Nick Hewer: Hello and welcome to this episode of Tomorrow with me Nick Hewer and Allianz. In this series we're exploring the global trends that will affect and shape businesses and whole industries in the years and decades to come. We want to discuss how such trends will influence businesses, identify the risks they bring and the opportunities they present. Today we're focusing on the growth of AI, a subject that excites many but worries many others especially in the business world. AI already touches many aspects of our lives from smart metering in our homes, automated driver assist in our vehicles and voice activation on smart phones. If some reports are to be believed the future could see robotic doctors carrying out diagnosis and treatment in hospitals, autonomous vehicles delivering shopping to your front door and your financial services being managed entirely by computer instead of a human. A report issued by PwC in 2017 suggested that AI could contribute up to $15.7 trillion to the global economy in 2030. That's more than the current output of China and India combined, so should businesses embrace this fast growing technology or be fearful about its capability to take all our jobs?
Joining me today to discuss this Adam Rates is the Head of Strategy and architecture at Allianz. Matt Hervey is the Director of Intellectual Property at law firm Gowling WLG and Paul Ryan who runs IBM Watson Artificial Intelligence for the UK and Ireland.
So let me turn first of all to Adam. Adam, AI, artificial intelligence banded around but exactly what is it and how's it currently being used?
Adam Rates: Well I think the important thing to remember is the term AI is a bit like the term vehicle, it covers a huge range of things, a huge set of technologies and capabilities from natural language processing to image recognition to neural networks to algorithm processing. So actually what we're talking about is a set of quite complex capabilities in the computer world, but the thing that makes it stand out is, in old style computing if you like, what you had to do is programme everything and think about what you might do and how it might work. So if you think of a postcode, think about the variabilities that you have in a postcode, if you write a piece of computer code to read a postcode then you have to think about all of those variabilities. For an AI you effectively train it on the outcome, so let's think of a picture processing, you want the AI to learn what a wheel looks like on a car, you show it lots and lots and lots and lots of pictures of wheels and eventually its probability of recognising a wheel goes up and up and up and up. If you're of a mathematical vent that's called vector geometry, so when you show it a picture it says I'm 99% sure that's a wheel, I'm 80% sure that's a wheel. What that means is two things, first of all you need a level of computing power and you need a level of data storage, so 10,000 pictures, 20,000 pictures, and the machine will get to about 80% of understanding. But the thing is that's quicker than thinking about how you would write the definition of a wheel for each of those photographs. So AI is about using lots of data and lots of processing power to think about what the outcome is that you want and training the machine to do that.
Matt Hervey: Got it, I mean as a lawyer I would make a stark distinction between what's called an expert system which has always been part of the sort of AI universe, where you sit down with an expert, let's say it's a lawyer or a doctor, you figure out their thought processes and you turn that into computer code 3.48 so it will follow their logic, but in the last five to seven years because of processing power when we talk about artificial intelligence, certainly in the press, we're really talking about machine learning which is what you've been describing where you take a training set and the computer itself figures out how to get from A to B.
Nick: Got it Matt, but it depends on big big big increasingly huge data in order to, you know, to get you to that, the wheel idea that you were proposing.
Adam: Yes that's correct. But the interesting thing is why this is so relevant now and if you go back at look at computing over the last 20 odd years, I went to an IT exhibition the other day and they gave me a one gig data stick for free, if I'd bought that in 1980 I wouldn't have got much change out of a US$1 million.
Yeah.
Adam: So that's just shows you how data storage has become more important and more relevant and how possible it is to do.
Matt: Oh, so what about the uses then? You gave us the example of the wheel, you know, how is it being used and increasingly what is the most popular?
Adam: So, from a customer end point, you, the sorts of things that we're starting to see are chatbots, so you phone up and have a conversation with, you know, like Siri or Cortana, those sorts of things, the other areas that you start to see are the sorts of things that we talked about a moment ago in terms of specialist systems, so can I recognise the damage on a car, can I recognise faces?
Yeah.
Adam: It's those sorts of areas that you're starting to see it being used in practice rather than in the theory.
Nick: Right so when I got through the passport control at Stansted for instance …
Adam: Those sort of things.
Nick: … I take my specs off and put my feet on that thing.
Adam: Yes.
Nick: That's AI?
Adam: Yes. At the easier end of the AI spectrum yes.
Alright. I think as a consumer another key point is just trying to navigate the vast world of content and information and so AI enables search, so I can now use Shazam to search for a song just by its sound but equally I can now point my camera if I use Blippar for example, at any make of car manufacture since 2000 and as long as it's going less than 50 miles an hour it will accurately recognise it for me.
Yeah.
Nick: Shall we just turn now to Paul, Paul Ryan who runs in the UK and Ireland IBM Watson, tell us a little bit about this extraordinary organisation, IBM Watson
Paul Ryan: So what we've built with Watson is the artificial intelligence platform for businesses and for professionals, this is a system that understands the world and interacts like humans do so it has senses, it understands language and it learns and experiences and that means it gets expression, it can see and it can read and it can hear. Watson senses emotion and sentiment which means it can shine a light on how we feel about things and that offers very very interesting and new ways to interact. Watson makes sense of vast amounts of unstructured data which is images, videos, language, sensor information as well as company data that lives behind the corporate firewall and it can turn all of that into insights that improve decision making to make people better at the work that they do and to provide competitive advantage. Watson represents cognitive computing and it came from the recognition that the rigidity of programmable systems, that's going to struggle to keep pace with the volume and the variety of data that's in the world today, and 80% of that data is very much invisible to computers because it's images, it's social media, it's books and videos. So Watson came from the need to make sense of all of this and the aspiration to have a technology that learns and that you train rather than programme …
Nick: Yeah.
Paul: … because that makes it available to more people to solve more problems.
Nick: And which sector do you think is moving fastest with IBM Watson? Healthcare I think is a huge area.
Paul: So today certainly Watson is huge in healthcare, it's also in financial services, insurance, travel, retail and governments so I really don't know of an industry it won't impact.
Nick: Alright, but healthcare certainly, and insurance indeed. Matt?
Matt: Well I was going to say that there are, sort of two classes of changes that AI will bring to insurance. One is the generic stuff that any business can benefit from so any business looks at its data and it's trying to recommend products to customers and you group customers according to their nearest neighbours and you say, well, this person wanted this product, this person is probably likely to want the same product, so when you're selling an insurance product you can identify the right time to contact a customer, proper parallel products and you sell more. But with insurance in particular the really interesting points are one touchless processing, so I understand insurance at the moment involves a lot of people looking at claims, assessing claims and if you can automate that you can save a lot of money and medical records is a classic example. If you can get AI to go through reams of medical reports to assess a claim or, or to identify the fingerprints of fraud which is estimated to be worth $40 billion a year you can save a lot of money.
Nick: Yeah, I get the healthcare and I read Paul's papers before coming on and healthcare is perfect, you know? And actually if I as a consumer, as a patient if you like, I'm informed that the diagnosis has been arrived at by a series of artificial intelligence systems, that's fine by me, because actually there was a trillion tonnes of it, alright, and it got to me in 30 seconds, that's great, and I believe it. But let's go back to the insurance industry shall we. Now then Adam, Adam how prepared for these changes is the industry? And will AI take so many jobs in the sector?
Adam: Ooh that's quite a complex question, so I think, I think we touched a moment ago on the value of the customer interaction so for customer insurance terms you need to think about brokers as well as end retail customers and the operational cost there's also an influence in the underwriting pricing and financial reserving space as well where processing large amounts of data very well makes an improvement there, so I think that's worth noting, it's not just in terms of the customer interaction. Will AI take jobs away? I think if you believe the papers then we're all going to be sitting in the Bahamas on deckchairs whilst the machines do all our work, I don't think that's necessarily true. I think when we were talking about machine learning earlier, what you still need is a level of expertise to interpret the outcome, so when you're talking about the machine giving you your medical diagnosis, probably what you want is the machine augmenting the information about the diagnosis, I'm not sure you'd want the machine to give you the diagnosis or to interact with you immediately after the diagnosis. That starts to impinge on the human world there and that level of interaction. So I think what you'll see is machines improving the information available and what you'll see is human beings involved in that process and I think you'll increasingly lean on expertise in that space. Will machines replace basic mechanistic processes, reconciliation, those sorts of things? Yes absolutely. So I think what you'll see is not swathes of the job being lost but the shape of the job market changing.
Nick: What about the skills gap? You sort of hinted at that.
Adam: Yeah, so I think that's an interesting challenge because I think, so if you take the sort of theoretical idea of underwriting, so you decide that a machine is going to do lots of underwriting and in five, ten years' time you have a machine doing underwriting, when we talked a bit earlier about machine learning and the expert input you need somebody to train the machine in what the outcomes are in good underwriting. If you have machines doing all of your underwriting where do your experts come from? And I think that the same applies to the medical profession, to the legal profession, where you've got machines going through case law, so I think you do have a challenge there. I think the other interesting challenge and you talked about it a moment ago, is in the trust space. How do we as individual normal people interact with machines and what do we understand about it.
Nick: There are an awful lot of brokers listening to this.
Adam: Yes.
Nick: What's going to happen to them?
Adam: I think that fits very nicely into the expert and the medical space. So a broker's role is to interact with their customer to understand the customer needs and to go away and find the appropriate set of covers and value and you know financial coverage for the problems that they have. The more capabilities and technologies they have to help them peruse the market, balance off, provide expert insight the better that relationship with the customer is going to be and also, you know, broker markets are about relationship with the customer and one thing machines are not particularly good at is relationships.
Nick: Now, Paul, can you come in on that?
Paul: I think to the question about jobs, the broker example is a really good one. When we think about IBM Watson we use the term AI to mean augmenting intelligence, so we see cognitive computing as a collaboration between man and machine, so in that sense with a broker, yes there is some work that can be automated but this is about being the best broker and giving that broker the tools that connect them and connect them to the customer so they understand the right offer at the right time and figure out the best way to interact based on [unclear 13:04.4] ancient history.
Matt: Well I mean, let's be a little bit more frightening, Excentia did a survey and found that 74% of consumers would be happy to get computer generated insurance advise, though all the areas of insurance, particularly car insurance, which are already commoditised to the extent that people are buying it through a website, they're not having any human interaction in order to choose their insurance, and there are disruptive entrants in the US who have investment in the tens of millions who are offering insurance cover within 90 seconds and a pay out within three minutes and with zero deductions.
Nick: Is this Lemonade you're talking?
Matt: It was indeed and they're saying you could claim for a pair of headphones or flip flops and all of that predicated on their being no human involved in settling those claims.
Nick: You're getting very animated there.
Adam: And I think you're absolutely correct but what you're primarily talking about there is the end customer retail market. When you start to move into commercial insurance, what in insurance land we would call mid corporate, where you're talking about may be multi site, multi location, multi vehicle, multi cover type options, then it becomes a little bit more complicated, so you know, motor insurance is relatively straightforward to write, there's only a, you know, there's a very few sets of criteria, you know how old is the driver, how old is the car, where do they live, where is it kept, those sort of things. When you talk about commercial insurance then you're into maybe thousands of factors that you need to take into account.
Nick: Also for instance, wearables, but wearables in health. What do you reckon about that?
I mean you can pretty much be sheathed in electronic gear that warn you about almost anything.
Nick: Yes so when we talk about AI there's a lot of overlap with smart technology generally because AI is about understanding data and telematics wearables is all about collecting vast amounts of data, and in the healthcare sector the idea is if you wear a Fitbit or other monitoring product on your body it will encourage general fitness and the hope is that will diminish chronic conditions such as heart disease and these chronic conditions account for maybe three quarters of the total healthcare expense so if you can make the public more healthy in this way you reduce healthcare costs but also you can identify sleep problems, depression and the like through these, through this equipment and also if you have a health claim and you've gone into hospital and an insurer is paying out for your treatment if you can then send a patient home by using a wearable, so it can monitor their heartrate etc., you can save on the cost of a hospital bed for a night which is about £400 a night and also you can have happier patients because they're allowed to go home.
For sure.
Adam: I think there's an important thing to talk about in terms of predictives as well, so we talked about telematics boxes and being able to predict a driver's future driving habits based on their past driving habits. In the health market this is an important thing a well, so we talked about sending a patient home, most of us are creatures of habit, we get up at sort of the same time, and actually if you have a predictive piece and you've sent patient X home and the device can then spot that they haven't made their cup of tea at the normal time or they haven't switched their lights off at the normal time there's something wrong, so there's that level of care that goes with that predictive ability as well.
Paul: And at Watson, Watson is being very successful at spotting things like depression and getting ahead of that because it can see changes in, very very subtle, in how people narrate themselves in the world so you can help people before something even becomes a problem so you've got the empirical that's very visible because it's about data, it's about a visit to the doctor, but you've also got the softer kind of longer term interaction.
Adam: And I think the final part of that is effectively the behavioural change so we, you know, a Fitbit or other similar device and a telematics box actually encourages a proportion of the people using them to change their behaviour, you know telematics are proven to improve driving skills and reduce accident levels. Fitbits do have an impact on people's health so, there's an element of sort of positivity that comes from that as well.
Nick: Absolutely, absolutely. One's very own blood pressure monitor of course, that should be an important part of every household.
Now come on what about cost, nobody's talked about cost. What are the cost implications of using AI in this area?
Adam: I think there's an interesting point, I mean, there's a point here where AI is no different to anything else in the sense that you know, I work for Allianz, it's a large organisation, we have a budget for changing things, whatever we do to change things has to pay back, so artificial intelligence or particular functions in that space, so to know what benefit is it going to drive, how much is it going to cost, does that pay back happen in the right sort of time period to make it justifiable to, you know, my boss and his boss and so on.
Nick: Matt, you're a lawyer and going to law is an expensive old business, going to, you know, litigation is an expensive businesses.
Very high value.
[laughter]
Nick: Are you guys getting involved in AI?
Matt: Yeah absolutely, and well, to give you a very dramatic example JP Morgan Chase introduced the system to review commercial loan contracts and they claim that work that used to take loan officers 360,000 hours can be done in a few seconds, which sounds like a lot of savings and certainly a lot of the more menial work in the law is looking to be automated so things like identifying problematic clauses in a contract automatically and then having a more senior lawyer be able to look at it efficiently. But what I will say is I'm a litigator and we have had what's called disclosure review tools which use AI for decades now to help you go through a mountain of documents and find out what's relevant and it just means you become more ambitious of what you can do with the budget you had and it hasn't actually changed the cost of litigation.
Adam: I think there's a part here where, you know, we're talking about a particular technology but actually we're talking about technology change, that happens all the time, you know, you wouldn't do adding up without a calculator now but, you know, if you go back, you know, a few years you wouldn't dream of using a calculator, now calculators you know are common place, you just have to think about how you do it differently. You go to your favourite supermarket, you can go to an automatic checkout, you don't have to speak to a person, it's just a level of technology change and you just, you look at how you're going to absorb it and derive benefit from it and you use those areas that you're [unclear 19:16.6].
Nick: So that sort of begs the question who can afford to ignore AI then? In business, and I guess there's nobody, but anyway.
Adam: I think if you're running a business then, you know, your role in the business is to look at where you can improve your costs, where you can improve your customer experience, where you can improve your penetration of the market so, you can't ignore this just as you can't ignore a whole range of other things.
Paul: There was a study by MIT that said that 85% of executives believe that AI will allow their companies to obtain or sustain a competitive advantage, so I think it will be the minority that choose not to implement or build something around artificial intelligence.
Nick: If you want to be competitive you've got to stick with it. What are the risks then? Who's going to talk about the risks that come with AI?
Adam: I think it's the lawyer's job isn't it to talk about risk?
[laughter]
Matt: It's a huge list I'm afraid, so on a practical level it's the issue of trust, so on the one side chatbots putting people off and on the other side excessive trust. So we've had people literally die because they've followed their satnavs slavishly because they believe the computer too readily. Then we've got the issue that AI particularly machine learning it's an algorithm generated by a computer which is just a pile of numbers, so if you want to roll back the clock and figure out what went wrong, no one can really unpick it, you can't audit what happened which means from a legal point of view unpicking who's liable can be very difficult. Then you can have a related issue, dramatic failure such as flash crashes and the stock market and that's when a single AI is remorselessly following its programming or when you have an unforeseen interaction between two AIs and then I would also add cyber security, AI is now controlling or will control maybe autonomous vehicles that may control aspects of our important infrastructure and you've got a risk of really catastrophic failures which the insurance market may have to pay for, and then on the social side you've got the power of the incumbents if data is the new oil and Google has so much data how does anyone actually compete with them?
Nick: Are you coming in on this Adam?
Adam: Yeah I think one of the challenges here is I think all those points are very valid and very valid indeed, I think one of the challenges here is this is a very fast moving space and the law notoriously isn't that quick so I think there'll always be a bit of a lag so I think, I think one of the things we talked about a bit earlier was trust and I don't think it's just about protecting the data, it is absolutely about protecting the data but one of the things we need to think about is what are we doing with this that is to the benefit of the customer? And if we're doing things that are not to the benefit of the customer it's not the right thing to be doing and I think that trust, that ethical code, that sort of moral balance is really important.
Nick: That brings us really into the impact on the consumer. Who's going to take up the challenge of explaining or telling us about the public's perception of AI? Who's got a steer on that?
Adam: I think, sorry, I think it's going to be quite interesting, I think you know the public tend to, so a generic term, I think as we become more familiar with technology they become part of every day life. We talked a bit earlier about satnavs leading to some unfortunate consequences in their early days, now nobody goes anywhere in their car without switching a satnav on even if it's to the corner shop and I think we're unfamiliar with some of this technology, we will be uncomfortable around it but as we become more familiar interacting with chatbots, as we become more familiar with being told that this decision was made or informed by an AI we will become more inured to it and more relaxed about it over time, it's just an inevitability and that's part of human nature.
Nick: And Adam you were talking about wow, you know, let AI deal with the diagnosis but the doctors can then tell you, I think maybe for a while that's true, but after a while, you know, we're very happy to hear directly from the chatbot or whatever it's called.
Adam: I think you might be right, you might want to hear it from the chatbot, you might want to hear it from the chatbot, you wouldn't want necessarily for the chatbot to be your only interaction. I think you would want, you know, if you have a serious diagnosis you're going to want another human being to interact and relate to, you don't just want a chatbot to talk to.
Yeah, things vary obviously.
Adam: We would see Watson as a tool to support the decision process that the clinician goes through, so helping them access the thousands of reports, helping them find the individual that's most like the patient, but still very much the clinician making the decision and making the diagnosis just based on a tool that helped them get to a better set of evidence.
Nick: I think Matt was talking earlier on about the use of AI in the legal profession and I think you hinted at a point that perhaps the costs wouldn't change?
Matt: The charges …
Nick: What about insurance? Is that true? The way we buy insurance will change, will the cost change as a result?
Adam: I think with all these things you're talking about ways of getting more efficient and staying competitive, so everything an insurance company does changes the cost, we're trying to drive down costs all the time. I think if you look at the cost of motor insurance over time, if you look at the cost of commercial insurance over time, I think you see that, you see that efficiency coming through.
Nick: The dream is surely that with telematics maybe all objects being smart, if something gets lost or stolen the claim will just be processed.
Adam: Yes.
Nick: You will not even ever contact your insurer, you'll just get the money.
Adam: And you can start to see that in things like travel insurance where if you have a flight delayed there are some insurers already that will automatically pay out your flight delay because it's receiving data from the airline, it knows it's been delayed, it's, you've hit a particular point in the contract, it pays automatically.
Paul: So I think this is about meeting your customer where she wants to interact, if she wants to pick up the phone, or call into a branch or interact through a conversational natural language system, and that mirrors the way the way that we like to use our phone, often we spend more time jabbing messages into our smart phones and the problem of making a high quality channel available that would give your customers the same experience that they would get through one of your skilled contact centre staff is not a trivial one to solve.
Nick: Adam, earlier on you said we're talking here about business sorry, but if there is no benefit to the consumer then we're doing the wrong thing.
Adam: Yeah.
Nick: Surely we're offering people in business speed and cost saving giving them a competitive edge. Now where is the consumer, the end user benefit in all of this?
Adam: Well I think Paul made a very good point, it's about offering the consumer what it is they want and if they want to pick up the phone and talk to a person that's one thing, if they want a quick query answered by a chatbot, so I had to phone my bank the other day to find out how to do something, it was a very straightforward enquiry, it took me a while to get through, the person was able to answer it in about 30 seconds, if there was a chatbot interaction available I could have just asked the machine. So it's about how I would like to deal with that organisation or that service and the most efficient way for the organisation to answer and respond and service that enquiry. So it is, that is really important because effectively we're in the market of selling things to customers. If they don't like it and don't enjoy the experience they're not going to buy anything.
Nick: It's an interesting point, I mean there's a lot of focus on chatbots in particular and to give you a sense of how important people feel it is Amazon's Alexa isn't very conversational at the moment and can only handle basic queries and part of their prize is a $1 million reward if you can come up with an AI which can converse coherently and engagingly with humans on popular topics for 20 minutes and no one's anywhere near winning that. But I actually thing what we've said today is, I think chatbots may in fact be the wrong direction, people don't want necessarily to speak to their insurers ever, they want it all to be frictionless, they want to do it through a text message, they want to be able to track progress automatically.
Paul: So in that sense it's a, we think of this as digital messaging rather than a chatbot and it's meeting the client down whatever channel they want, if that's a text message, if that's email, and that's an important decision to make sure you choose an AI that's not just providing you a point chatbot.
Adam: You're dead right about the interactions, I think there was a statin the other day that said something like 20% of conversations with Siri and Alexa involve swearwords, so the idea that, you know, you're having a positive interaction there I think we're a bit away from that.
Nick: And Matt?
Matt: One of the great commentators on AI, Pedro Domingos, actually predicts that we will have our own individual AIs, our own virtual people who will negotiate with the insurer's AI or the employer's AI and it will settle the parameters for the discussion and then you as a person will only be involved once your AI has a deal you like.
Can I have one to deal with HMRC?
Adam: That would be awesome wouldn't it?
[laughter] brilliant.
Nick: Alright, just one final thing, I am a lot older than you guys and I'm not in the business and when I got involved and was asked to sort of have a look at all this, I felt it was a dismal sort of area, I slightly worried, I suppose I read the Eagle when I was a kid, I was slightly frightened. Now am I being ridiculous? Come on, Adam?
Adam: I think it'd be rude to say yes, but yes.
Yes.
Adam: I think there's, I think there's two answers there, I think I have an 11 year old, you look at how easily she adapts to technology, I don't think she has the same fears that, you know, the generation of my mother has. At the same time you think if autonomous cars are available now how much easier her life would be.
Nick: What about Matt?
Matt: Well I do think that the promise of frictionless insurance and frictionless business interactions generally is very, very appealing but we do have to make sure that the efficiency with which I can measured up for medical insurance doesn't create a society of haves and have nots, people who just cannot be insured. There's also the issue of the singularity when the machines turn on us of course which I think is fantasy and also the looming threat of mass unemployment which is leading to another form of insurance, apocalypse insurance where the hyper rich are buying self sufficient resorts where they can have their armed guards keeping out the unemployed masses.
Nick: Yeah we haven't talked about the military but we'll pass on for a second there ….
Adam: [unclear 29:26.7] phallacy there isn't there?
Paul: The jobs that the graduates who work for me in Watson do today are far far more interesting than the job that I had in information technology when I graduated in 2000 so I think the job landscape gets much much more interesting and will continue to find new and interesting things for people to do that they'll enjoy.
Adam: I think there's a, there's a nice IBM stat isn't there that says something like 90% of the jobs that will exist in ten years' time don't exist now.
Nick: Absolutely, absolutely.
Adam: I can't remember the exact numbers but it's a nice idea.
Nick: That brings this episode to an end but please do subscribe to the series through your podcast app, that way you'll be sure of never missing an episode and we'd really appreciate you leaving a review too. We'll be back to explore another major global trend in the next episode of Tomorrow. In the meantime, from me, Nick Hewer, it's goodbye.