# Aegis Studio Transcription

> Generated entirely on-device via Aegis V2.0 Core for absolute privacy.
> Applied Lexicon Domain: **LAW**

---

### [00:00 - 02:00]

Good morning ladies and gentlemen it's great to see so many of you here and so many old friends as well. We have a lot to do over the next hour or so in terms of me speaking at you so im going to speak for about an hour or so and then hopefully there will be an opportunity for some questions and answers and what I thought it would take you is here im going to talk a little about the future that Daniel and I see for the professions, and provide some evidence that we gathered in our preparation for the book and trying to still this down into a model of we regard as the evolution of professional

service. This you will see rapidly is underpinned by technology and well devote a little bit of time to expressing our views and how technology is best understood which will lead me on to the heart of this conference, artificial intelligence. That will take me as one will see from the line of argument the question of the future of jobs and finally I will close by considering how it is that we share expertise in society.

So lets start with future in fact start with two futures because the more Daniel and I solve the professions as we explore and investigate it we recognize there were two futures unfolding and the first we say is reassuringly familiar. This is where technology comes along and automates our traditional ways of working. It streamlines and optimizes the profession. So it's not fundamental change here im broadly speaking we find professionals comfortable with this idea that the role of technology is to automate pre-existing practices.

But in our travels and I know everyone in this room who is simply there with this we also saw a second future that we believe is unfolding in parallel and this is a future where many of the problems to which historically only professionals could provide the answers. Many of these problems are now being solved in entirely new ways, innovative ways through technology. So technology is coming along

### [02:00 - 04:00]

in this second role, not simply to automate what is going on for years, but to innovate, to give rise to new ways, essentially, of delivering professional service. And what we see is that these two futures will run along and parallel for a while, but in the long run, we believe the second will dominate. And in due course, the traditional professions, as we know them, will come to be dismantled. And this is not a prediction over the next few years, this is over the next few decades. And it might be a problem.

It sounds horrific, and many professionals say what a gloomy message you have, but I want to see the outset that our interest has always been in the recipients of service. While we have sympathy with those who need to adapt, those who currently provide, our overwhelming interest is impatience in clients, in students, those who receive service. And it seems to us that the second future provides an opportunity to make the expertise of professionals far more widely available than was possible, indeed was conceivable in the past.

We also thought that we should take a step back and ask a more fundamental question, why do we have the professions at all? And we spent the first chapter of our book looking at this. What basic problem do our professions exist to solve? And then what we thought about it, we recognize in some extent we take this from Herbert Hart, the great legal philosopher and the concept of law, who talks about true sums of a human nature. And one of them, he says, is that human beings have limited understanding. It's unarguable. We all have limited understanding.

And so when were faced with problems in our world beyond our knowledge and experience, serious problems, we tend in the print-based industrial society to turn to these people called professionals when we have a problem with our health in relation to law, in relation to our buildings and so forth, our tax affairs. We turn to these people we call professionals.

### [04:00 - 06:00]

And they have what we in the book define as practical expertise and thats a mix of knowledge and expertise and experience but also know-how and skills. The whole cluster of attributes that we find in professionals, we call this practical expertise. And the only way in the print-based industrial society that we as non-experts can gain access to this practical expertise is through these professionals. And they're operating under this grand bargain.

And this is essentially the monopoly, the exclusivity that we grant professionals in many areas and certainly a lot as people who are the only ones who are allowed to conduct particular tasks, whether it be appearing in a courtroom or conducting an operation. So these professionals we argue are gatekeepers. They're the custodians, the curators of knowledge and expertise that they dispense to make available to people who need this. And this is the solution of the print-based industrial society.

But were no longer in a print-based industrial society were in a technology-based internet society and if were really honest around the world, our professions are creaking, our health service, our legal services, our educational services. Even in advanced economies there are concerns, complaints, worries. We face a crisis, it seems to us, in how it is that our professions deliver their services. For most people, for the most part, professional services are unaffordable. We know this is a law in particular. Many of the practices are antiquated. They're opaque.

Sometimes this is intentional obfuscation, but sometimes it's the jargon, the legal we all use to communicate with one another is impenetrable to the non-experts. And also the professions underperform in a very specific way were not saying the professions do their job badly. But what we are saying is that the expertise of the very best can, because of the

### [06:00 - 08:00]

way they make this expertise available, only be made available to a very few. And so our professions are unaffordable, antiquated, or pecan underperforming. We think this is a fairly severe indictment. So the question we ask in the book is this, can we solve the problem differently? Can we find a new way of offering access to this practical expertise that doesn't involve human beings acting as gatekeepers? Do we really need the gatekeepers any longer? And our argument, of course, is that through the emergence of a variety of technologies, were finding new techniques, new ways of making

practical expertise available. And indeed, at the end of the presentation, I'll suggest to you that there's six different ways we can make expertise available. But lets look at the evidence. And this is what largely done by Daniel. It seemed to ask that it would be useful to speak to thought leaders, to market leaders, right across various professions, to get a sense of what was happening at what we call the vanguard. So in education, we found at Harvard that more people signed on in one year to their online courses than had graduated from the university in its 377 years of existence.

Then with the taken Academy, you may be familiar with this, online and instructional videos in mathematics. Daniel teaches maths in economics at Oxford. He points people towards Kann Academy. These videos own for 10 million unique visitors each month thats more than the effective school population in England and Wales. These are first-rate, traditional videos that can help people both with easy and hard problems in mathematics. In medicine, Web MD, a network of health website, more than 190 million people every month visit these, they're not all hypercontacts.

### [10:00 - 12:00]

Others affirm architects in Holland are doing precisely that. Using d printing to print and assemble parts that are put together that will constitute a building. Printing parts that are around a meters in length. In consultancy, look at adventure, with regard to them, I suppose, traditionally, as a firm of management consultants. They employ 750 hospital nurses. The boundaries between professionals are breaking down, or take delight, founded 170 years ago as an audit firm. They now have over 200,000 people working within that firm.

They have their own university, for goodness sake, 700,000 square feet in Texas. And even when we looked at the clergy, we looked at divinity, we found people doing strange things online. This is Second Life where people have their avatars. It's a gloomy looking, Anglican Cathedral. But every day, and there's a large attendance, they have a global, they have daily worship classes, they have weekly Bible discussions, they have counseling as well. But our all time favorite is an app called Confession.

And the Vatican, believe it or not, in 2011, granted its first digital imprimatur, as the license thats given to official publications in Catholic religion. This is remarkable. This is good tools for tracking sin, and it has drop down menus, options for contrition. So we regarded this as remarkable, and we've just given you there a flavor of whats going on.

So in the profession we looked at, we saw sometimes in using, sometimes amazing, but certainly alerting instances, not simply of automation, thats to say of systems coming along and computerizing our traditional ways of working systems that were so...

### [12:00 - 14:00]

the problems, providing the service that historically had to involve human professionals and gatekeepers, we were saying that systems were taking on a different role. Now, a lot of our book is trying to make sense of whats going on, and one of the models, the structures we use is this, and it's a development from something that I've done in previous work in my book, The End of Lore's Question Mark and tomorrow's Lore, so people might see some resemblance there.

But what we argue is that the professions are evolving from the traditional model of some kind of craft, while the professional services delivered as a craft, what I call my other one, a bespoke service, a highly tailored, customized service, an assumption that each particular case requires a fresh canvas. You start with a blank sheet of paper. This is a notion of professional service and law, for example, is very much inculcated into us in law school, where one gets the sense that every case could reach the Supreme Court.

Or when you read fiction, people are running around looking for loopholes or for smoking guns, but actually in daily life, in law and elsewhere, we find in the professions a high degree of standardization, consultants use methodologies, doctors use protocols and checklists, lawyers use precedent templates, standard form documents, teachers use last years notes. It's not the case that we start from scratch and we craft every bit of service on this blank sheet metaphorically speaking.

We go further than standardization, though, we systematize, we use within our institutions, our schools, our universities, our hospitals, our professional firms. We use technology sometimes to automate checklists and we have work flow systems, and in law, for example, even automatically to generate documents. Now this is my video.

### [14:00 - 16:00]

It's all within the professions, and at this stage in evolution, we reach a significant line. We call this the line of externalization, where the content, the systems that have been used internally, are then made available across the internet to users, to recipients of service. Significantly, this can happen in three different ways. It can happen when, for example, Allen and Overrail, one of the world's leading law firms, charges for its online legal services. 15 million pounds a year, they generate in revenue through that.

But thats one model where the content you make available is as an online chargeable service. And many professional firms are rushing into this box. If they cant sell their time, then they should surely encapsulate their expertise and make it available online. Charities and governments, however, tend to be moving in a different direction. They're making guidance and content and advice, materials available online, on a non-chargeable basis. They're still controlling the content, but the service is not one for which users pay.

And the final model, which is alien to many professionals, but for us as deep attractions, is the idea of the content being made available on a common spaces. That is to say, it's not owned or controlled by major professional providers, by gatekeepers, by governments, in the spirit of open source software, in the spirit of wikipedia. This is content that we all can contribute to, can draw from, and in a sense, we create and look after it ourselves. And these are the three different models.

And when ones thinking ahead, one of the big strategic questions is how it is that practical expertise in different fields will be made available online. And it's often hard for those who are lawyers to think

### [16:00 - 18:00]

objectively about this. But I think we can often think more clearly when we think of other professions. In medicine, for example, when you think of the billions of people around the world who would benefit from ready availability of medical expertise, do we really want to evolve, why we still have some control, into a society or into a community or into a global network, where that expertise can only be made available in the world. And we cant be available on a chargeable basis.

Or when it's rather qualmsily controlled by the state, should we not be thinking in terms of medicine, of building up this commons of expertise, and the same arguments apply across all professions. So we've got a lot of information from left to right, from craft to externalization, the commoditization of professional service. But as you can imagine, much of this is underpinned by technology is at the heart of this future, these futures that we identify. And I want to talk a little bit about technology to you today.

And to some extent, I know many of you are familiar, intimately familiar, with technology. I share with you the way that we try to express the developments we see in technology. Because I speak to you, I suspect us, converts. But I also speak to you as ambassadors. One of the things you need to do is go back to your universities, your firms, your state bars, as the case may be. And if you're convinced, you need to plead the case. And I just want to share with you the way that we plead the case.

However, it is that we unfold the story about technology in a way that we hope is compelling and interesting. And I started by going back to 1996 when I wrote a book called The Future of Law. And bizarre, this may seem in retrospect. When I wrote that book, one of my main preoccupations

### [18:00 - 20:00]

of electronic communication was email. And I was getting lectures all around the world and so forth. And one of the running themes was of mine was that the dominant way that lawyers and clients would come to communicate in the future would be by email. Now this, of course, seems entirely unobjectionable, indeed, but now all today. But at the time, I joke not. The Law Society of England and Wales said I shouldn't be allowed to speak in public. They said I didn't understand confidentiality.

They argued that I was bringing the legal profession into disrepute by suggesting that lawyers and clients would use email together in the future. And that gives you a flavor of where we were just 20 years ago. That wasn't outlandish thinking. That was fairly mainstream. And today we believe that we can understand technology by considering four different phenomena. The first is the exponential explosive growth in the underpinning technologies. Secondly, and to some extent this is a defining phrase of our work. And I suppose it maps quite well onto the notion of AI.

But what we argue is our machines are becoming increasingly capable. We also say that machines are becoming increasingly pervasive. And finally, as human beings, were becoming increasingly connected. And I want to run you through each of these and tell you the story that we associate under each heading. And the law one needs to start with here is not a law of the land. It's Moore's law. Gordon Moore, the man who 5g years ago, in 1965, predicted very approximately that every 80 months or two years that changed slightly over time, processing power would double.

And this gives rise to what mathematicians and others call an explosive and exponential growth in processing power. So it starts off quite shallowly when it doubles every month and then it explodes. And we are...

### [20:00 - 22:00]

to use a phrase of "Red Kurzweil" in my view, the knee of the curve just now. We think we've come a long way in terms of processing power, but actually were just warming up whats hard I think as human beings is to grasp the effects of exponential growth. So let me give you an example. Imagine a piece of paper a, just a.06 millimeters in thickness. Imagine you fold it over once, you fold the result over twice, again four times. After four folds it's the thickness of a credit card. After a living fold, if you keep doubling, it's the height of a coat cap.

After q1 folds it's the height of big bend. After q3 folds it's distant from here to the moon. And after a hundred folds it's eight billion light years in thickness. This is quite outlandish isn't it? What on earth is going on here? It's doubling a bit of paper over a hundred times, but this is whats happening to processing power. And this leads a number of commentators to observe that by 2020 the average desktop machine will have the same processing power as the human brain thats about 10 to the 16th, 10 to 17th calculations per second, which I think we would all say is remarkable.

But as not as remarkable as this, by 2050 if processing power continues to double every year, and there's some controversy here, but most material scientists and computer scientists say is we move away from silicon into other ways of other techniques for processing, particularly in quantum community. It's entirely a reasonable prediction that doubling will continue. By 2050 if the doubling continues, the average desktop machine will have more processing power than all of humanity put together. This is this unbelievable doubling every couple of years. He must be exaggerating.

Well this in 2001 was the prize lecture, the Nobel Prize lecture of Michael Spence, and he said then that roughly there had been a 10 billion times reduction in the cost of processing power in the first 5g years of the computer age. And remember he was giving

### [22:00 - 24:00]

that lecture at the beginning of 2000. So every couple of years I'll have gone 20 billion, 160 billion and so on. It's not just processing power though. Think about data. According to Eric Schmidt, Google's Chairman, every two days now we create as much information as we did from the dawn of civilization up until 2003. If you project forward by about 2021, that will be every hour or so thats the amount of data were creating.

On a more prosaic level for the those of you who fight hard as all of us at Sesame due to grass, the enormity of this, just reflect on a good memory card in your camera in 2005, probably 128 megabytes. In 2014 a good card, 128 gigabytes thats more than a thousand fold increase in less than 10 years thats more than a doubling of a year. This is remarkable whats happening. You can store all the world's recorded music in a little box you can fit in your hand. That just gives us the sense of whats happening.

And then we can say the same of bandwidth, we can say the same of most technical measures that were seeing this explosive growth. And this is fueling enabling what we call this increasing capability of machines. And we look at that currently but this is going to change over time under four headings. The idea that machines using big data techniques can actually begin to outperform human beings in a variety of areas. And when machines can solve problems, the machines, this is a field called effective computing, can both express and detect human emotions. And finally the whole area of robotics.

So we think these are the four areas in which were currently seeing machines becoming increasingly capable. And if you look at big data, this is the idea that data can yield insights, patterns, correlations that were creating as users of the Internet of Technology, a data exhaust, huge bodies of information that actually encapsulate capital.

### [24:00 - 26:00]

And from that data and information, we can make very valuable predictions. You'll have heard of the system called Lexmic. You know which is bought recently by lexus Nexus. That helps predict decisions of the patent court. You'll be hearing later from Dan kate, who's working in Supreme Court prediction is fascinating too. The underlying point in these and other projects is fascinating.

It's that to predict the decisions of the court, one can, as we've traditionally, as we've seen in the past, they have done, engage in legal reasoning, look at the substantive issues and see where we think of the needness. But we can also make statistical predictions about the behavior of the court. We can look at phenomena such as the judges involved, the court involved, the volume of the claim. You may know the research that was done in Israel and parole hearings, which showed the difference between hearings before lunch and after lunch.

A whole bundle of criteria that we can bring to bear in thinking about court's behavior well beyond the substantive law. And this fascinates me because the question that all clients ask when they have a dispute is whats our chances of winning. Now, if you can come to a more accurate answer to that question, not by reasoning in law, but by undertaking some kind of statistical prediction, you won't find that she's executive having a problem with that. Or let me put it another way.

If you've got a great case in England, a leading commercial barrister will never say that the chances of winning are greater than 70%. Never. So whats happening with the rest of the 30%. Why is it you cant be more confident given the merits of the claim? And the barrister will say, well, there's many other factors involved, for example, who the judge is And I would say precisely, they are the criteria that Dan and others are looking at when they're engaging in the prediction of judicial decisions. And so what were seeing, and this is just one example.

### [26:00 - 28:00]

is that from the data that we created the byproduct of users of systems, we actually can identify correlations, patterns. We can yield insights that sometimes can even outperform us as human beings. And then the eponymous Watson. One of the reasons were gathered here today, this computer system that remarkably in 2011, on your TV quiz show, "Jeopardy! We dont Have It in the United Kingdom", beat the two best of our human-Japanese champions.

When you think of the technologies there, the speech recognition, the forms of natural language processing, the knowledge, the storage, the manipulation, the inference, the reasoning mechanisms, the speech synthesis, apparently even in an earlier model on mechanical arm that could press a button. But when you think of whats going on here, this is definitely a move forward in the whole field of artificial intelligence.

A system, ladies and gentlemen, that frankly, in its own bizarre way, given the nature of the program, "Jeopardy!" can answer questions about anything in the world more accurately and more rapidly than any human being. Now, as is often said, IBM did not invest in Watson because they wanted to be good at quiz shows. And the early signs of there were particularly in medicine where I think they've invested most heavily.

In terms of the quality of treatment plan, the quality of diagnosis, you can see how these systems, in the not too distant future, will outperform not just individual human beings, but teams of human beings. And we also know that it's early days yet, but Watson's moving beyond jeopardy, moving beyond medicine into areas of law as well. So thats a little bit about data, and thats a little bit about problem solving. A third element, what I call, I dont call it, is the effect of computing. I suspect this is an area that most of you will be less familiar with.

This idea that computers can both detect and express human emotions, that a computer could look at your face and tell you,

### [28:00 - 30:00]

whether or not you're happy, surprised, angry, or disgusted. Indeed, Apple just bought a company that specializes in precisely this idea that there are systems now that more accurately than any human being can look at a smile on a human face and tell whether or not that smile is a genuine smile or a fake smile. Computers more accurately than any human being can listen to two female voices and tell whether or not they belong to others. They are a mother and a daughter. By the early 20s it's anticipated our smartphones will know what kind of mood were in and they'll respond accordingly.

Our jackets will give us a little hug which someone sends a friendly message to us. Our machines will begin to interact with us in a way that reflects their perception of our mood and also they'll begin to express emotions if you believe that makes sense in the context. But they will engender in us. Already they do this idea of a cute machine that we can look at a little robot and find that cute. They can actually engender emotions in us and they'll like us to those we feel we were interacting with other human beings. And of course all of this leads in due course.

It's a whole notion of transhumanism I suppose where the borders between human beings and machines become blurred. But it's an interesting step in this direction I suppose if you look at this man Professor Ishiguru in my view the world's leading academic roboticist who builds androids and this is an android of himself. And it's quite hard to tell which is him and which is the android. And partly this android gives lectures to people and when you get an android at the podium and you're more than a few rules back you cant actually tell that it's not Professor Ishiguru himself.

You may wonder as I fittingly did how it is that an android travels well finally what happens.

### [30:00 - 32:00]

is the head comes off, and the head goes as hand luggage, and the body goes in the hole. I just have this vision of the hand luggage opening up, and there's a headline. [LAUGHTER] Disconcerts and passengers. I think it's also important to say a little bit of a robotic. So what im just talking about is the ways in which our machines are becoming increasingly capable, handling data, solving problems, even expressing and detecting emotions. Robotics fascinates me for a whole bundle of reasons, and how rapidly it's progressed over the last few years.

It's interesting if you read a book that was published in 2004 by two MIT economists. It's called The New Division of Labor. And in that book, these economists were trying to identify the kind of work that all the human beings could do in the future, and those that machines could take on. And they make the observation then, and these are leading experts. But there's certain things you cannot really imagine a computer doing. You cant imagine, they said, a computer cutting your hair. You cant imagine a computer doing your garden.

And thirdly, they said, you cant imagine a computer driving a truck. Now this was, in 2004, the leading commentators on this whole question. And of course, we now have fleets of self-driving cars, And every series, motor manufacturer in the world, is investing in this area. I dont think it'll be 25 years before we look back and say, it's amazing we used to drive cars. It used to sit there behind a wheel and do that. Why and after? What an absurd waste of time. So this technology is coming along.

And I think it's a good example of how the unimaginable becomes an everyday reality in a decade or so. The machines are becoming increasingly pervasive as well. Already with a billion mobile phone subscriptions, a billion smartphones. But beyond that, those are your intranetive things. This idea that chips will be embedded in everyday objects. 40 to 5g billion devices is anticipated just within the next five years. And actually, these will be chips also embedded in us. The whole idea of it.

### [32:00 - 34:00]

Ingestible machines, little... You can see on-line all sorts of illustrations of this, where people actually can swallow pills with chips on them. These chips roam around the body monitoring dispensing, communicating. So chips in everything, not just inert objects, but human beings too. And then a different way human beings though, fourthly, are going to become increasingly, or we are becoming increasingly connected. And it's not just the high definition... Desktop to Desktop, Video conferencing, which will transform all our lives.

We also notice in our research, the development of interesting social networks, that were closed to the general public. Social networks like Cermot over 200,000 doctors in the United States. No pharmaceutical companies, no patients. Doctors online exchanging news and views and research findings about medications, therapies, and so forth. The same in education and Ed model. Less successful, but it's the same idea, legal on-brand. But I think we will, in due course, have the same in law. And in architecture too, in our connect.

Another form of social network, which we find fascinating, more of which in a minute, but this idea of recipients of professional service coming online and saving experiences. Patients like me, about a third of a million patients, are particularly people suffering from chronic illnesses, who find greater comfort and also getting better advice in some cases, not by regular visits to the doctor, but from their peers. And secondly, the emotional support of the empathy, the one who can gain online from patients who are suffering from different disorders.

We solve the same though in religion and architecture as well. Even in tax. Fascinates us in tax. If you're filling out a Turbo Tax and you have a problem, you dont then pick up a phone to traditional tax advisor. You go on to the answer exchange. And in the same way as I dont know how many of you still use Windows, I've been a Mac convert for many years. But I remember when you used to use Windows, they came up with some completely unintelligible error message. What you would do is,

### [34:00 - 36:00]

cut it pastor into google and up would come some clever people out there who could tell you how to start out the problem. This community of experience, communities of professional experience, refining in areas such as medicine architecture as well. So they are, our underpinning technologies are growing at an exponential weight, our machines are becoming increasingly capable, they're becoming increasingly pervasive, and we as human beings are becoming increasingly connected. And the big news is there's no finishing line thats the heart of the world. The hardest thing for all of us to accept.

The technologies that may well transform our lives by 2025 probably haven't been invented yet. It's hard for us to grasp because we can all remember a pre-internet world. But this takes me on to AI, and I believe it's important to understand AI in at least two waves. And for me this is quite nostalgic because in the 1980s I spent much of my time in AI, and I just thought id say a little bit of an act. In 1981, when I was an undergraduate in law in Glasgow University, I wrote this dissertation called Computers in the Judicial Process, and this was the beginning of my journey.

Because part of this work exposed me to the field of artificial intelligence and law, and in 1981 there were 26 publications globally in English on the subject. And I thought that would be quite a good thing to do some research in, so I went to Oxford, I did my doctorate in artificial intelligence and law. It was called Expert Systems in Law, a Jewish potential inquiry, a snappy title, I accept. And then I wrote my first book in '87, where I think about four people read, called Expert Systems in the Law.

But then from 1986 to 1988, I was involved in a project that I think is of relevance today. Look at this piece of legislation. Section two of this act shall not apply to an action to which this section applies. This is serious. Some legislature sat down and thought this was a good job well done. So this turns out to be one of the more intelligible.

### [36:00 - 38:00]

clauses or sections rather, in the latent damage act 1986. And this was a new piece of legislation written by an Uncle Phillip capper, who at the time was the dean of the law school at Oxford University, head be one of my examiners. And he wrote this book on this unintelligible piece of legislation, or the case law that previously had been developed, that was relevant for it. And also he wrote this book about how he anticipated the field would develop. And he said to me that he thought, "id finish my doctor." He said, "This would be a good area for some expert systems work.

Why dont we build an expert system of his knowledge?" So at the time of this branch of artificial intelligence known as expert systems, the idea was, and it was all sorts of different areas. You sat down with a human expert, you mined the jewels from their heads, you represented their knowledge in a computer system, you essentially developed a decision tree around which they could roam users. So an expert, I wanted to say just by with content, there's two ways the machines can do clever things.

On the one hand, where all the clever stuff is done by human beings outside the system, you develop that complex decision tree, you drop it into a system, and the system becomes as aware of the method for distributing the knowledge. And then there's the systems, the clever systems, that both produce and distribute the knowledge, where the systems offer anything on say unstructured data, do clever things themselves. This is very much the first wave. And we developed the system, and I need to explain myself here, at the time, ladies and gentlemen, this was as cool as you could imagine.

This was state-of-the-art design. We were very proud of that. So this was a system, essentially a big decision tree of philips knowledge. We, it was published, and this was when floppy disks genuinely were floppy, 500 quarter in floppy disks. We published a system that was appeared in the back of a book we wrote about the system. The system itself was a kind of standard interactive expert system. It asked questions, yes, no answers, pull down menus, and eventually, and it roamed through this tree.

### [38:00 - 40:00]

The huge decision tree answering the question, "When can my action no longer be raised because it's time-bought?" thats the, it belongs to the law of limitation. A complex area of law, over two million passes through the system. Our system reduced research time from two hours to two minutes. And this was done in the late '80s. I dont exaggerate this as exactly as we reported it. And I was heavily involved in the AI and law community. Indeed, in 1991, I chaired this conference, the third international conference. The AI and law that was held in Oxford. So it was a big part of my life.

And at the time, after the latent damage system, but maybe as this conference crept along, I was becoming less comfortable. But it seemed to be truly inevitable that within the next 25 years, everyone would be using this kind of technology. These questions and answers systems are outperforming. Philip said the system was better than him. Surely everyone would be using these systems in 25 years. And also just in law, it was also in medicine and tax and audit and consulting. But these systems didn't really take off. I want to qualify that in one way they did.

If you look at any tax-compliant systems of the big accounting firms, they're all based on rule-based expert system technology. If you look at Automation Document Assembly, it's all based on rule-based expert systems technology. If you learn a lot of online legal guidance, it's all based, for example, neotrologics work, for example, Alan over his work. It's all based in 1980s, rule-based expert systems technology. So I won't have it when people say it died away completely. But what I anticipated would be pervasive.

It seemed to me thats just surely the way that people will solve their problems. And I think it didn't succeed for three reasons. First, it's very costly to develop these systems. Secondly, this was at a time where the cost pressure in the legal profession weren't as great to another professions too. And everyone was charging by the hour. So why would you really want to develop a system that reduced your research from two hours to two minutes? If clients weren't clamoring for that kind of reduction in time and your competitors weren't doing otherwise, the commercial industry.

### [40:00 - 42:00]

incentive wasn't there to develop these systems. But thirdly, there was this, the web. Came along the first public accessible web site just 40 days after the end of that AI law conference, this thing called the web. And a lot of us just saw immediately that this is a far more exciting way to make legal content, legal guidance, legal advice available online. To be true, these systems that we developed, those of us who then could heavily evolve with systems in the web, they weren't nearly sophisticated. They didn't answer questions in the way the expert systems did in the 1980s.

But there was an immediacy and there was a usability and there was an inexpensive evolve that trumped, to be honest, the development, where we were able to come in the 1980s. So I largely, as AI law behind. And then this happened as we know in 1997, when the World Chess Champion, Gary Kasparov, was beaten by Deep Blue, IBM, computer system. Now in the 1980s, when I worked in the obviously university computing laboratory, we talked about this a lot. Could a computer system ever beat the best chess player? And our answer generally was no.

And the reason our answer was no is absolutely pivotal for today's discussion. We thought the answer was no for this reason. In chess playing, and as many other areas of expertise, often human experts couldn't explain how they performed. Some people call this tacit knowledge. The human experts could identify rules for their mundane level of playing, but they're real pieces of genius, their real creativity, their real imagination seemed to be ineffable, inexplicable. No one could actually reduce that to rules.

And remember the model then was that the only way you could get these systems to perform at the level of experts was by representing experts knowledge in a computer system. So if you couldn't articulate--

### [42:00 - 44:00]

the rules-- these systems were destined to perform at quite a high level, but none of the level of experts were intuition, creativity, genius seem to be required. But what we hadn't banked on was the exponential creasing processing part. And by the time IBM's deep blue beat Gaddick as far off, this was a system that could look at 330 million possible moves in a second. Even the world's best chess playing champion struggle to more than about 100 moves in their head at any time. So we had a very simple, single moment. Gaddick as far off was beaten by brute force computing by huge amounts of data.

And this is absolutely vital. Because it's Patrick Winston, one of the fathers of AI puts it, there are lots of ways of being smart that aren't smart like us. It's a phenomenal arrogance that we had in the '80s thinking that the only way we could develop computer systems to work at the level of the best lawyers was by representing the way lawyers solve legal problems. And we find the AI fallacy committed by some of the world's leading AI scientists. We find it certainly in the media. And I suspect people in this room tend to fall into the trap as well.

It's the mistaken assumption that the only way to develop systems that perform tasks at the level of experts are higher is to replicate the thinking processes of human specialists. That defined AI and expert systems in the 1980s and much of the '90s as well. That the only way to get these systems performed at a high level was to model it to human beings. And yet we saw with Capacasparov that he was beaten at a different game. He was outperformed by a system that had huge amounts of data and fingertips and was capable of astounding processing. Now let me give an example of this.

Many people say to me, "AI is impossible in law because why my clients come to me is because of my judgment. How can a computer ever exercise judgment?" There's a little bit of a few of...

### [44:00 - 46:00]

I also say the same with creativity and empathy. Just as a psychomite, there's a little bit of a definitional trick there. If you think that judgment, creativity, empathy are uniquely human attributes, then, of course, machines can never exhibit any of these. But lets put that to one side. And to say to you that how can computers exercise the judgment in any event is the wrong question? This is the better question.

To what problem is judgment the solution? Why in law, and right across our professions, do we call upon the judgment of our human experts? And the answer to that, we think, is we live in a world of uncertainty under certainty, facts, uncertainty of knowledge. And when we have problems fettered by this uncertainty, we go to experts, human experts, and we say to them, "In your experience, using your judgment, how is it, do you think we should proceed? whats the answer?" So what were calling upon our experts to do is to manage uncertainty.

And so the question we should ask is not can computers exercise judgment? It's can computers manage uncertainty better than human beings? The answer to that is almost certainly yes. It's the kind of thing, again, Dan will be talking about literature. It's the whole story of big data that if you've get tens of millions of data items stored, then the uncertainty that human beings usually cope with by drawing on what they call their experience. A couple of thousand examples is well-trumped by the technology.

I've related question myself and asked, which I love because im interested in philosophy, but I think it's not entirely relevant for this discussion, is can machines think? And my favourite illustration of this, and this is just, I think, simply wonderful. The day after Watson won on Jeopardy, John Sarah, a backlit philosopher weed rather not be able to do that.

### [46:00 - 48:00]

He was present a lot on AI, and he wrote an op-ed that was headed this. Watson doesn't know at one of Jeopardy, and that is perfect. It didn't want to go down to the pub to celebrate, it didn't want to phone its family to tell how it felt, but it still was. And what we are finding therefore, and we call this the second wave of AI, forget the distractions but whether our machines have these emotions or exercise, judgment and so forth were seeing their emergence of increasingly capable non-thinking machines.

These are machines that outperform human beings, but dont operate like human beings thats the second wave of AI. And it amazes me just even over the last few weeks we saw, we've seen yet another illustration of this. The game goal, the board game, apparently is more permutations than there are atoms in the universe. And not many months ago, leading commentators were saying it'll be ten years before a system can beat the best Go! And a few weeks ago, four to one, the world champion was beaten by the system called Alpha Go.

Now whats particularly fascinating about this kind of system, which is based on neural networks, not the neural nets we use in the 80s as a form of representation language, but neural nets is a way of supporting machine learning whats particularly interesting in the development of the system is how it played the game with itself. This idea of recursive self-improvement. And im not yet sure that looks like in law were certain seeing machine learning and thats to say machines that perform better as they're used.

But actually the idea, I've only acknowledged that I can think of just now, imagine a computer system that could draft a contract, then correct itself, redraft it, and then correct itself. This kind of recursive self-improvement, which has taken the deep mind people part of google in a number of game-playing scenarios from systems that could barely play the game, playing themselves many millions of time to a level that outstrips human beings. Here's another way of looking at the whole area of the eye.

### [48:00 - 50:00]

Consider this, there's really four faculties of professionals, it seems to us. What do we have? When you think of any professionals, with our cognitive capacity, our ability to think, to solve problems, with manual cycle motorability, dexterity, with effective skills, emotional, detecting, expressing human emotions. And whats a moral capacity that as professionals, not only do we offer guidance, but in some sense we take responsibility for the advice weed have to do. We have to take responsibility for the advice we have to do.

We have to take responsibility, and we have to take responsibility for the advice we have to do.. We really want a machine to make an excuse to turn off a life support system. Do we really want a machine to make the decision and to pass a life sentence? Do we really want moral capacity? Do we want to take responsibility for the advice we have to do we want to take responsibility for the advice we have

### [50:00 - 52:00]

to do in the 20s which we call the medium run here are the roles that we reckon will be taken on by the people who be solving problems that used to be the province of traditional professionals. Now most young professionals and frankly older professionals look slightly horrified by this list partly because they dont understand some of the words but partly because they dont sign that exciting either I didn't go to law school to be a process analyst I didn't go to medical school to be a data scientist but actually they in this time of greater technological progress more rapid progress than the

world has ever seen we need to rethink the roles we take on. There'll still be roles certainly in the 20s for people who are interested in improving access to justice and better health but they won't look like the traditional doctor and lawyer. In the long run though we often refer to this is an article review happily a pleasant review written of our book in the Economist in the Shupita column and it is wonderful cartoon of Professor Dr. Robot c you can see as good as way gone as good as the count of his leger status open so forth.

And very many people when they think of the future of jobs once they hear about the AI narrative that I've been unfolding they haven't made that one day they'll come in and sitting in their chair will be Professor Dr. Robot c taking on the job and this just isn't how it's going to unfold. Just a couple of things incidentally if you were a device for example devising a system to draft documents why on earth would you give arms and legs.

I dont show this idea and it's in the im afraid many of the room commit this error I wish weed stop talking about robots it's an amazingly anthropocentric way of representing legal problem solving the future why should why on earth would we want a machine that looks like a human being to do a lot of these tasks thats just a side coming the bigger comment is this then in the long run when we look at it machines will become as I said increasingly

### [52:00 - 54:00]

capable and they've got to take on more and more of today's tasks you shouldn't think so much in terms of jobs for the future you should unpack jobs into the tasks involved take nurse for example and nurse 25 years ago a lot of their work was about bedside conversation and bedpads today nurses in the k are writing prescriptions they're doing minor surgery we still call them nurses but the tasks they take on have changed it's not that the job has disappeared the job has changed and what were seeing is uh...

a switching of tasks many of the tasks that human lawyers used to undertake will be replaced be replenished by the roles and tasks I've just mentioned so at the same time however machines are going to be taking on more of those tasks and as we show it in the 20s no doubt new tasks will emerge but here's the key point it's likely the machines will take on these as well there's no reason to think that list of roles and tasks I mentioned won't itself or themselves be susceptible to the same technology so we find it hard to avoid the conclusion in the very long run by which remaining 40s 40s 50s

the best steady decline in the need for human professionals in the long run once one unpacks the jobs you realize many of the tasks will be replaced by machines some new tasks will emerge but if the rate at which machines are taking on the new tasks outstrips the rate in which they're emerging you come im afraid ladies and gentlemen to stay where there's less and less for human professionals to do now we started off writing a book we thought it was asking and answering the question whats the future of the profession in fact were asking and answering a more profound question how do you produce

and distribute practical expertise in society you see asking the question whats the future of lawyers whats the future of professionals in a sense assumes an answer if you will it assumes there will be lawyers or professionals or it somehow assumes that they are part of the future landscape we dont deny that but the more important question and good battery analysis of what professionals for is how do we produce and

### [54:00 - 56:00]

distribute practical expertise? That is the problem for which the traditional professions have been the answer. The traditional models professionals in firms and schools and hospitals. But what we see is there are six alternative models emerging and are run through them very quickly because im conscious of time and im finished in about five minutes. The first is the network experts model. The economists call this workers in tap, axis is an example of this is this idea that actually professionals dont need to work in traditional institutions, hospitals, schools, firms.

They can be self-employed, they can be organized in different ways. So it's still human experts delivering the service but challenging in a network society whether or not the traditional institutions that add in their overhead so much cost to the process are necessarily needed. Then there's a paraprofessional model, the classic model in law. People all say the broad-based triangle, junior laws at the bottom, deep experts at the top.

And conventional wisdom has it that what will happen with AI and if the law is simply with this well see this certainly is that the machines will take on tasks from the bottom up. But also what you're seeing with Watson, of which thats a logo there's a different model. Imagine in health for example where a diagnostic Watson system outstrips a human expert but you still want someone offering empathy. So actually the paraprofessional model on this view is that the expertise, the diagnostic power come from the machine and the inter-personal service might be delivered by a nurse.

So a nurse with Watson will outperform in terms of empathy and in terms of diagnostic performance and treatment planning the traditional model. Now there's a knowledge engineering model that one I mentioned earlier. It's not dead, the role-based expert system model. As I say it underpins much that we see in tax, much that we see actually in law too. This idea that we do, I think for maybe not world-class performance but very high performance, we remodeled.

### [56:00 - 58:00]

complex expertise in the form of decision trees that others navigate around. Fourthly, and I've mentioned this, communities of experience model, where the way in which the knowledge is produced and distributed actually doesn't really involve experts at all. It's where recipients, patients, clients, students, share their experiences themselves. And often people find it more comforting, communicating with a peer, who summer to similar problem or face to similar challenge.

Now we believe, in this one of the roles, that these communities of experience would benefit from being moderated thats to say the experts roam around and perhaps identify areas where perhaps dangerous or highly misleading guys being offered. But there is no question that already exists. This is a flourishing way in which problems to which human beings used to be the only answers in the form of professionals are now being sorted out. Now you may see, of course, this is terribly dangerous, you need to speak to a lawyer lets remember again our grave problem of access to justice.

That almost no person in advanced jurisdictions any longer, other than in situations of life or death, metaphorically speaking, can actually afford to seek the advice so that they can understand their entitlement and enforce their entitlement. We need different ways. Then in Bay the Knowledge Model, think of this game we all play in our handheld machines. When I was younger we played this with atoms. We called it playing cards, we called the game patients. I dont know if those are your memories to play.

Now what used to happen when you played that game with cards if you tried to put a red five under a red six? What would happen? I dont know why you want to do it's called cheating, in fact. But you could do it thats the point. What happens when you try out the system? The cards flicked away. The rules are embedded in the system. And this is actually crucial because in the future, particularly in law, rules are going to be embedded in systems. As our clients and manufacturers,

### [58:00 - 60:00]

of production, more and more of their processes are automated. It won't be that suddenly there's a human being in the loop to look at a regulatory and legal issue, the law will be embedded in the workflow. And the same in financial services as well. The global financial services essentially, hugely underpinned by technology, a high degree of automation and workflow and knowledge processing. The idea that lawyers are the only human beings left involved to be advising when you can actually embed the knowledge seems to be unimaginable. Then there's the machine generated model.

And this is the one most people fixate on. They say, "Oh, you're saying machines are replacing human professionals." I suppose in the long run were saying this is very likely. But really that is decades from now. I do want to urge people who are thinking they're five alternative models before you get to the idea of big data systems outperforming human beings, or of whats it like systems outperforming human beings, or indeed of alpha-go systems outperforming human beings. Final thoughts, I believe the legal profession is evolving through three stages. The first stage is denial.

And thats the stage. And it's lasting up until about this year where leaders, both infirms and state bars and so forth, were frankly in denial that the profession was going to change fundamentally. I find it hard now to look at it when straight in the face, we say, "were going to go back to 2004." It seems to me we are where we are And thats a stage of greater technological change for the world's ever known. Unimaginable, we won't change.

But over the next five years, up until about 2020, the dominant change, I believe, will be resourcing of outsourcing off-shoring subcontracting, finding new ways, essentially, of delivering human service, taking the cost out of the more routine work through sourcing in different ways. Stage three, the disruption is going to occur in the 20s. And this is going to be underpinned, ladies and gentlemen, by artificial intelligence. The systems that are evolving, that are taking on more and more of the tasks that historically we thought only human beings could undertake.

For many lawyers when they hear this, they hope they can hold out to retirement before I hear it.

### [60:00 - 62:00]

They're going to stuff engulfs them. And thats a nervous laugh I hear in the audience. I argue this is a window of opportunity. I love this petition from Thomas Edison. The opportunity is missed by most people because it's dressed in overalls and it looks like work. And I say this because there is a certain glamour, I think, a certain excitement about AI and machines taking on tasks that we used to think require human intelligence. But actually, a lot of people are going to be able to do that. But actually, a lot of it's going to be about hard work.

It's going to be rolling up our sleeves, clever people thinking imaginatively and creatively about the development of new systems. And my advice to the legal world is from Jack Welch as this, "Change before you have to." I think what we've seen in something like Alpha Go and were seeing in robotics, were seeing this dramatic acceleration in some technologies. Some lawyers say, "Oh, I dont want to be a pioneer. I want to be a fast second." thats usually our rationalization for doing nothing. I dont think we should be in our back seat.

I think if we look at the legal world, I have never argued that the need for legal and compliance services shrinking, quite the opposite is increasing. The only question for legal profession, traditional lawyers, is whether or not they have a significant role in delivering that service were seeing a whole bunch of alternative providers, which are great, providing greater choice for consumers. But also, were seeing a whole set of technologies coming through.

Now, the people who seem to me who are going to survive and thrive through the 20s in the legal industry are those who invest in the underpinning technologies and in particular in the kinds of AI we've been discussing today. Thank you very much. Thank you. I understand well get time for a few questions and then well break. I'll be guided by Lyre as to where we should call a halt. But sir, the question in the middle there. Yes, I thank you thats a wonderful presentation. Thank you.

### [62:00 - 64:00]

im Mary Ann of Questions in my mind. You spoke a little bit about morality. We've all seen with certain elements of disruption, user, Airbnb, which has challenged global legislation. I think if you apply that in a broader scope with artificial intelligence, coming from a project manager background world about auditing and ensuring that quality control levels can be the reality standpoint. You know, raising some questions.

As the processing power, as perhaps a lot of these technologies in advance, this gives distributed to larger organizations, or consulting firms you refer to, will have the where will fall, the funds, the... Okay, I should rush you because I want to get there. I forget me, but we should give you a couple of questions. Where do you see, if there are going to be college industries that are developed to ensure the security and security that these approaches sustain the greater good? Well, I think you raised my fundamental issue. We raised two moral issues in the book.

One is the one I mentioned about whether or not there are certain tasks and activities that even machines cannot perform us. We dont want them to go. The second is more profound, I think, and that is the question who owns and controls the systems and machines. And our worry that we share is that we might go from the old geek, keeper to retrospect, made the rub of benevolence, the doctors, the lawyers, professionals, to major technology providers, Internet search. And there's an even bigger question there.

The economic question is income is derived increasingly from capital rather than from employment. Fewer and fewer people will have jobs to do. And if that content and system is held by a very few, we live in a very fragile society, a polarized society. And our answer, and we do have the rhythm this at the end, digging into a little political velocity there, is that we should really prefer. And again, we need to debate on this now.

### [64:00 - 66:00]

a commons approach so that these systems and content in so far as we can make this happen now, should not be held by many corporates, nor indeed by major governments, but we should be thinking of common holding of them. So, we haven't fooled that argument in the book. Great question. Thank you. Other question in the back, sir? I wanted to respond to your comments about AI model.

I think the problem for AI model in the years has been that the techniques for connecting their models, computational models, of legal argument, to legal texts did not exist, but now with the technology underlying Watson and the techniques that are available, that can. Yeah. And so, I think that will improve, it will certainly reduce the cost to develop the systems and help the field to make a real contribution to the legal field. Yeah, it's very different from the field we worked on in the 80s, though. Yeah, thats absolutely right.

And thats what fastening, I sometimes joke and say that I write the same book every four years. But in a way I do, because the technology has changed, and in a sense what you're seeing is the technology now are there to realize we were working heavily in the 80s in the same community to realize a lot of the vision we had. And then I dont think the visions change, which I always just say functionally was to make scarce legal expertise and knowledge more widely available and more easily accessible. And we did that some crude ways, and now were finding new techniques that are evolving.

And I guess in 10 years time well look back and think how naive we were and well be a third wave, but we cant yet anticipate. But thank you for that. One, two. You've blown my mind, that was absolutely... im glad you reminded me of it im going to see that my practice is going to change in about 20 minutes. I wonder though why in the discussion of the legal profession.

### [66:00 - 68:00]

And the professions as a whole, you chose not to apply this same rigorous analysis to politicians who have to create the law that we will have to provide sage advice upon. Well, first of all, on your comment, you've been waiting for me for a little bit, joke. One of the things I do want to say, by way of reaching out well understand if you leave the room. I dont think this is as urgent as maybe, as to say, the world is not going to, particularly clients, move slowly as well. So this is not something we need to sort out over the next six months.

I think as we go through this period of five years of resourcing, we should also be trying to prepare for the 20s. So it's not if you dont do something in the next couple of months you've done, it really isn't that. I never want to be thought of in that way. Politics is an interesting friend of mine who wrote the Times, Daniel Lord, Daniel Finkelstein, heard one of my lectures and then wrote the article. You probably want to read, which is why we apply this to political thinking.

More, in the sense, science is probably another son who's actually writing a book on precisely this or not the impact of it. If you take the same technological changes, what does this mean for political ideas? What does this mean for political institutions? What does this mean for democracy and liberty? So, certainly, our re-family business, the task is at hand. So thats the answer. Okay, keep it in the family. No, sorry, it was one of... Yes.

Actually, that was very similar to my question, which is, when you talk about the example of applying AI to medical context, the human body, human evolution doesn't happen on that fast for timescale. Social evolution, loss and change in the day. How does your work take into account the fact that AI is a moving target law, and the legal landscape can change depending on these technologies being implemented? Yeah, it seems to me that this is a problem in the 80s. We used to say that one of the criteria for a good domain of application for expert systems is around the world.