| podcast interview Ð microsoft artiÞcial intelligence · 2018-03-12 · beyond the hype:...

14
1 Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft Beyond The Hype: Artificial Intelligence Podcast Interview – Microsoft Recorded 2nd November 2017 In association with Dr. Joseph Sirosh Corporate Vice President, Cloud AI Platform, MicrosoThe paradigm shito AI is as big as the shito the internet. Welcome to the MMC Ventures Podcast. We’re going beyond the hype in artificial intelligence. A warm welcome to listeners. I’m David Kelnar, Partner and Head of Research at MMC Ventures, the insight-led venture capital firm based in London. In this six part series, we’ll be hearing deep insights from some of the world’s leading AI technologists, entrepreneurs and corporate executives – while keepings things accessible for the non-specialist. I think AI is today’s most important enabling technology, but it’s not easy to separate fact from fiction. My goal for this series is for us to come away better informed about the reality of AI today, what’s to come, and how to take advantage. This series is sponsored by Barclays. I asked Barclays for a strapline they’d like to include as sponsor and I thought their response was really interesting. “Thanks. I’m not sure about slogans. Here’s just how we think about AI. We think AI is incredibly important - a whole new field that’s as significant as anything that has gone before. And we think about it a lot. We think AI is vital to our business and we’re working hard to take advantage of it for our customers. And we need to learn from, and collaborate with, a wide range of people to ensure success. Technology advances fastest not when it’s held close, but when people go out, listen and contribute.”I thought that was better than any slogan, so I asked if we might run with that. I have pleasure in doing so. My guest today is Joseph Sirosh, PhD, Corporate Vice President of Microso’s Cloud AI Platform. It’s a great privilege to speak with Joseph. Joseph leads Microsoenterprise AI strategy, and Microso’s cloud AI products including the company’s Azure machine learning, search and chatbot initiatives. I’m excited to talk with Joseph today about the implications of AI, the role of Microso’s cloud AI platform and how it will evolve in the future, the adoption of AI in the enterprise and more. Prior to his current role, Joseph was Corporate Vice President for Microso’s Data Platform. Joseph joined Microsofrom Amazon, where he was Vice President for Amazon’s Global Inventory Platform. Joseph was responsible for the science and soware behind Amazon’s supply chain and [00:01] Joseph [00:05] David Interviewed by: David Kelnar, Partner and Head of Research, MMC Ventures

Upload: others

Post on 29-May-2020

3 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

1

Beyond The Hype: Artificial Intelligence | Podcast Interview – MicrosoftBeyond The Hype:

Artificial IntelligencePodcast Interview – MicrosoftRecorded 2nd November 2017

In association with

Dr. Joseph SiroshCorporate Vice President, Cloud AI Platform,Microsoft

The paradigm shift to AI is as big as the shift to the internet.

Welcome to the MMC Ventures Podcast. We’re going beyond the hype in artificial intelligence.

A warm welcome to listeners. I’m David Kelnar, Partner and Head of Research at MMC Ventures, the insight-led venture capital firm based in London.

In this six part series, we’ll be hearing deep insights from some of the world’s leading AI technologists, entrepreneurs and corporate executives – while keepings things accessible for the non-specialist.

I think AI is today’s most important enabling technology, but it’s not easy to separate fact from fiction. My goal for this series is for us to come away better informed about the reality of AI today, what’s to come, and how to take advantage.

This series is sponsored by Barclays. I asked Barclays for a strapline they’d like to include as sponsor and I thought their response was really interesting. “Thanks. I’m not sure about slogans. Here’s just how we think about AI. We think AI is incredibly important - a whole new field that’s as significant as anything that has gone before. And we think about it a lot. We think AI is vital to our business and we’re working hard to take advantage of it for our customers. And we need to learn from, and collaborate with, a wide range of people to ensure success. Technology advances fastest not when it’s held close, but when people go out, listen and contribute.”I thought that was better than any slogan, so I asked if we might run with that. I have pleasure in doing so.

My guest today is Joseph Sirosh, PhD, Corporate Vice President of Microsoft’s Cloud AI Platform. It’s a great privilege to speak with Joseph. Joseph leads Microsoft enterprise AI strategy, and Microsoft’s cloud AI products including the company’s Azure machine learning, search and chatbot initiatives.

I’m excited to talk with Joseph today about the implications of AI, the role of Microsoft’s cloud AI platform and how it will evolve in the future, the adoption of AI in the enterprise and more.

Prior to his current role, Joseph was Corporate Vice President for Microsoft’s Data Platform.

Joseph joined Microsoft from Amazon, where he was Vice President for Amazon’s Global Inventory Platform. Joseph was responsible for the science and software behind Amazon’s supply chain and

[00:01] Joseph

[00:05] David

Interviewed by:David Kelnar, Partner and Head of Research, MMC Ventures

Page 2: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

2

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

order fulfilment systems, and Amazon’s central Machine Learning group, which he built and led.

Joseph holds a PhD in computer science from the University of Texas and has been a leader in the field of data science and machine learning since the 1990s.

Joseph, a very warm welcome.

Thank you for that introduction.

Joseph I’d like to start by thinking about the impact AI will have quite broadly. In a recent speech, you described the Cambrian Acceleration – the time, four billion years ago, when organisms’ ability to see, among other factors, enabled a radical expansion in organisms’ capabilities and sophistication. You suggested that with AI, we’re at a similar moment in software to that Cambrian Acceleration. What do you expect to unfold in the decade ahead?

It’s a great question. Think about it this way: until now software was really about digitising the world into bits and allowing programmers to write programs. With the power of AI, now for the very first time, you have the ability for computers to understand the data. Remember: to see, for example, you need to understand what’s in the data – not just register bits in a pixel and pixels in a camera for example. So, AI is for the first time giving the computers the ability to understand what’s in the digitised bits. And second, computers are going to be able to bring to act without being programmed. And that’s a fundamental change in what computing itself is.

And so, until now software was really simple and stupid. In the future, software is able to understand its data; it is intelligent, it is predictive, it is able to act. It’s able to help augment the human power in untold ways. That’s the future that AI enables and AI in many ways is going to become the new normal.

Now if I had said in the early 1990’s that internet is going to become the new normal, a lot of people may not have really believed. It feels like a very strong statement. But it happened at that time, and there were thousands of start-ups during the dot-com boom – and out of that a whole number of start-ups survived, and a tremendous transformation happened. The same is happening with AI today. There are thousands of start-ups and very large number of companies leveraging AI. AI has seeped into the fabric of everything we do on the mobile phone and very soon it will be part of all software. And that’s a new normal.

Do you think AI is different from previous paradigm shifts in software we’ve seen – such as the moves to mobile or cloud computing? Does the ability for software to learn through training make it fundamentally different from previous paradigm shifts?

It is absolutely as big, but I’m not sure its fundamentally different because it’s still software. AI is an evolution of software, and much as the term artificial intelligence leads one to think in human-like forms, it is far from that. It is algorithms that are very mathematically describable; it is statistical in nature, it analyses data. It is not human intelligence, at least not yet. You know, remember, humans learn differently from machines. Humans are able to learn with one shot learning. They’re able to use context in ways that are otherwise are never possible. Computer software, even with artificial intelligence algorithms, is not able to approach that. And I have a saying, all artificial intelligence is dumb, but some artificial intelligence is useful. And you can create a lot of utility from AI algorithms, that’s a very powerful revolution in itself, but it is an evolution of software.

What will be the most significant implications of AI do you think?

[02:18] Joseph

[02:20] David

[02:47] Joseph

[04:34] David

[04:48] Joseph

[05:46] David

Page 3: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

3

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

I think there are incredible, possible things that AI can help with. Let me give some examples. Healthcare is an area where AI is going to be extremely revolutionary. Now we have a partner called Epic software in the US, Epic Systems. They build electronic medical record software, covers about 65% of the US population. They have built an Epic cognitive computing platform on Microsoft Cloud, Microsoft Azure. And they’re enabling their customers, which are hospital chains, to analyse patient data.

And so, some of the insights people are able to find are amazing. For example, from historical electronic medical records you can predict hypertension, blood pressure up to two years in advance. You can look at the common workflows in a hospital that physicians and nurses follow and simplify that so dramatically. You know for example when a physician treats a child with an earache, he has to follow a very standard procedure, which is very similar to what he follows with a lot of other patients and it’s slow work many times. You know exactly what kind of tests you will prescribe. You might also compress that workflow and make it all automated for the common cases so that you use very little of the physician’s time for the simplest cases and take that precious time and apply it to the most complex cases which truly need attention. It can lead to great optimisations in healthcare.

One more example: In India, we have a partnership with an eye-care institute for a variety of eye care related applications of AI. For example, we can actually use historical data and estimate how a patient’s eye heals after Lasik surgery. It turns out every patient heals differently. And the final vision you may end up with after a surgery may be different in different people. Now you have before and after data and when you can actually apply machine learning, you can actually predict what the vision is going to be, many can actually predict what the vision is going to be a month after healing. You can correct it ahead of the surgery. So, you can have perfect vision every time you do Lasik surgery. And that’s powerful and so on so forth.

Healthcare is rich with applications for machine learning. Assistive technology…at the keynote today I am going to be able to show a demo of in fact one of my colleagues, he’s actually blind – visually impaired, he has an application on a mobile phone that uses powerful cognitive resources in Microsoft cloud and is able to help him scan documents, recognise his environment, recognise people and really assist him in performing in a complex world dramatically better than otherwise, very powerful. So, I think you will see AI seep into the fabric of our lives, empowering us in every aspect of our life.

What’s most misunderstood about AI today, do you think?

I think it is the sci-fi nature of AI. It couldn’t be further from the truth. I think we tend to see AI, computers and software in human-like terms, which it is not.

By the way, this reminds me of something somebody said, which was I forget who, said, you know we should never be afraid of AI because AI does not have a survival instinct. Humans and life do. And AI will never take over, even if it becomes something that it’s not today. You know, because it doesn’t have a survival instinct. I really think AI is useful software first and foremost, applied in the right way and Microsoft has espoused a set of principles for AI. The acronym for it is FATE: Fair. Accountable. Transparent. Ethical. And I think if we, and by the way that applies to all technology, and I think if we apply those principles – Fair, Accountable, Transparent, Ethical – to AI, we build the products the right way, I think we have a very powerful engine for the economy.

Fantastic. Let’s talk a little bit about Microsoft’s cloud AI platform. Your platform includes AI infrastructure, and services, and tools. Can you explain to the corporate executives, entrepreneurs and investors who are listening exactly what you provide and how using those can help people gain

[05:52] Joseph

[08:47] David

[08:51] Joseph

[09:59] David

Page 4: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

4

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

value from AI? Because it’s not always very easy for people to, kind of, onboard to this ramp.

Yeah, great question. The way to think about it is: to apply AI you have to build systems of intelligence. And every company has to. Systems that help process credit cards detect fraud interact with their customers, have conversations, do marketing targeting, wherever AI is applied.

Now, building the systems of intelligence is a lot of heavy lifting. There is a lot of work to be done in bringing the data together, applying it to computer infrastructure, applying the algorithms, creating the web services and then driving actions and taking feedback from it.

Microsoft as a platform provider eliminates that heavy lifting. It creates the platforms and tools with which our developers, any developers, you know, corporate developers can easily assemble these systems of intelligence. So, Microsoft is playing a role very similar to what it has always played in software – in earlier days we provided operating systems, we provided tools like the SQL Server database. We now have the Microsoft Cloud. These are platforms that make the building of software applications dramatically easier. So, we provide the Microsoft AI platform to make AI as easy to build for the common developer as possible and as easy for any company to get value out of as it has been for a small group of extremely select companies. Okay, a small group of extremely select companies have been able to build their own infrastructure for AI, have the data scientists, the advanced machine learning people and the PhDs – and only they have been able to connect up all the components to make AI work. And now we are democratising it; we are bringing it to everybody.

Let’s explore a bit how your machine learning services will evolve in time. You currently offer pre-trained machine learning services related to computer vision, written and spoken language, knowledge and search. And as you described, developers can call on these services to deploy capabilities within their own projects. Will you be deepening existing capabilities in those areas, or developing broader capabilities within those above areas – or both?

It is both. It is both deepening and broadening.

Now think about an example of speech and translation. It is incredibly hard for every company to start developing accurate speech recognition or accurate translation – because it is not only about algorithms. It’s also about the data, it’s about the expertise in speech as a domain or translation as a domain.

So, what Microsoft does is bring all of those things together: the data, the expertise, and the software and algorithms together, to create the best-in-class services for speech recognition, or translation, or search, or language understanding, or face detection, or image understanding and so on and so forth. And that’s pre-built AI for you, right? So that any developer can tap into that pre-built AI and infuse AI into their application without knowing AI. And so, a big part of what we bring as a partner to companies is broader collections of such prebuilt components that make it possible to attach AI in all meaningful situations. And then in each of these areas, once there is already there, it’s going to be continually improving services. Speech services become more and more accurate over time, so you don’t have to adopt new technology. Just by subscribing to that service in the cloud, your application constantly improves. Translation keeps on improving, video understanding constantly so, you are now getting on a conveyor belt that is taking you towards a future along with our innovation as it continues.

I can clearly see how the capabilities of how each of those services will evolve and deepen over time. Can you give me some examples of how they might broaden? If you take an area like speech or computer vision, can you give a sense of the kind of capabilities you’ll be able to deliver in years to

[10:24] Joseph

[12:18] David

[12:49] Joseph

[14:25] David

Page 5: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

5

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

come that you don’t deliver today?

Great question. So, for example, in the case of speech, we have now a custom speech service that allows you to bring your vocabulary, then your vocabulary might be different from the vocabulary in another environment. Your collection of accents, your domain information so that it performs extremely well in those particular situations. When you take translation, for example, you can do the same thing – your customer vocabularies, all the languages you care about and being able to be very accurate in estimates. And this is going to be a key thing that we keep driving: customisation of your own data, to your own situation, and still not making you do the heavy lifting.

To what extent will Microsoft want to develop sector-specific services or applications in AI? I mean, for example, object recognition that goes beyond recognition of everyday objects. You alluded just now to speech recognition – like vocabulary, perhaps, applicable to the financial services sector. To what extent do you want to tackle the verticals?

This is a great question. The customisation is one step to helping people to get there. But then there are specific areas like customer care, where we can bring tremendous amounts of components together to make customer care great.

So, for example, Microsoft in our own support centre, we have deployed an AI-based conversational interface – a conversational bot. That has been able to field approximately 30% of all the questions that come to it well so that customers are self-serving at resolving it without escalating it to a human agent. That is really big in the world of customer care. It is much better for the customer because the instant self-serving resolution, it’s very cost-effective for the call centre and again like in the situation I talked about in the case of doctors and hospitals, you’re able to have your expert customer care agents spend time where that time needs to be spent. And that is very, very important. And so, we have the Microsoft Dynamic 365 solution for customer care, an AI solution for customer care now – and that sums up the areas where we will bring specific intelligence. And then the rest of it, by the way, is really going to be through partners. Microsoft has always been a very partner-friendly company. That’s our DNA. A platform company cherishes its partners and our partners bring, very often, the domain expertise and the data to be able to create custom solutions in vertical industries.

Fantastic. And do you, or to what extent do you, intend to expand into technological areas beyond vision, language, knowledge and search?

So, it is a very broad area of AI, that is…I want you to think about, for example, a marketplace or an app store on the phone. There are millions of apps. In the same way, you should imagine a future where there are a million APIs in the cloud. They can be for speech and language and vision and any number of things. And the task of a software developer is then going to be about glueing together these APIs that are hosted in the cloud. API’s that are reliable that never have an outage, that’s backed up, that is secure, that Microsoft stands behind, right. And the software development task is then, by taking these pre-fabricated parts, customisable, assembling them together into an intelligent software application. And that is the future in many ways.

Help me have a crystal ball here. What do you expect Microsoft’s machine learning services to be able to do in, let’s say, two years – and in five years? If we’re having this conversation in five years’ time…?

So, let’s take examples; it’s very broad question. Let’s take conversational bots for example.

Bots are becoming more and more useful. Now my first caveat to all this applies: artificial

[14:44] Joseph

[15:27] David

[15:45] Joseph

[17:15] David

[17:26] Joseph

[18:19] David

[18:31] Joseph

Page 6: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

6

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

intelligence is always dumb, but some artificial intelligence can be useful. And bots are becoming very, very useful with AI for getting tasks done much faster than otherwise. I really believe there is a browser-like moment coming with a combination of bots and AI. You know, when the browser was first invented in the early days of the internet, although the internet existed, right, and the internet evolved over 30 years, and hypertext protocol HTTP existed since 1988, [19]92 – when the browser came – usability of the internet dramatically changed. And the same thing is happening now, with bots and the power of AI, to help understand conversations and get tasks done.

And so, one of the things I foresee in the future, in two to five years, is that you will have incredibly intelligent conversational bots in every enterprise. In fact, I am able to show one of them at Barclays, which took just all of 30 mins to create. By scraping the Barclays web page, an intelligent bot was created that could answer questions. And so, you imagine conversational interfaces taking over as the primary ways that you interact with every service online. Not through a webpage – all web pages are designed for the PC era. And conversational bots and AI is designed for the mobile, intelligent era. And that’s one of the big things that you will see.

Fascinating. So, you think the language interfaces represent the next paradigm shift in human-computer interaction?

Natural language interfaces – not just language, but AI that’s able to understand the intent, and the context, understand you, the history. So, for example, I mean, I give you a simple example. If I tell my Microsoft HR bot “I’m taking a vacation tomorrow for time off” it has to understand who I am, as a Microsoft employee, salaried, vacation means eight hours. It has to go to the time and expense report application, invokes its API with tomorrow’s date, it has to look up what tomorrow is. Then it has to go find and see if that field of vacation, not sick leave, vacation is actually filled already. If it is not filled, already it’s reflected back to me and asks for my confirmation. That’s a task completion workflow. I would have gone and done that over a PC or a web page. It would have taken many steps and would have taken logins and all of those things. But a bot can complete that for me. And I think that’s how bots become really useful. Talking of chatbots: in the home, digital assistants’ roles I think will clearly continue to grow in capability and in ubiquity. Now Amazon, Google and Apple are all releasing their own in-home smart speakers with embedded voice control assistants. Microsoft is embedding its AI assistant, Cortana, into the Xbox and into some third-party devices like the…

Harmon Kardon Invoke speaker

… exactly but isn’t producing its own device. Does your strategic approach to assistants in the home differ from other companies?

Well so, I think we will continue to invest in virtual assistants, both Cortana on Windows, Cortana in these other devices. They are one of the dimensions that are exciting – but not the only one. We are going to see conversational bots everywhere.

For example, Unilever has created a Unibot. That’s a meta-bot that then federates to a lot of sub-bots for specialised tasks. It can be HR, or it can be you know, servicing, or any of those things. So, you are actually going to see companies create federations of bots even, that actually complete tasks and certain bots are skilled in certain things and so on.

And then one of the other things I should add also, bots are in many ways “front office”. I believe the vast majority of AI will be invisible AI….

[20:09] David

[20:17] Joseph

[22:41] David

[21:34] Joseph

[21:35] David

[21:42] Joseph

Page 7: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

7

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

Machine-to-machine AI?

Correct. Invisible AI that is in the fabric of the economy. Helping prevent fraud in payment systems. Helping behind the scenes with healthcare. Do targeted marketing and ad targeting. And, essentially, helping regulate the entire fabric of the economy. And you’re not going to be directly interacting with it. It’s going to make us safer by doing predictive maintenance. It is going to help electric utilities and telephone utilities optimise their networks. And so, on and so forth. The vast amount of AI. It’s sort of like the iceberg. 90% of the AI will be invisible AI in the fabric of our economy – and making our economy dramatically more efficient.

Fantastic. Let’s talk a bit about AI company strategies. You alluded to this a little bit earlier. Strategically, what does Microsoft aim to achieve in the long term through the provision of cloud AI infrastructure and services, and the open sourcing of AI frameworks and libraries such as Microsoft Cognitive Toolkit? Is your goal to democratise AI, and then monetise the infrastructure and services at the heart of the AI paradigm shift? Or do you think about it a little differently?

Well, Microsoft is first and foremost a software and technology company, and we’re a platform provider. Our strategy is: build the next intelligent cloud, intelligent edge platform and the productivity tools. So, longer term, our strategy will continue to be along that direction, which is: provide the best-in-class services, an intelligent engine, an intelligent cloud, for both software development and for productivity – and empowering everyone. It’s a broad charter.

If we think about paradigm shifts that occur in software: in what ways, if any, do you think the paradigm shift to AI will be different from the shifts that we’ve seen recently to the cloud and mobile computing?

I think the paradigm shift to AI is as big as the shift to the internet. Which means, like for as I said, for the first-time computers are not going to have to be programmed for a whole collection of tasks. And they are going to be able to learn from data. And I think what you will see is, very rapidly, ambient intelligence coming in. You know, everything around you, every device, every software – you know, lighting in a room, all of that – will be able to use data to be better. And the environment that you work – you know, humans shape their environment, and AI will be the agent that humans use to adapt the environment of the planet.

Fantastic.

For at least fifty years, technology has accelerated cycles of creative destruction and has reduced the length of time for which large companies tend to maintain value. I wonder, though, whether we might see a bifurcation in companies’ longevity – and the emergence of a small number of super-competitors in the technology sector who capture and maintain value to a greater extent that has been the case in the past?

They might emerge because of data network effects in AI because technology companies are expanding horizontally – more imaginatively – into emerging technologies like quantum computing, and finally because technologies companies are expanding more forcefully up and down the software stack as well – for example, technology companies developing their own silicon.

Do you think some super-competitors might emerge – or do you think ultimately this cycle will be pretty similar to others?

I think you said that at the beginning itself that the longevities of companies are and half-lives of companies are constantly decreasing. I think as if you have a civilised society if you have a fair

[22:39] David [22:40] Joseph

[23:23] David

[23:51] Joseph

[24:26] David

[24:39] Joseph

[25:28] David

[26:24] Joseph

Page 8: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

8

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

education system, you have fair healthcare and an economy that is properly regulated by democratic government…I think we will; we will be insulated no matter what technology comes in. I think we will have fair competition; we will have incredible creative destruction. I think the world will be a better place and it’s not really about technology. It’s really about how we govern ourselves, and I think AI gives governments, as well, a lot of ability to create fair powerful infrastructure and systems. There are cities tapping into AI for citizens’ projects and making citizens better empowered, healthier and so on and so forth.

So, that said, one of the things that we have seen in the world in general is: in pretty much every sector of the economy, there are a small number of dominant players, whether it be in banking, in oil, or in metallurgy or iron ores or any of those areas, there are a small number of players. That is part of what is the network effect generally is – which is that winners tend to be able to capture economies of scale that allow them to, for a period of time, while that particular sector is still hot, achieve a dominant position. I think you will see that in the world of software, cloud and AI. But I do see those things as being transient.

Right, there isn’t something uniquely different for instance [with AI]…?

I don’t think there is anything uniquely different. Everything, each sector… yeah…sectors of the economy itself are dynamic, and some sectors are hot at one period of time because of structural factors – and they change.

Microsoft itself was challenged by previous paradigm shifts to mobile and cloud computing. But today Microsoft really has become a leader in cloud computing, and the next paradigm shift of AI. I’m curious to ask: from an insider’s point of view, what did you see that were the keys to turning around a company the size and complexity of Microsoft?

I think really it is… A company is an organism, and the right leadership makes a huge difference. I think having the right culture…at the end of the day, look, we are all knowledge workers. What matters the most is creativity. What matters the most is being able to create value for customers – creating a customer-centric culture, focusing on unleashing the creativity of the employees that were already there, ensuring we are great partners, ensuring that we are open, that we are a learning company. Putting all of those things in place, which our leadership was able to do very effectively – I think that is what helped us become competitive again and those are fundamental principles.

In many ways, while we talk about platforms, there is also a concept of management platform that you have to think about. A platform that has the right ingredients in terms of management structures, culture, orientation and so on. That no matter what changes, allows you to innovate and stay competitive and create new products and services of high value in many, many, many areas. That’s what some of the leading companies like Microsoft have been able to do.

We’re still very early in the paradigm shift to AI. But what do you think is the biggest disruptor beyond AI? What’s the biggest opportunity and threat do you think to Microsoft in ten years’ time?

One of the areas that we have been pursuing very aggressively in a scientific endeavour is quantum computing. I think…imagine computers that are able to solve problems that classical computers will take a billion years to solve. And it does not at this point in time, with the state of the technology, look too far, far away – maybe five to ten years – and that is real.

And it’s not just Microsoft. There are multiple companies working in the area of quantum computing. We believe we have a scientific advantage in the technological venture at this point in time, with

[28:07] David

[28:07] Joseph

[28:19] David

[28:41] Joseph

[29:50] David

[30:02] Joseph

Page 9: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

9

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

some of the innovations we’ve made. And then we’re realising that in practice we’ll change the face of software and computing yet again. And machine learning will… and AI will become different again.

In fact, life itself wouldn’t exist only with classical physics, right. For example, certain phenomena like photosynthesis are believed to be quantum mechanical – not just traditional chemistry. In the same way, certain types of computation and AI will be quantum mechanical in nature. Certain things that can be done will be possible only with quantum computers. That will be an exciting future to look to.

We can do another podcast then!

Let’s talk a little bit about the adoption of AI within the enterprise today. To what extent do CEOs or CTOs understand AI today, first of all?

Well let me say it this way: at no other time in history have so many people understood so little about so much. And that is true not just about AI but computing and our own lives itself. Our lives were much simpler when we were younger. And so, all of these technology waves come so fast; there is a dire need for companies like Microsoft, that reduce these things to pre-built AI or recipes you can follow. Without having to understand it in depth, we take it for granted that we actually don’t understand what a CPU is, or how your mobile phone works or what the connectivity is. We say Wi-Fi, but we don’t understand what Wi-Fi is. We don’t have to; we shouldn’t have to. That’s how technology is, and civilisation itself is built – there are specialised people who build the components for civilised economies. There’ll be lots of specialised component makers for AI; you should be able to use those and build higher order value.

If we envisage a classic adoption curve of innovators, early adopters, the early mainstream, the late mainstream, and laggards – where are we today within the enterprise around AI?

So, there are a few companies that get it really well. If you look at the top five market cap companies in the world. They’re all building at the intersection of three of the most important trends in computing: cloud, data and artificial intelligence. All of their software is at the corner of that cube defined by cloud, data and intelligence. The axis of the three.

So significant numbers of the top companies have used AI in some way or another, perhaps classical AI, but the rest of those companies are soon going to follow.

And we are early in the adoption lifecycle. We are very much in the – for a lot of AI – at the middle, maybe, or the early part of crossing the chasm.

From a practical perspective, what would you recommend to enterprise executives that recognise AI could be important, could be valuable, but aren’t sure how to embrace the technology? What should they do next?

I think, like any of these other technologies, you should avoid the tendency to dismiss these things off-hand. Very often you see a lot of companies being, or rushing, to be fifth in line. And rushing to be fifth in line means you absolutely miss the advantages of the virtuous flywheel it will kick off. And the person who got there first or second in line is going to have a dominant advantage in their industry. That is what technologies do. Technologies create a winner takes all effects. Now it is incredibly important, for folks, for everyone, to lean into these disruptions and understand in their domain what are the likely areas which are likely to have the highest ROI only if AI could do something for you. And then go see if you can make that happen. You need to experiment.

[31:15] David

[31:25] Joseph

[32:31] David

[32:43] Joseph

[33:28] David

[33:40] Joseph

Page 10: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

10

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

And today every company has to be learning, experimenting company. And today you have to experiment with AI. It is the technology wave that is, today, real – and the top companies in the world have shown how AI gives them incredible advantages and create new products and new experiences.

According to Gartner survey, lack of skills within organisations is the greatest impediment to adoption of AI. How can companies address this issue?

I think by using pre-built parts. I think using recipes.

I think using the cloud is another one, by the way. Cloud is the greatest invention since the invention of the steam engine. It simplifies software, like a steam engine, eliminating the need for a horse and buggy. It is removing the barriers to building and deploying entire applications. Because a cloud provides, you know, the cloud turns hardware into software. You never have to buy and provision hardware in a physical data centre space and networking and wiring all of those things. It changed the game. The cloud turns software into services. You don’t license and install software; it just comes to you as a service. Leveraging all of that agility that a cloud provides you is incredibly important for you to be able to compete. And I think leveraging the cloud, leveraging the pre-built parts available in the cloud and then building the applications on top of that is going to be incredibly important for you to have the short turnaround and experiment extremely rapidly. And in many ways, your success as a company, any company, is going to be dependent upon the number of experiments you can run. And that’s the key that you have to realise.

Could I ask you briefly about LinkedIn? So, Microsoft bought LinkedIn for $26B. LinkedIn obviously offers a variety of assets - including its professional user base, the content people post on it, and its information regarding people’s expertise, employment history, and influencers and contacts.

To what extent, if any, was LinkedIn a data play to support Microsoft’s AI ambitions as opposed to more its goals, say, in Customer Relationship Management software? To what extent is LinkedIn a source of data for Microsoft’s broader AI strategy – or is that just a happy side effect?

So, LinkedIn has a very trusted relationship with its users. We have a user agreement that permits only certain uses of the data. It is in the interest of the LinkedIn members. And we hold at Microsoft and at LinkedIn, that user trust as the number one thing. Because if our users don’t trust us and if they are not getting the value first, we are actually not going to be able to have any asset over there at all. And so, I think, this is one of the things at Microsoft in general holds, as opposed to certain competitors whom you know, certain competitors explicitly say, they will use that data, because their monetisation is from an ad business or one of these other things. That is not Microsoft’s primary policy. So, LinkedIn is an incredible asset that helps its members first and foremost. And where that member group gives its permission to use it primarily in their interest, that’s where we start.

How, though, with users’ permission, do you think you can derive value from their data? Is that an AI play?

Take examples. Like, say, you have an office 365 subscription and say Barclays has an Office 365 subscription and you have all of the employees in Barclays. It’s potentially very helpful for you to see, associated with each of those employees, their LinkedIn profile. So, you know exactly what you know. Anyway, people do look up LinkedIn profiles as you know, for business cards. So being able to connect all of that information, help foster social networks in the workplace. Being able to help improve employees’ work lives, there are a lot of applications like that for LinkedIn.

[34:50] David

[34:59] Joseph

[36:19] David

[36:53] Joseph

[37:56] David

[38:02] Joseph

Page 11: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

11

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

Let’s talk a little bit about AI technology. What do you see as the greatest limitations, and challenges, in AI technology today?

I think, technologically, we still have a long way to go in learning like human beings. Human beings learn instantly, they can follow directions. Machine learning and AI today requires a lot of specific data to be able to learn. And it’s not just data – data about the specific thing you want to learn, with the labels and so on. I think this limitation is huge.

While we are talking about the topic of AI…it has still been hard to even create a two-legged walking robot that can navigate every environment. I can’t create a robot that will get me my lunch or get me a coffee. It is still very hard to do. So, I think there…is that again I will keep going back to my original statement, you are to keep it simple and structured – and all AI is dumb, but some AI is useful. So, you’ve got to tailor to the area where AI is really useful. And it is not general human intelligence, where you can apply to every problem. And that is still the primary challenge around AI.

So, it’s interesting: we’re often reminded that training data is critical for AI – that training data is the new oil, and so on. And clearly, good quality training data is extremely valuable.

But I feel, it sounds like you’re describing there, that perhaps the value of proprietary algorithms – or the potential for what proprietary algorithms are under-stated? In some areas, like processing language, lack of data isn’t the issue. It’s actually the algorithms we have. Do you think, in other words, there could be a new wave of capability by perhaps new algorithms that enable us to do useful things with less training data?

Correct. There is a tremendous amount of innovation happening with new algorithms.

But it is not only with new algorithms. It is also capabilities such as what is called transfer learning. Meaning learns in a related domain, where there is the right data available, and carry over that learning to adjacent domains. And so, if you have great data from translating from English to French can you then now make it much easier to translate from English to Spanish, or English to some other language, right. And I think that kind of transfer learning is actually very powerful.

So, new algorithms will definitely help. And also remember, when it comes to data, it is the right data that matters. A small number of bits of right-on data is dramatically more useful than petabytes of useless data. And sometimes people over-index on the value of data. And people say data is the new oil. In my mind data is not the new oil. Data is the new noise, and big data is a big noise. It’s only the right data married to the specific problem that you solve for and the right framing of the problem all coming together that allows the right AI to be built.

You alluded to the problem of transfer knowledge, their transferability. Do you think we can realistically expect a lot of progress there in the medium term, or do you think transferability will really remain quite limited over the next three to four years?

I think in the specific domains like speech, like language, vision, and so on, transfer learning will be very powerful. I think it is already turning out to be very powerful.

I think in any new domain there is a cold start problem. Initially, you have to go gather the data and the labels and frame the problem and solve it once. And once a particular problem is solved it becomes much easier to transfer that ability.

[38:40] David

[38:48] Joseph

[40:00] David

[40:35] Joseph

[41:46] David

[41:58] Joseph

Page 12: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

12

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

Explainability of deep learning algorithms is obviously a well-understood difficulty. In many cases, we know that deep learning algorithms work well – but we just can’t look inside to see the basis of their recommendations. Do you expect this explainability challenge to be addressed anytime soon?

I think explainability is a challenge for all complex systems. It’s not unique to AI. We can’t explain most of the things we interact with. We can’t explain human judgement or even corporate behaviour many times.

There is an illusion of explainability very often. And that is true, and I think AI will be challenged in terms of explaining. But here is what will be possible. It will be testable, and I think that’s an important difference.

Look, most of the medicines in today’s world, they’re not explainable. You don’t know exactly the specific processes by which a vast majority of medicines you consume work and are effective. However, when they launch a new drug, we go through randomised clinical trials. We have a very specific process for qualifying the drug for human use. We have a regulatory authority that looks at it. In areas where AI directly impacts human life in meaningful ways, where we care about many of those things, in particular, I think we will have similar structures, similar regulations, similar guidance. And that’s what will help regulate complex technology – period. And what I think we can do, is use those principles – statistics, to create frameworks for testing AI applications – that provide a high level of transparency into why certain things work the way they do. And that is in my mind the right direction to go scientifically.

So quite a… I guess sort of what philosophers would call an “instrumentalist approach” – you know, software is a tool, if it acts in the way we think it will act, and it does so in a predictable way, and if it’s useful, that’s kind of enough. And your point is that that’s the way most of our tools work today as it is.

And that’s how most software is. The software is not provable, right; there have been attempts in the world of software to create provability for software. It not provable – it’s tested, its debugged, it’s tested, it’s continuously maintained, it’s continuously improved. And that will be the case with AI. AI is software at the end of the day.

Just finally on AI technology. While recent advances in computer vision have been striking, AI’s ability to understand language is still, in reality, I think very limited. We can transcribe and translate speech really well, but our ability to derive meaning from information and solicit related information I think is quite limited. Put another way, my Amazon Alexa can transcribe my speech very well, but it can’t actually perform many functions.

To what extent do you think this will change? I mean, how long will it take before I can ask quite a broad range of questions and can get specific, full answers do you think? I think these services will be rapidly improving. The nature of most of these services is sort of constantly collecting more information, more data; algorithms are improving. I think in specific task areas I think they will become extremely good and useful. However, again, it will never give you the appearance of being truly capable of a human being. Again, I don’t think, aspiring to that is necessarily worth it. But I think in each of these areas, it’s a matter of a small number of years for these to become extremely useful.

Great. I’d like to finish just by raising two points about the social impact of AI. The potential benefits of AI are numerous and fairly well understood. We’re all excited about better healthcare, cheaper transportation, enhanced manufacturing and increased agricultural output. But I’d like to discuss a

[42:24] David

[42:40] Joseph

[44:16] David

[44:34] Joseph

[44:49] David

[45:30] Joseph

[46:11] David

Page 13: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

13

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

couple of societal risks. Regarding job displacement, do you think AI might destroy more jobs than it creates? I think every technology revolution, starting from the luddites, have feared that technology destroys jobs. What technology actually does, is it changes jobs. In the economy, it changes what’s called a production possibility frontier – one of the foundational curves in economics that you learn – which is fundamentally a change in what the economy is capable of producing and creates incredible efficiencies.

So, as a result, in my opinion, what is going to be happening is, sevenbillion people on the planet, we’re going to find you the economic opportunity because productive capacity of the economy is so much more enhanced. And therefore, back to good governance, if you have great, democratic governments, which are really at the end of the day, governments for the people, then governments are going to be able to create economic opportunity for all that actually make the world a better place. And then it’s really about how we govern ourselves and lessabout the technology itself.

Do you think it might be different this time if this technology is enabling us to automate some more repetitive tasks?

But that automation is power, right? It allows you to... it amplifies human power. Automation augments you. Automation allows you to share that time and creativity with the nerve cells which you only have. And you know, by the way, no AI is creative, creativity is not what AI is born with. And so, it allows you to tap into creativity and instead of being engrossed in drudgery and tasks that have low economic value, to create new economic value. And so, I’m very optimistic about it.

Let’s talk about bias. AI has the potential, rather excitingly, to free decision-making from human biases by making better, more objective decisions. But AI algorithms, of course, are trained on historical datasets that reflect the historical prejudices we have - particularly regarding race and gender. How can we prevent AI reinforcing historical biases?

By testing! The answer is by testing. Just like the medical analogy. When you develop a new drug, you are going to take it through a collection of steps to validate that particular drug. Randomised clinical trials are, for example, a standard in drug testing. The same way, where AI really impacts human lives and decision making, it has to be tested, it has to be debugged. You have to have the testing framework for it. And I think that’s part of the technology that we all have to develop together.

So, effectively, good governance frameworks for good QA?

And even scientific principles for testing, right? Statistics, even, there’s a very sound methodology for testing statistical systems and black box systems and so on. And I think all those technologies apply.

I’ll finish with a quick-fire round, if I may! Six questions, maybe just one to two-word answers for each.

Firstly, is the promise of AI over-hyped?

It’s now in the plateau of reality.

[46:38] Joseph

[47:45] David

[47:54] Joseph

[48:28] David

[48:53] Joseph

[49:25] David

[49:28] Joseph

[49:41] David

[49:51] Joseph

Page 14: | Podcast Interview Ð Microsoft ArtiÞcial Intelligence · 2018-03-12 · Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft order fulfilment systems, and

14

Beyond The Hype: Artificial Intelligence | Podcast Interview – Microsoft

In which sector will AI have the most profound impact?

Any internet businesses.

Do you think AI will destroy more jobs than it creates?

No.

Should we worry a lot about autonomous weapon systems?

No.

Will we achieve the AI singularity, where general AI triggers a period of unprecedented technological change and If so, when?

Not in my lifetime.

And finally, should AI systems of sufficient intelligence have rights?

The sufficient intelligence is going to be so far beyond what I can imagine that I don’t think that will be a problem we face.

Joseph, this has been a great pleasure. Thank for your expertise.

Thanks very much.

We hope you’ve enjoyed this episode of MMC Ventures’ “Beyond The Hype” podcast, presented in association with Barclays.

Follow up on Twitter @MMC_Ventures and explore our research at mmcventures.com.

Don’t miss our next episode, Healing Healthcare, where leading entrepreneur Ali Parsa, CEO of Babylon Health, describes how AI will transform the delivery of healthcare worldwide.

[49:54] David

[49:58] Joseph

[50:00] David

[50:03] Joseph

[50:04] David

[50:07] Joseph

[50:08] David

[50:15] Joseph

[50:17] David

[50:21] Joseph

[50:28] David

[50:28] Joseph

[50:31] David

Founded in 2000, MMC Ventures is a research-driven venture capital firm investing in high-growth technology businesses. MMC has backed more than 50 enterprise software and consumer internet companies that have the potential to change the future of financial services, the workplace and retail.

MMC invests on behalf of private and institutional investors. MMC has over £200 million under management and is investing approximately £35 million annually.

MMC’s portfolio includes: Appear Here, Bloom & Wild, CloudSense, DigitalGenius, Elder, Gousto, Interactive Investor, Masabi, NewVoiceMedia, Peak, Safeguard, Senseye, Signal Media, Sky-Futures, StoryStream and Tyres on the Drive.