The Next Wave with Young Sohn - Technology Megatrends: From AI to Quantum Computing to Cybersecurity

The Next Wave with Young Sohn - Technology Megatrends: From AI to Quantum Computing to Cybersecurity

Show Video

Good morning, Alex, how are you? Good morning, Young, I'm very well it's nice to see you here! So,  tell the audience about your investment fund  ideas. What are some of the things that you're   looking at how big you are and what are some of  your recent investments you get really excited   about? So, Amadeus Capital Partners is a venture  firm here in Cambridge and London in the UK   and we've been around for about 23 years  now. We've made more than 165 investments   in technology companies all over the world and  we focus on deep technology. So, technology with  

a lot of science at the sort of cutting edge of  innovation. The early stage team which I helped to   manage here at Amadeus particularly focuses on  the first couple of financing rounds, so called   Seed and Series A financings which startups need  to get themselves established. You know, we start to   see in the history of Amadeus, our firm, a number  of different what we would now call unicorns, so   you know, billion dollar plus exits from  investments we've made including CSR,   four scout and network computing network security  business and also more recently businesses like   Improbable and Graphcore which have also achieved  this kind of valuation. Very exciting! Some of   the topics that I want to discuss is really  around the key mega trending core technology.  

As you know, 2020 was supposed to be a great year  but it turned out to be a pandemic here. But   when I was coming into 2020 I was thinking that  there are four C's that are influencing our lives.   And the first C was consumer and how the consumer  behavior and experience is really changing the way of  buying things, experiencing products, the way we  communicate. And then the second C was really   around what I call cyber security, as more and  more people are connected in cloud. What does that   mean in terms of their data, privacy, security? And  then the third one is about cloud. It seems   to me that we're all working on the clouds now  and that we're connected and the cloud business is the one   that is really growing to be a much bigger size  than ever and driving our economy. And the last C  

was China, given there is ongoing geo-fencing  that is going on but technology in particular.   How do you work with two different economies in  terms of standardization, in terms of partnership,   in terms of IEP, in terms of access? So these are  the four mega C's that we've been discussing in   the beginning of 2020. Of course, now we have one  more C which is Covid-19. So those are the   five subjects that I want to kind of focus on this  morning. So, start with your view of how do you see   AI impacting our lives and how do you see that AI  is actually delivering in reality? Because I know   you've been doing a lot of early stage investment  in AI and other related deep technology areas. And  

given your background as the one of the earliest  and longest venture capital from out of Cambridge,   Hermann Hauser and others have been building this.  Alex, and you've been running it now   and I just would like to get your perspective of what you see in AI. Sure, so I think the AI environment has been evolving a lot in the last  couple of years and first of all people started   out believing that delivering complete solutions  with AI would be feasible. In practice, I think that   mostly what's really happening is that the  solutions are around using machine learning to   provide some analytics function. But very little  full stack solutions are being delivered. Think   about autonomous vehicles. Many of the autonomous  vehicle businesses are having to reconsider or  

consolidate. And you know that's because finishing  the whole project from top to bottom in the time   scales that were originally envisaged hasn't  proved feasible. On the other hand, machine   learning for data analytics is becoming really a  significant factor in a lot of companies lives and   problems of course that arise from that involving areas like bias detection, algorithm   drift, you know these sorts of things need work  and startups are starting to address them. So   we're finding quite fertile opportunities for  investing by looking at the infrastructure of the AI market where machine learning needs to be  deployed rather than the complete solutions in AI.  

So it's in a way, if you kind of use the football  analogy, we are still in a very early part, first   half, I guess. I think that's right and I think that,  you know, in technology there are many optimists   and we're excited to build the future. But reality  usually intrudes pretty quickly, you know. We   think we're heading for a win just because we have a good start to the match but   actually the reality is it's a long game and many  things will happen along the way. And we have to  

accept that the market is evolving  around us. So tell me about when you   look at the European landscape, the AI companies  that you see, what are some of the interesting   killer apps that you see and what are some  of the companies that you get excited about?   So I think in the particular context of  the current situation with the pandemic,   one of the most important developments has  actually been in the area of Covid-19 detection   and the development of new treatments. It's  very clear that one of the great things which   AI can do is to do things like rapidly discover  genomic markers which indicate when some kinds of   people may be more susceptible to the Covid  virus than other people. And also in the drug   detection area to identify molecules which can be  therapeutic and which can be used to help people   find treatments. Of course not just for  covid but also for diseases more generally.  So we have for example here in Cambridge a great  business called Congenica. Congenica's business  

is to do the genomic analysis of very large  amounts of human data and use that to detect   new kinds of treatments for rare diseases. And  we've also a business that was invested by Amadeus   but it's actually in Valencia in Spain called  QUIBIM. That business does analytics on images of   for example X-rays or MRI scans and it became  very important quite quickly in the pandemic   to help doctors detect early signs of the Covid  virus in people who were suffering from it, even   before they were able to feel themselves that they  actually were suffering seriously. So I think those   kinds of technologies in the healthcare market are  really attractive and you know good evidence of   the opportunity that AI holds to improve the  quality of all of our lives. So, Alex, I think   one of the interesting things about the emerging  applications and others of which clearly benefit   humankind is about the how you train the data  and using those data you can discriminate   and being able to filter out what is possible,  what isn't and it seems to me those things are now even   better than what humans can detect. Because machines  can detect each things with much better accuracy than   humans of what I understand in imaging applications.  So I'd like to get your perspective where are we?  

So I think you're absolutely right. I think  one of the great attractions of using AI is that   it's capable of sifting through incredibly large  volumes of data, much larger than humans would be   able to do. And then to give support to humans to  help make important decisions. If we think about   the healthcare applications, for example screening  lung X-rays to detect early signs of diseases   like Covid, the important thing is to support  the radiologist and to help the radiologist   to pay particular attention, perhaps to some area  of the X-ray or to think more seriously about an   X-ray which looks borderline to a human, but which  the algorithm can detect is actually showing   serious signs of disease. So in this way the AI  is supportive of the expertise of the human and   at the same time it's able to do much more  analysis than the human could feasibly do   in a short period of time. The thing that we  also should think about is who gets all the   data, right? Even the bigger company, the bigger you  are the more data you can have and they'll become   a better model and then therefore it actually has  an advantage for platform players. So do you see  

a breakthrough behind just data-driven learning  to other ways to make it more efficient?   So I think the important thing to be  aware of is that new machine learning   algorithms are emerging which actually work  on significantly smaller amounts of data.   At the moment, the history has largely been that  more data is better and greater accuracy comes   from greater amounts of data. And, of course, first  of all that potentially deprives some users of the   opportunity because they don't have access to the  data. But it also actually is very inefficient, it's   not very satisfactory. From a sustainability point  of view, data centers have become notorious in, you  

know, consuming huge amounts of energy. So these  new algorithms are emerging which allow effective   machine learning to occur on much smaller amounts  of data and there's also an interesting trend   in how the mass is actually done. The idea of  binarized neural networks, so actually using neural   networks where only one bit of information is  being processed instead of much wider 16 or 32-bit   numbers, which have been the tradition, this can  increase the efficiency of these algorithms and so   both the requirement for data and the efficiency  of the algorithms are actually improving. And these   trends, I think, will continue and make this kind of  machine learning accessible to a much wider base   of users. So it's not just a big large database  application but it could even apply to edge   computing where you need a smaller set of data and  being able to also run it with much less compute   intensive by narrowing the bid you can be able to  run a much more power efficient way of running the   AI applications. So that's really interesting, I'm  sure you're looking at some exciting startups in  

that area. For entrepreneurs out there: If you have  great ideas around that, contact Alex. I'm sure   he'll be excited to talk to you about this.  One more question around this, it's about AI bias.   So as we entrust more and more of our tests  on AI, such as screening bank loan applications   or rating job applications or whatever the  data sorting through AI. We can see more and   more risk from biases and AI models. How are  the setups or others addressing these issues?   This is really important because, you know, clearly the tradition in computing, there's a good   phrase to describe the situation "Garbage In  Garbage Out", if the data which you're feeding   machine learning algorithms doesn't accurately  represent all of the people that you're going to   need to ultimately study when you're making  inferences from the data, then the machine   learning algorithm is going to deliver the wrong  answer. So it's really important first of all to  

curate the data which is being provided in the  first place. The second critical thing is to have   techniques available to detect when bias or drift  in those algorithms so that means the algorithm   is becoming less accurate over time. Then, if  either of those things are happening, we need to know.   There are a number of startups which are offering  this kind of analytic capability to monitor the   quality of the work carried out by the algorithm.  We've invested in a business called Seldon  

which has an open source stack widely used  by people for deploying machine learning   algorithms and Seldon's alibi tool allows  people to understand how was a decision made   by a machine learning algorithm, what was the basis  for the inference. And that helps the operators   of that algorithm to develop a clear steer on  whether or not there is some kind of bias or drift   occurring and as a result to take the appropriate  measures to prevent that from being a problem.   Here is a very interesting area it's probably too  early to say but it's something that we have to   watch carefully because clearly we've been  living with a lot of bias for a long time. What   we don't want to do is AI and machine learning  and others accelerate even more to that direction   because the data the training is actually feeding  that into our sidecar. So I'm glad that you're   looking at that as a part of your thinking. How  do you make sure the bias is not built in in our   thinking and our data and our AI algorithm? So let  me change it a little bit to now quantum compute.  

I know Hermann called me about five years ago  talked about quantum computing so it's been a   while that we've been thinking on this concept. So  what is the state of quantum startups in Europe?   And what are the European  unions as well as the institutions that are   working on this issue, are they collaborating,  are they competing, what's the state of union?   So the European quantum scene is very fertile  and what's exciting about it is that there are so   many different quantum technology application  areas, so everything from quantum photonics, quantum   sensing, right through to quantum computing is  delivering really interesting capabilities which   were simply impossible in previous generations of  technologies. In the quantum photonics area being   able to make sensors which are accurate to detect  even as little as a single photon of light can   enable imaging with a level of detail which has  never been possible before. And it's also possible   to build secure communications links using quantum  photonics with single photon emitters and single   photon detectors, we can essentially make a optical  communication link which will be uninterceptable   without us knowing that somebody has interfered  with it. And that kind of security is extremely   valuable. In the quantum computing area computing  is really at an early stage. There are six or   eight different technologies which exploit  quantum properties to carry out computing and   as, you know very well, Young, some of the world  leading work in this area is delivering only maybe   50 qubits or so of capability right now. So it's  demonstrating the possibility of quantum computing  

but it's not really yet delivering functional  computers which are capable of working on   the hard problems in the real world. I think that's  the trend which is going to change over the course   of the next, probably only two to three years. We're  going to start to see quantum computing scaling up   into the hundreds then thousands and tens of  thousands of qubits. I think most people would   agree that when we can get to one million qubits  it will start to be possible to carry out really   meaningful work with these kinds of quantum  computers. So that's probably five years away  

but it's actually a horizon which is collapsing  pretty quickly. So, you know, just kind of give   background to readers of our audience of our  program, clearly the benefit of having large number   of qubit is about being able to solve problems  simultaneously. So think about linear algebra when   you have multiple values variables and you want  to solve the same time by having this compute   requirements in classical computing, clearly we're  using zero and ones to identify and be able to   manage to take a huge number of iterations. But if  you have, imagine it's not zero to one, now you have  

many many numbers between zero to one, so you can  be able to use all those things at the same time   and deploy that and being able to solve very  complex linear equations - I try to simplify it   for the readers - but I think that is the benefit  of having a large number of quantum bits, I guess.   Absolutely. This ability of the quantum computer to  be able to perform effectively the equivalent of   many millions of simultaneous calculations which  would be required on a classical computer, means   that the ability of the quantum computer to  solve problems which were almost unsolvable   in classical computers, is really a huge step  forward. And so this has applications in areas   such as cryptography, in breaking codes but also  in things like modeling the way proteins work   inside human cells. So helping us to discover new  drugs or identifying the chemical reactions that  

catalysts go through to accelerate various kinds  of industrial processes, enabling us to develop   new materials more rapidly than would be possible  without this kind of computing capability. Great, so   what it does is being able to solve  very large variables very quickly   with lower power, higher accuracy and these  are some of the advantages over class compute.   Obviously, quantum compute isn't for  everything, right? There are certain things   class compute will still exist, still go on but  quantum compute can be a complementary to the   traditional compute that we've been living with.  Let's change the subject a little bit to security.   If you look at today, the number of issues that we  are dealing with with cyber security is increasing.  

As more and more people are working at home,  the number of instances is now at a record, the number   of ransomware has been going up and they are even  going after mission critical space like hospitals   and schools and others. So what is your view of  the whole cyber security space? Thank you. So,   certainly, the security environment has been  evolving fast in the last few years. As you   mentioned, I've worked myself in the cyber security  area for many many years. And what I've seen   most recently is that the traditional model where  we tried to keep the bad guys out of our castle   by building a really high wall around the outside  of our organization, this model just doesn't work   anymore. Nowadays, we're all carrying mobile devices  with us, we're all connecting from home into the  

office and so, actually, the wall is already  full of holes and is starting to be torn down.   So we can't have a defensive model for security  any longer, which thinks about just keeping the bad   guys on one side of a wall. Really, the significant  challenge here is all about people being trained   to detect when they are being offered, maybe  an email which is a phishing attack, or   somebody is trying to persuade them to give up  their password details by talking to them over   the phone. These kinds of human connections  are actually where most significant security  

breaches happen. There's a person in the loop,  they click on a link or they answer a question,   when actually, if they looked really carefully, they  could probably figure out they shouldn't do that.   So the challenge is to help people to behave  better, to help to first of all train them, also   to have detection on the network. So we know  when breaches occur it's no longer a question of   believing that we're going to keep the bad guys  out completely. We actually really have to accept   that breaches of networks will occur, so we want to  detect them early and we also want to have tools   in place that help us to slow down or stop those  attacks after they've occurred. So that's really  

where the focus of a lot of investment activity  is now occurring in the cyber security area.   So, if you think about cyber security analogous to,  let's say a virus in a way that we cannot avoid the  virus. If we're going to have it, it's a question  of how are we prepared, how do you detect it   and how do you response to that. I guess, it's probably  what you're saying. Because virus is just not going   to be able to protect it. Always unseen, difficult  to predict and very costly to creating ongoing  

vaccines around you all the time, that you  don't even know what's going to come at you.   Right. So, just as people really can't avoid disease,  it's not practical to live inside a bubble all the   time. So we have to have an immune system and  we have to have an ability to detect when a  

foreign virus or bacteria has entered our body and  then to fight it off. And so this is the function   of the various mechanisms inside people which  help us to defend ourselves against disease.   And so, as a metaphor, as a way to think about  cyber security, it's quite helpful to think about   the enterprise security in the same way to have  both systems which detect when an infection occurs   and also to have systems which fight it off.  But we do have to accept that we are going to  

sometimes get sick and so it's necessary to plan  for that. And so some of the other implications   of that are things like taking good care of the  critical information inside your organization,   so you actually keep it off the network in places  which are better defended when it has really high   importance. And you don't allow duplicate copies  of sensitive personal information about people   for instance or corporate intellectual property  to be lying around on the network and able to   potentially be picked up if a breach does occur.  That kind of good hygiene is actually just common  

sense in making sure your enterprise is less  vulnerable. So virus and knowledge is pretty   good, about having a good hygiene, being able to  have the right response and being able to attack   the virus. Once you detect it with your antibody  system, they can be able to do autoimmune response.   All those things actually do have a lot  of analogies to our human body.   Tell me, in your view, what are some of the exciting  companies in cyber security that you've seen?   So, I think that the kind of technologies we've  been discussing here, which allow us to detect   when those attacks or infections  occur, they're really important.   We invested in an exciting business called Sension  which has the ability to detect when traffic on   the network contains suspicious behavior and  flag it up to the IT security team. So they can  

take actions against it. Another investment is a  business called Exonar. Exonar helps us to classify   all of the data on a corporate network  and figure out when it's a duplicate of   sensitive information we probably shouldn't  keep lots of copies of, so help us to focus the   protection of for example encrypting files  to only really do that where necessary for   particularly sensitive information. Because it's  cumbersome and difficult to use encrypted files,   most people will recognize the annoyance of having  to type a password in to unlock a document they   want to read, it's a nuisance to do that, you  only want to do it for those documents which   really really need it. But figuring out which  documents those are and how many copies of this   critical information you have in your enterprise  actually turns out to be pretty hard. So businesses   like Exonar help to actually make that a reality  by classifying the information first and then   enabling people to decide which information is the  most important and protect it to a higher standard.  

Great, well thank you for sharing some of  your great companies out there. Well, I'm   so happy that we can have a conversation  around this key technology mega trends that   are impacting all of our lives and I really  appreciate all the help you have given to   me and also my team. I look forward to continuing  our journey and have great holidays! Thanks   very much, Young. It's nice to chat with  you here and we appreciate your support.   Wishing you healthy and happy holidays and  a much better 2021. Absolutely. Thank you!

2021-01-21 13:12

Show Video

Other news