getting really good satisfaction levels. It’s pretty exciting and we talk about that a little bit in the primer. It’s a great example of how this can work well.
MO: Yes, so at We Build Bots, we essentially build chatbots which are capable of dealing with the natural language process and sentiment and it deals with 80 to 90% of the queries because of that. So, we’ve got a seamless handover which then goes to the customer service representatives, should they be required. The reason that you maybe heard a bit of uneasiness in my previous question was that I think there’s certain points. So, if, for example I’ve got a complaint or there’s been a bereavement in my family or something like that, I might not want to talk to a chatbot. I might actually want the personal touch. So, I think maybe there’s always this possibility of requiring a handover. Maybe we can’t get away from that, but maybe that just depends on the business that’s implementing the chatbot.
JH: I agree. I think that making things concrete, the technology, if we think about for the listeners who are not maybe familiar with exactly how this is happening today, the thing that’s different about conventional customer support and what chatbot technology is now allowing is for there to be a more empathic conversation. When it comes to escalations, for sure you need to have people around - that’s really important and that you agree that the degree in which you have those people is really going to be a function of your overall service objectives and how much you’re going to be a quality differentiator versus a price differentiator or something like that. That’s a business model decision but what’s really exciting and I think this is probably riding on something that you said, is that the natural language processing is getting better every day to help people be able to understand what their customers are saying through chatbot technology.
MO: That’s very true, yes.
JH: Yes. We’re a long way away from it. I mean, I think the stats that we picked up from our fairly informal survey was that typical natural language processing components of conversations today generally feel to be around about 20% or less. That is when you have a chatbot conversation, it’ll be mostly buttons and menus which is fine when that’s the right engagement but when you have buttons to choose from and you want to complain then you probably just want to have a conversation and that’s where people are best for now. That piece of being able to understand the nature of the complaint and emphasise and engage is something that people are actively working on and getting better so that even the initial engagement of a complaint can be triaged and dealt with by bots. I think to your point, I really relate to your concern of just being given a bot, the word I really heard when you said it was that it doesn’t feel respectful, perhaps.
MO: Yeah, that’s right. I think you’ve hit the nail on the head. I think with the developments in the natural language processing and then the sentiment detection, it’s going to be easier as time goes on to work out if somebody’s angry or distressed when they’re contacting the chatbot and then deal with that accordingly. I think maybe the really exciting crossover points might be the combination of the task bots and the social bots and maybe using elements of both to come up with something that’s quite universal.
JH: Yeah I agree and in fact, there are initiatives to do that. There’s a services company I momentarily forget, but their basic model is that they are for horizontal chit chat to any bot, so their whole proposition is if you just drop our service in then you can have this broader engagement conversation.
I think it’s an interesting idea, there’s also a trap - there’s always one to watch out for which is around expectations of what people can do. So, if I look at my son – my eldest is 7. I interviewed him actually on my LinkedIn profile, you can see it and it’s quite interesting because I basically said, how do you solve these problems? How do you find out what the time is? How do you spell a word? And different things like that, and, you know, it was, like, Alexa and then the question, Alexa, and then the question. The interesting thing was when I said to him well, how do you make sure you wake up early in the morning? And he said well, I set my alarm clock. And I said did you know that Alexa can set an alarm clock too? And he said ‘no, that’s great! I didn’t realise’. So, the problem is when you have an interface that’s in front of you, it communicates all the possibilities with the design. And so, you know the expectations around what it can do is clear, the utility and the scope but with a voice interface it’s unclear. And so, Alexa can probably do a thousand more things than we dreamed of and the biggest challenge with this stuff is the milestone prior to a very natural her-like interface where you just have a very natural flow from the movie Her. The challenge is to enhance that scope of possibilities and make the features discoverable in an easy way.
Now, as a user experience, as I am myself, you can’t expect everything to be effortless. There’s going to be training, that’s going to be part of it and studies have shown that people are pretty receptive to changing their engagement with voice computers when the voice computer actually guides them and corrects them and says did you mean this or did you mean that? People are smart enough to realise that next time just say that or do this and it’ll be faster. The reality is that between a kiosk which requires no training at all and a cockpit of a fighter pilot, voice interfaces will naturally (in the transitional period to the ultimate chatbot), there will be this part where the best interfaces, the winning interfaces, will have built-in guidance to help people know how to best engage with their voice. They will include discovery mechanisms as well, so that if you say ‘hey, I’ve done this thing’ or ‘I want to do this thing’, then they can say and by the way, next time you can also just say this and I’ll do that. That won’t 100% always work but I think that’s what’s going to be part of the success in future, is when people are connecting or there is discoverability as part of the journey.
MO: Yes, I 100% agree. That’s probably a really good point to end, if that’s okay with you, Julian?
MO: So, is your primer still available for people to access?
JH: Yes. Currently, it’s available complimentary for a limited time, so if you do want to access it then head here. So yes, for limited time we’re making it available for people to download and as part of the download you also get automatically invited to the first month experience of our chatbot research. So again, a summary is that the chatbot primer is showing you the landscape and the subscription is the way to keep you up to date and dig into it a lot more deeply. We haven’t specifically closed it off shortly but it’s likely that we’re going to be updating so that you’ll get maybe just a preview of that and then you will need to sign up to get the full primer and future editions - so get it while you can.
MO: Yes. And people definitely should, I would highly recommend it personally. Right, so, just to wrap up, we’re planning on asking every guest that we have on TheBotcast, if they had to live with only one application on their phone or on their computer and couldn’t have any others, which one couldn’t they live without?
JH: Let’s assume that I’m with all of the people that I want to talk to as part of this constraint and communication tools aren’t a problem. So, therefore I think the one app that I could not do without would be GarageBand. It’s a musical composition tool and it’s what I do in my spare time. It’s an amazing tool to create music in a very powerful way. So, it’s the most substantial improvement in my creative abilities of all the tools I have on my phone so that’s the one I’d go for. But obviously, if the other scenario is well actually, you can’t guarantee people around you all the time, then I’d probably have to say WhatsApp.
MO: I don’t think I would have guessed that actually. But thank you so much for joining us Julian. You’ve been a fantastic first guest for us to have on TheBotcast.
JH: Thank you. Thank you so much for your time.
MO: So, that was the interview with Julian Harris and as I said, there were some really interesting discussion points in there and it hopefully gives everybody a taster of the sort of things we’ll be going into. I think it’ll be really good when we can focus specifically on different industries and look at how AI is impacting those industries. Actually, on that topic, when this podcast goes live, we will also be going live with the second episode at the same time where Dr Zara Nanu, CEO of GapSquare will be joining us. Their primary focus is to reduce the gender pay gap so hopefully that should be really interesting to understand how AI is helping them do that.
Listen here and subscribe here to receive reminders when new episodes go live and gain access to all Botcast related resources. Lucy, our Digital Marketing Manager at WBB will also be sending out sneak previews of upcoming episodes.
If you’d like to send in any questions for future guests, then you can send those via our social media channels using #thebotcast, you can also contact us via the Anchor app where you can send in your actual voice and ask questions directly to the guests, and we’ll be putting those directly to them which will make for quite interesting listening.