Making Data Simple: Tanmay Bakshi Podcast
Gain a fresh perspective on data and AI in the healthcare industry from Tanmay Bakshi. Recorded live at the IBM Toronto Lab in Canada, Bakshi and host Al Martin discuss the impact of data in healthcare, including mental healthcare for youth. They also chat about how to help people communicate through the use of brain sensors, machine learning and deep data analysis. Stay tuned to find out what’s next for Bakshi. (Spoiler: there is practically no limit to what this 14-year-old data genius can do with ML and AI).
00.30 Connect and tweet to @AskIBMAnalytics
00.35 Subscribe and rate this podcast on iTunes under Analytics Insights
00.40 Subscribe and listen to other Making Data Simple podcasts on the IBM Big Data and Analytics Hub
02.00 Learn more about the IBM Toronto Lab
02.15 Want to know more about Db2? Click here.
02.30 Pick up Hello Swift! iOS app programming for kids and other beginners by Tanmay Bakshi here.
02.30 Learn more about Fox Pro here.
02.40 Learn more about Tanmay’s app, T-Tables.
03.39 Watch and learn about IBM Watson playing jeopardy here.
05.10 IBM Interconnect 2016.
09.15 Learn more about IBM Watson Initiatives here.
10.20 Read Hitchhikers Guide to the Galaxy by Douglas Adams.
16.15 Learn more about how Apple collects it’s data here.
17.10 Want to know more about suicide prevention and resources for teens and adults struggling with suicide? Cick here. you can also contact your local suicide prevention association.
21.25 What is EEG? Click here to find out more.
27.00 Read Homo Deus: A Brief History of Tomorrow by Yuval Noah Harari.
27.25 Read more about the 2013 Study, The Future of Employment: How susceptible are jobs to computerization by Carl Frey and Michael Osborne here.
27.25 Connect with Michael Osborne on Twitter.
30.25 Watson Olli bus sound like a cool invention? Learn more about it here.
31.24 Learn more about what Watson is doing with Oncology.
48.00 Pixel for Pixel face swapping
51.33 Connect with Jeremy Snape on Twitter
51.40 Learn more about the Watson Summit in Denmark
53.20 Learn more about what the University of Toronto is doing in tech and research.
53.31 Learn more about what Ryerson University is doing in tech and research.
53.23 Learn more about what the University of Stanford is doing in tech and research.
53.24 Learn more about what MIT is doing in tech and research.
53.25 Learn more about what the University of California is doing in tech and research.
53.26 Learn more about what the Berkley is doing in tech and research.
Ready to dig deeper? Check out our previous podcast episodes of Making Data Simple.
Fatima Sirhindi: Hello, this is Fatima Sirhindi, producer of your favorite podcast, Making Data Simple. Just a heads up, this podcast was recorded live at the Toronto Lab on March 1 and any references to dates are from that time. Also, because of the live recording, our audio may sound a bit different. The live recording is available on our Facebook page @IBMAnalytics and will be available in the show notes.
Tweet us ideas of what you want to hear @AskIBMAnalytics and don’t forget to subscribe to us on iTunes under Analytics Insights or on the IBM Big Data and Analytics Hub. The link to that will also be in the show notes. From there you can listen to all our previous podcasts including last weeks on data and retail. Join us next week and thank you all for listening and enjoy.
Al Martin: HI, my name is Al Martin, I'm the vice president of hybrid data management in IBM analytics and I'm the host of the Making Data Simple podcast, which is on the IBM Big Data Hub as well as iTunes, under the analytics insights podcast. Today's special. We were doing this to live via webcast on Facebook. The title is “Mastering Big Data, Machine Learning and Watson” with Tanmay Bakshi. All right, you are the big celebrity here.
Tanmay Bakshi: Thank you, glad to be here.
Al Martin: And Tanmay is in, we are in the IBM Toronto Lab. And this is also a special occasion for the folks in the Toronto Lab, it is the 25th anniversary for the DB2 distribution so a little round of applause for everybody here.
So you didn’t know you are a part of a celebration.
Tanmay Bakshi: No I did not.
Al Martin: You're the special guest. So I'm going to give you a proper introduction here. Okay. You are a Keynote and TED Speaker, algorithmist, author, IBM champion and Honorary Cloud Advisor, YouTuber.
Your current goal to inspire and train 100,000 kids and novice developers through your books, YouTube tutorials and you're a thought leader on big data machine learning on IBM Watson. Did you know that?
Tanmay Bakshi: Yeah I think so.
Al Martin: That's the short version. I went with the short version. Hopefully I got it all right. Did I get it all?
Tanmay Bakshi: Absolutely.
Al Martin: So I thought what we'd do to start is I would get the elephant out of the room in that I was thinking when I became a developer, when I first started coding and that was, you know, when I had to get my priorities straight and I had to figure out what career I was going to choose. And I went into electrical engineering. And I quickly found out at age 19, that I liked software development. When did you get your priorities straight?
Tanmay Bakshi: I mean initially I started coding when I was five. My dad used to work as a computer programmer so as you can imagine, as a five-year-old watching my dad program almost all day was so fascinating to me. It would be like magic. I would look at the computer and it would display my name on the screen, adding numbers, really doing whatever. I wanted to find out how that worked in the back end and how the computer would display my name on the screen or really do whatever else. So my dad saw that curiosity and the questions I had about computers and introduced me to very simple programming languages like Fox Pro and eventually, my interest grew.
I got more and more curious, because of that I started using more learning resources, books, the internet, visual basic and in fact at 9-years-old I even had my very first iOS app T-Tables into the app store. So around 9 years old I continued to develop iOS apps but then eventually I started to lose my interest in programming. I didn’t really find programming as fun anymore because I just thought computers are very rigid, very literal. The moment you start programming something in, it becomes obsolete.
However, when I was 11, it all changed because I stumbled across IBM Watson. And so I stumbled across a documentary on how IBM Watson played “Jeopardy” and won against two of the best human competitors and it immediately peaked my interest and I got hooked to machine learning and AI. My faith in technology was reinstated, and in fact after I had done some work with IBM Watson, I had my very first major keynote at IBM Interconnect in 2016. And around June 2016, I had some summer training in this lab that was very cool so that lasted for quite a few weeks.
Learned a lot here, and ever since then I been really into artificial intelligence and how to design your own neural networks and integrate systems like IBM Watson, and all the other services on the IBM Cloud.
Al Martin: Wow, I think I've got to get to work. Fantastic. So here's the good news we got a lot in common because, how old are you now?
Tanmay Bakshi: 14!
Al Martin: I feel 14. We’ve got a lot of energy, and I probably get called an adolescent probably twice a week.
Most of the time I think that’s in a good light I think. So I went through a lot of the accomplishments that you had outlined there. But here's my first question for you. How do you describe yourself? I mean what do you want to be known as? What do you - how do you describe yourself?
Tanmay Bakshi: Well, mainly, I would say that I love to develop with technology. More specifically, intelligent technology. Algorithms that allow machines to think for themselves and understand abstract concepts and data that really up until now if you think about it, only we humans were known to be able to understand.
That is mainly what I love to do. But once I develop these algorithms and work with them, I love to take them and share them back with the rest of the developer community and make it easier for them to develop these kinds of applications as well. Because then what is the point if I create a cognitive application and no other developer can actually benefit from that. That is actually where my other goal originates from that you mentioned, the goal to reach out to at least 100,000 aspiring beginners. In fact, I am already around 7,400 there and of course always work towards that goal.
Al Martin: So for the folks in this room for them not to know some of your accomplishments, they must be living under a rock at this point in time. But my question would be: do you have a favorite of what you’ve done that stands out?
Tanmay Bakshi: Well, it would be hard to choose a favorite. I love to do all this type of stuff. I love developing AI although that is the number one thing that I like to do, I couldn’t say it’s my favorite though, I love sharing that back with the rest of the developer community. I love doing things like for example this podcast, the keynotes talk, I love all this kind of stuff.
But really what I really don’t like though, I wouldn’t be able to say that but what I love is developing technology and bringing that to the hands of developers especially using tools like, Watson and all the other services that IBM provides. I am able to take that and create really simple tutorials as to how these technologies actually work and allow more developers to work with those technologies. Again, Artificial Intelligence, Machine Learning, Big Data, these are powerful technologies but unless we have the developers to work with them. What’s the point?
So that is what I love, is that I can create that technology and bring that into the hands of even more developers.
Al Martin: Sounds like it's a process getting there.
Tanmay Bakshi: yes exactly.
Al Martin: So I've got a couple of questions. What I thought we'd do here is I'll talk to some questions about yourself; then we'll talk about technology, talk about big data, machine learning, Watson. We'll bring it back to yourself. And if we have time we'll put it out to questions in the room as well. But so here's the thing, the toughest question of the night. Why do you think so many people are intrigued by your story?
Tanmay Bakshi: Well, I wouldn’t necessarily say that my life is a story.
I mean you could say that really, I love all the stuff that I am doing. I am really passionate about the programming and the technology that I am using. What I love to do is again, also spread awareness of these kinds of technologies and what I love to do.
What I like to do is say, for example if I can develop cognitive applications with IBM Watson, I can do all these kinds of things, why shouldn’t more developers, in fact you start developing with these technologies? Of course, not only bring them to the hands of developers but then also into the hands of your users. So, what I would say is that I am able to take what I learn about these technologies, whatever it may be, really simplify that and make it available as a small package you could say. I am really giving it to as many people as possible so they can take advantage of that and implement into their applications or their systems or whatever else it maybe. It’s really not only AI, it’s really anything that relates to this kind of topics.
So, again I am really passionate about that, I really, really love doing it and I would think that would be why.
Al Martin: I can tell, you’ve got a lot of energy, this is great.
So, you know, Kate Nichols who's actually in the audience today who leads the podcast with Fatima Sirhindi, hi Fatima. And Kate mentioned something to me the other day that's kind of interesting. And it was a quote that she got from Douglas Adams who wrote I think, Hitchhiker's Guide to the Galaxy.
Anyway, he had stated that there's a set of rules that describe our reaction to technologies. And he puts people in three buckets. First bucket is anything in the world that you're born with is more normal and ordinary.
Number 2 was anything that's invented between when you're 15 to 35 new and exciting. And Number 3 was anything that's invented after you're 35 against the natural order of things.
So I see you in the first bucket right now. But the whole thing that caught me that was interesting about that to me is I do believe we're in that (non-precedent) period of disruptions.
However, I think we've been in periods of disruption previously. But I think we're also democratizing technologies which is different today. In other words, that it's acceptable at any age whether you're 90, whether you're five; it depends if you're going to embrace it.
And I would just, you know, I was thinking about this. I think one of the reasons people are interested in your story or what you've accomplished so far is you have a unique perspective, a unique point of view. You're an expert today, you know, and you're counted on to be a thought leader tomorrow of course.
And you see a ton of possibilities right now. And I don't think you even hit the other two, doubtful. But I think a lot of people look at, you know, the creative disruption as, you know, painful. Whereas I can say I don't think anything from a technology point scares you right now.
Tanmay Bakshi: No, not at all. Not at all.
Al Martin: You embrace is.
Tanmay Bakshi: Yes.
Al Martin: All right. So let's talk about big data, machine learning, Watson. The podcast that I do is Making Data Simple for a reason. And it's because I'm trying to make data simple and acceptable so I can push that up the stack where machine learning, analytics, augmented, artificial intelligence can take advantage of it simply
…and for the developer, for the data scientists, for or whatever. What do you think of when you think of big data? Let's start there. Okay? You can give me data, big data, whatever.
Tanmay Bakshi: Sure, essentially, what I think of Big Data as is really every second of our lives, as we’re driving, as we’re doing our work, whatever it may be, we are always creating data. We are always generating data. The thing is that with Big Data is that we are able to actually able to capture all that data store it, secure it and from there enable technologies to understand it like for example machine learning.
So that we can for example, let’s say that we track someone throughout their daily life, we can find out what they’re doing as an inefficiency, make their lives more efficient, give them a better schedule. Allow them to do what they already do, in a better way.
Through big data, we enable technologies like machine learning to take advantage of the data were creating. That’s really what I believe and what comes to my mind when I think of big data, is to collect and make sense of all the huge amounts of data that we are creating every single second of our lives.
That’s really what I think when it comes to big data, but then again though, once we have gotten that big data, it’s always a part of being able to understand and that’s where the machine learning comes in. Which is of course, what I am mainly passionate about.
Al Martin: I'm going to get there in just a moment.
Tanmay Bakshi: Alright
Al Martin: So you said it is - this kind of leads into my next question about tracking data. You know, you talk about it - us tracking data every day of our daily lives.
Tanmay Bakshi: Yes.
Al Martin: The more data that we're able to aggregate typically -- you know, we'll go into machine learning in a minute -- the better we're able to, you know, assess, you know, a medical solution, predict the weather. That data's important.
Where do you see, you know, from your standpoint the line of distinction between how to keep data private and those that say I don't care about data to the fact that more data we get the better help, you know, the better health, you know, we are going to be able to provide etcetera.
Tanmay Bakshi: Exactly. So you see Big Data is important, without being able to collect, terabytes, petabytes of data, it’s impossible to actually make sense of it all because again, computers have no sense of the world, all they can do is manipulate data. But if we give them enough data and the right algorithms, they can manipulate data in a way that makes them seem like they’re intelligent and they can do things only we were able to do.
However, where is the distinction between maybe this is data we shouldn’t collect or these are people that want to give their data or these are people that don’t want to give their data and we have to collect this data? So now again, since data collection is so important, I think that should be done entirely transparently.
You should be gaining the trust of your users before you gather their data. Like if a user trusts you with their data, they will be more open to giving you that data, and more open to you actually using that data for the AI, machine learning or whatever else you may want to use it for. There is a very good example, let’s start with a very small example.
If you have an IOS device, any IOS device, with I believe IOS 11 devices. When you type a text and you go over to the emoji keyboard, it actually highlights the words it thinks you can convert to emoji’s, how does it do that?
It collects data from everyone who is typing on their devices and whenever they replace a text with an emoji it collects that data, trains the neural network services and brings it down to all of your devices and now you can replace words intelligently and it actually adapts to new kinds of trends or whatever else in this. It knows all of that because it’s able to continuously learn. But again though, it always is depending on the users and how much they trust Apple with their SMS data. If the users don’t trust Apple getting their SMS data, they’re going to turn that off. They’re not going to allow able to access that data. But because the users trust Apple with that data, they can actually collect all that and use that to their advantage.
In this case, users advantage. Getting a little bit bigger, let’s talk about Tesla, 16.30, of course, as you drive a Tesla car, it is definitely collecting all different kinds of data from all different kinds of sensors and it’s trying to make sense of all that data and its actually sending it back over to Tesla. In turn, Tesla keeps fine tuning their neural network models and updating their data and it creates a better autopilot experience for everyone including yourself.
What I believe, actually another example, I am actually working on a project. This project deals with mental health. Of course, there are a lot of different initiatives. Trying to boost suicide prevention, it’s very important but the only problem is that by the time we get to the suicide prevention stage, it’s almost always too late.
And so, what we need to do is somehow do depression prevention in the first place, which is what I am trying to do through Machine Learning and Artificial Intelligence. Creating an early warning system for depression in teens. Now the system actually takes advantage of all the data this teen is generating. This includes GPS to see where the teen is going, SMS, Facebook, Twitter, browser history, everything you can possibly imagine. Of course, the users are not going to be comfortable with me taking that data and setting it up over the Cloud. So in this case, since we are dealing with so much and such sensitive data, I do all my machine learning on device unless the user specifically opts in to sending their data in the Cloud to create better machine learning models.
So, really what I believe here is that if we want to collect huge amounts of data for huge benefits for example this mental health project or Tesla’s self-driving cars or cancer treatment all these sorts of things, it needs to be done in a transparent way. The users need to know what kind of data is being collected and we need to make sure that this data remains as secure as we can possibly get it. To gain the trust that the users have in them.
Al Martin: I think it's also the younger you are, the more opt you are to share data, do you think that’s the case?
Tanmay Bakshi: Might be, because you see if you think about it, as a teenager, generally even, you’re talking about a human if generally isn’t too comfortable for you, you really think a teen would be like, okay I’m depressed, I should call this help line, of course these help lines really do help but do you really think that they take the initiative to call and actually get help?
So really what I believe is through machine learning, your data is not being looked up by anyone, in fact it’s just a machine learning model going through it locally trying to understand it on a device, you could even turn off the internet and it would still work. I think telling them that works and telling them that no other human even if its anonymous, no other human will be looking at that data no matter what. I think that really captures their trust and that’s what allows us to be able to actually work with this project.
Al Martin: So essentially, it’s about trust in your mind?
Tanmay Bakshi: Yes that is one of the main things yes.
Al Martin: How many initiatives are you working on at any given time, we talked about like three before and now you got another one, that’s amazing.
Tanmay Bakshi: Well, there are three, I mean there are a few main ones that I am working on including the mental health project that we just talked about. Also another project called The Cognitive Story, in fact the point of the Cognitive Story is to really generally augment people’s lives using the power of AI. The first goal that we are working towards is helping those with special needs and the first person we are trying to help is actually a quadriplegic girl named Boo.
She lives just north of Toronto, and as I mentioned she is a quadriplegic and so she cannot communicate whatsoever, she cannot move she cannot communicate so she is thirsty, hungry, uncomfortable. She can’t communicate any of those intents. So what we’re trying to do here and what is sort of my role is to use artificial neural networks to understand the electro brain waves that we can capture from Boo using a custom 3D printed headset. Then using those brain waves trying to say okay are there some patterns that indicate when she’s uncomfortable or when she wants to say yes or no. And then of course using those patterns into allowing her to communicate.
So I have quite a few projects but these are the main ones, in the field of healthcare and then one more project that I’m working on is taking what I learn and sharing that in my videos, YouTube channel, books I author things like that.
Al Martin: So we're jumping around a little bit but I want to hear a little bit more about the Rhett syndrome, where were you on that?
Tanmay Bakshi: Well, I mean that project is going along pretty nicely. That project is still in the research phase of course, in fact actually one of the team members, Ross who is actually in charge of all the hardware, the IoT behind the project, he's in Ecosystem, which is an IBM partner. He was in Toronto for the past three days, so first taking us here, we actually drove up to Boo’s house, it’s about a one-and-a-half-hour drive and we put a prototype on her head and collected some EEG data from her.
By using that EEG data, we are going to try and find patterns in that data, however the second day we did collect some data from Ross and what we are trying to do is find out how exactly these patterns look and how we can find those patterns in the first place. Once we find that we are going to apply that algorithm to Boo’s data and that is sort we’re hoping for.
In fact, there are going to be a few more updates at THINK, we have a session there including the headset that we have developed, the neural network architectures and a lot more.
Al Martin: Can you direct - you described a little bit of the headset earlier? I thought that was interesting because we made advancements in the headset.
Tanmay Bakshi: Oh yes.
Al Martin: Say a few words on that.
Tanmay Bakshi: Oh actually - sure.
So with the cognitive story, as you can imagine, we are dealing with extremely complex data. We are dealing with data that even for example the, noise coming out of that electricity socket on the wall right there will affect the reading that we get from Boo’s head. So we need to filter out all of that noise manually. It is extremely sensitive.
We are dealing with micro vaults and we are dealing with this on a scale of microseconds. We are actually gathering this data on 250 brain waves per second and then to add the cherry on top to file the complexity even more, we are not even dealing with regular EEG, we are dealing with EEG that is dry. We are not putting in any sort of Gel here or, she doesn’t need to shave her head, it works with her hair. Of course, we are using 8 channels, 8 electro magnets on Boo’s head. This adds to the complexity because usually you are working with 64 or 96 channels. You are working with a headset that has gel on it. You are working with a headset that is actually professionally wired through a system. This is also wireless.
You can imagine, a lot of complexity going through just the headset part and then using those 8 channels and using a deep neural network to actually understand it on device in real time. Quite difficult but we are always working on it and of course always improving the technology.
Al Martin: Yes. This is where technology can change the world. I mean I'm with you. What - let me step back and talk a little bit about the machine learning
Tanmay Bakshi: Yes.
Al Martin: When we were talking prior I asked you hey, what's your favorite piece of technology right now, and you said machine learning.
Tanmay Bakshi: Yes.
Al Martin: So I guess my - it's a two-part question. Remember we are through the podcast. So what is your definition of machine learning? And then why is it your favorite?
Tanmay Bakshi: So my definition or what I think of as machine learning is a set of algorithms that allow computers to not really think but to understand abstract data that up until now only humans can understand but not only that, also allowing the computers to adapt to that overtime by themselves without extra human input.
If you think about it, that sort of becomes the distinction between AI and Machine Learning, I won’t get to that right now but really, essentially why I love Machine Learning so much is because it allows our computers to be dynamic. It allows them to adapt and learn overtime.
Just for example, the Watson that played jeopardy, they’re not going to code rules into Watson of how to play jeopardy, that’s impossible. You cannot have a set of even a million If statements that if you said Okay Watson, if this is a question, analyze it like this, this is the answer. It just doesn’t work, you have to use machine learning technology that Watson this huge computer can find the best way out, by itself.
But then again though, one of the extra reasons why it’s so great is because it’s able to understand abstract concepts that we humans create as well. Like emotion, it means nothing. It’s not something that you can represent mathematically, it really just means nothing. It’s something that we humans have created to describe how we feel, that’s it.
However, through machine learning, we can now make computers understand emotion but manipulate data in a way that it looks like its understanding that emotion and then using that understanding, understanding different types of data and being able to communicate with us humans on a personal level as well. That’s really why I love machine Learning.
Al Martin: Then how much time of your day do you spend on machine learning?
Tanmay Bakshi: I mean really, most of the code that I write is with machine learning and Watson. I would say that about 50 percent of my day does go to coding.
Al Martin: Really?
Tanmay Bakshi: Yes because that’s what I like to do most of the time.
Al Martin: Right. Now I know a lot of the questions out there are probably going to be what language is your - out of languages, which is your preferred language?
Tanmay Bakshi: Well my preferred language for generally anything like Linux, Windows, Mac, web development all that kind of stuff is Swift. love Swift. However, if you’re talking more specifically AI, then we’re getting into python, or C. If you’re talking about Web Development, more intense things that I can’t do in Swift then... So I wouldn’t be able to say this is the language I use but mainly, Swift, Python and of course tones of different language for individual use cases.
Al Martin: So you're switching between any language of choice depending on the need?
Tanmay Bakshi: Yes depending on the need.
Al Martin: So why don't we come back - you mentioned - well let me back it up. So I'm reading a book right now. This ties into what you’re talking about.
And he says it's the first time in history intelligence to be coupled for consciousness. Kind of like what you were talking about before where feeling and stuff like that, you can still have intelligence at least according to the end without consciousness.
But also in the book he talks to a 2013 study by Oxford Economists Carl Frey and Michael Osborne. And he said that 47% of the U.S. jobs are at high risk in 20 years. So that'd be 2033.
And he also goes into other areas like 99% chance that human telemarketers, chefs, cashiers, et cetera. I don't necessarily buy into the doomsday scenario. But where do you see that from where you're sitting?
Tanmay Bakshi: I see AI as one of the most exciting new technologies in the world. The reason I say that, to get to your point of intelligence segregated from consciousness, in a way that’s true. With the current type of computing that we have fundamentally, the way that we compute on the silicone. It’s impossible for us to compute consciousness. But it’s possible, in theory, for us to simulate intelligence.
However, there is one small problem with that. We are not really simulating intelligence. We are simulating a way for computers to simulate data they’re given that looks to us like they’re intelligent. Watson to us looks like it is playing jeopardy. It is a simple input, output machine in this case, the processing that goes in between that input and output in a sense modelled off of the way that we think.
So really what I think, getting to what you said about the 47% of US jobs, you know at risk. That is theoretically true for some jobs. Now what I say there is a spectrum. There is a spectrum of different kinds of jobs, there are some that computers will not be useful in what so ever, even today, there are some where computers will be very helpful when we augment the human that is using them and there are some where computers may be able to replace humans.
This is only for – and I’ll tell you why in just a second. In previous industrial revolutions we have seen jobs being shifted from one area to another. With AI though, it’s going to flip around all together. It is going to be entirely different and entirely disruptive. You can take for example, driving. Right now, humans drive not as a want in most cases but as a need. For example, for me to get here today, we had to drive for example 20 minutes to where my sister works and then 40 minutes here. Okay that is very inefficient.
Luckily, we aren’t driving through peak office traffic but still, if we were that would’ve taken be an hour or two to get here today from my house. So that is not very efficient there. Look at all the things that are going wrong there. First of all I’m spending time not doing work and not relaxing either, I am taking time to do a task that only takes my energy away. I’m spending time doing something that you know in some cases be life-threatening. There could be an accident. Doing something that you know is causing pollution and all those sorts of stuff.
Traffic jams, you know all this wasted time. Now how can we fix that all? By implementing AI, through for example the Watson self-driving busses. Now we can combine the public transport with self-driving cars, no more traffic jams, hardly any accidents. Of course, it would get me here almost instantly, I don’t need to do anything. All my energy is conserved. I am focusing on working or relaxing whatever else I need to do. So in that case, the AI is able to take the sort of Manpower that is in the job of driving and is there for really no end – not end goal but no end benefit to humanity and we’re able to take- at least directly- and we take them and put them into different fields that require them like healthcare, like life sciences that need these people.
Just, for example, its estimated that by 2035, there will be a global healthcare working shortage of around I believe, 2.5- 2.9 million people. So of course through AI, we can enable them to get into those fields. Best part, not only do we enable them, we make it easier for them to get into this field. Just for example, Watson for oncology, makes it easier cancer specialist, not even necessarily easier, but it allows them to do what they do in so much of a better and quicker way.
Another example is the Woodside company in Australia, what they did is that they went to IBM and they said okay, develop us a Bot that can understand data that our engineers gather so that we can eliminate this percentage of our workforce. IBM said sure, they developed this Bot and it turned out that Woodside didn’t cut any of their workforce, instead they hired that extra percentage instead. That’s because they realized that the barrier of entry for those engineers has lowered so much that they can actually come in and gather all the experience from those top engineers and now everyone, I guess you can say, has the same level of experience.
They are all learning from the top engineers instead of all that knowledge going into an archive box. So, now we are using AI to be able to transfer knowledge from one person to another. So really what I believe, getting back to your first point of segregating intelligence for consciousness, in a way that’s correct, in a way you could say. That computers aren’t really intelligent. Watson doesn’t know it’s playing jeopardy, it just is. It’s an input output machine, just like all other computers. And then the 47% of jobs, I agree, but then again it’s not overnight. That will take a few decades for AI to fully saturate those jobs and in the meanwhile we can slowly shift that workforce from A to B and make sure right now we start prepping our youth for that future and make sure we get into job like the life sciences, technology and healthcare that we know will be augmented but not replaced.
Al Martin: Fair enough. Let me ask you a question. You know, specifically from machine learning and I hear a lot of stuff that it can do. It’s very exciting. What do you think it just cannot do? Be specific, where does the line of distinction end?
Tanmay Bakshi: Sure, now really what I believe where machine learning is not useful is where we need a true imagination. An inventive sort of, where you need to do something where you need to be creative. Now I don’t mean to say that computers can’t be creative what so ever. I even created an application that uses IBM Watson machine learning that actually generates new kinds of music based of 169 Mozart and Beethoven’s songs. It creates new music, it’s called Make your Own Mozart.
However, if you think about it, the computer is only limited to that one scope. It only knows Mozart and Beethoven’s music. If you try feeding in more music, it will just get confused. It doesn’t have that kind of creativity or that kind of imagination which is why our jobs like for example artists, aren’t at risk for being replaced by AI. But then you might take a look at-how Watson was able to help experts editors create a movie trailer.
We use the power of AI to allow these artists to be even more creative and allow them to create even better music and even better art and then for example help them create even better for example movies.
In fact there is actually one example, not sure exactly which one but I believe it’s the Superman movie, there were a few parts where editors through CGI manually go back and remove a moustache of that actor in all of that movie. However, this one person used a $500 computer and in three days used an AI neural network to remove that moustache for them. $500 compared to the millions of dollars out of budget they used to remove that moustache in the manual editing.
But then again though, it’s not perfect. The AI of course makes a mistake somewhere and so using AI in its involved states will help solve those problems and then using humans to fix and edit with the output from the AI.
Again. ML is not good where you need creativity, true imagination but ML is good when you need to understand something that is created by humans or understand something that only we can understand.
Al Martin: It is amazing what ML and AI can do. In addition to IBM, there are other companies that are doing some great stuff like Google wrote a program against the games and I think the only two criteria that it was the pixels on the screen and you have to have the highest score.
Tanmay Bakshi: Exactly, so you know what they did is- you think that machine learning can only work in one specific domain and that is true. But what google did is that they took 14 different games and they trained it to play around 10 of them and then it learned how to play the rest 4 automatically just like that. It did not need to be trained on them but just like a human it inferred, Hey Red is a dangerous color or don’t go towards the skull or don’t go towards the gorilla or whatever.
It learns those kinds of obvious patterns that we as humans learn and is able to play the game. Similarly what they did is something called Neural Machine Translation, uses neural networks to translate one language to another. Basically style transfer for natural language and so what Google did is that they trained one model to convert English to Japanese and Japanese to English. Then English to Korean and Korean to English. It never solved one pair of Japanese to Korean or Korean to Japanese but yet it did Korean to Japanese with near human accuracy because it was able to create an intermediate translation for the natural language that it says and convert that to the target language. It’s really interesting what machine learning can do.
Of course there are some limitations but what it can do allows us humans to work in so much more of a natural way.
Al Martin: It doesn’t scare you at all?
Tanmay Bakshi: No.
Al Martin: Don’t worry it doesn’t scare me either. Hey, Watson. We can’t get out of here without talking about Watson. What’s so great about IBM Watson?
Tanmay Bakshi: Well you see, actually a lot of people do ask me sort of why, I use IBM Watson and of course one of the reasons is because Watson was one of the first AI that I first stumbled upon but if you think about it, IBM Watson API are simple to use. They take out all the gruesome work of implementing the machine learning. For example, with any of the projects that you see custom neural network techniques, it takes months to work with custom machine models to work with the specific kind of data that you’re looking for.
And you might want to base it off of existing algorithms or create new algorithms, that’s very hard but through the Watson API you have a sort of simple approach, where you can understand natural language or whatever it may be and the best part is those API or whatever it may be are very broad but not very narrow either. Because there are so many of them they work on almost any use case and best part, you can work with your own custom data. It is not limited.
For example with Google natural language you are limited to the google model. Although, Google natural language is extremely powerful, what you just leave it to is what is trained. With Watson you can actually put in your own studio and create a classification or whatever you want to do. So sort of a flexibility that you get while making it simple and API accessible. It looks great and then it also looks great and is entirely secure and encrypted, IBM never collects the data. If anything you can even use an enterprise license to host on your own platform if you want.
And it works on anything. It will work on Raspberry Pi to a laptop to anything. It enables users to learn machine learning and prototyping really quickly. And in one single platform you can go from importing raw data to exporting machine learning platform and all of the work that you did is in one unified platform. You’re not going to tones of different platforms to do all the individual pieces of work. It’s all in one place.
Al Martin: All right, I think that’s, you just gave us our next commercial, that’s perfect. So is it the healthcare part of Watson I presume, Watson is useful in multiple different scenarios but I assume healthcare is that your primary interest right now?
Tanmay Bakshi: Yes
Sort of my primary interest and generally with AI is using with healthcare, and the reason I say that, because if you think about it, not only is AI good for healthcare, healthcare is perfect for AI. There is just so much data in healthcare. It is actually estimated that in the average persons lifetime, there are generating 300 million books worth of healthcare, not pages that is book worth.
Now imagine through AI, in fact 64 percent of radiologists time goes to non-interpretive tasks. So with a little bit of efficiency through AI, it makes that radiologist work so much better so much more efficent and in the end more economical for more people to be able to access.
There are so many instances where AI becomes so perfect for healthcare and vice versa that really I believe that AI has the power to save and augment millions of peoples lives in this field due to the fact that it can understand that abstract data that up until now only we could understand.
For example with cancer, IBM Watson can actually all the cancer return that comes out every week which would take a regular radiologist 29 hours a day to go through. Which is actually impossible to do. Which is (Unintelligible) that there is so much literature coming out that it would take you more than a day to go through all of that. So, simply impossible and of course ingesting all of that information, remembering it and applying it…simply impossible.
And then there’s also the problem of, for example, a doctor in (Unintelligible)
New Zealand could be gathering some experience and then a doctor in Dallas, Texas wouldn’t have had that experience. Through one single unified Cloud experience though, we can get experience from everywhere and apply it everywhere. So generally, yes, AI in healthcare is my main priority.
Al Martin: Do you share that data? You just put radiologists out of a job. I’m just kidding. It’s all complimentary, I’m good. So I’ve got to ask this question as well. You have been working with IBM for awhile, how is it like working with IBM, if you say anything bad we’re going to cut it out anyway so it doesn’t matter, ahaha.
Tanmay Bakshi: I think that it is really very exciting. I mean there is so much powerful technology, there are some great people here. I think it’s just really exciting experience for example some of the summer training I did around one of the (Unintelligible) here a little while ago. I learnt a lot while I was here.
That actually originally got me into the world of web development, which is a place that I really wasn’t interested in before. I think generally the technology here and everything here is just great, I love working with IBM especially because of the fact that IBM is so open toward Open Source technology how it not only is using tones of (Unintelligible) but actually contributing back to those libraries and helping them work in them like for example power AI, How they were able to achieve 95 percent distributing accuracy with deep learning. Training literally one of the biggest deep learning models with one of the biggest data sets in the world in 7 to 8 hours. Where as it usually takes numerous days with 256 pascal GPUs using power. So this is all really exciting technology and I really love it and that’s why I love working here.
Al Martin: Good answer. You also have a goal, you started out with 100,000 kids you’re looking to train on the technology. And I guess, you know, my question is, any advice you would give those kids in terms of how to get into data, machine learning, Watson. Maybe some advice for the Vice President of Hybrid Data Management too.
Tanmay Bakshi: Well, to answer your first question. Really what I would say is – My first tip here is, AI is the next level of coding. Don’t expect it to be something that you learn individually. Coding, development AI, these things are basically next level of each other. We have to start off with the basics, code. One thing I do always say is, if you are not passionate about coding, don’t do it.
The reason I say that is because if you’re not passionate about that, you won’t be perseverant, you won’t find coding fun. If you don’t find code fun, there is no point. You’re just going to give up. This leads me to my second point. Make sure you’re very perseverant. Don’t stop on the first 50th, 100th or even 1000th error because then you won’t really get anywhere.
Coding is just full of road blocks because again when you’re trying to do something no one else has done before, you’re going to experience those roadblocks because you don’t know where you’re going. Just remember that once you get over that roadblock, you won’t ever experience that roadblock or a similar road block ever again.
Apart from that though, that tip that I gave really becomes useful when you’re using AI. For example, when you are training machine learning model and you don’t get accuracy. It’s not a simple, oh there’s a bug here, let’s change this code. You have to try, tens of different solutions, each one will take a lot of time to train. Each one you have to test in combination with other solutions so it’s a little bit time consuming, a little repetitive but the results are amazing once you’ve actually done it.
So, I would recommend being perseverant, follow your heart, follow your passion. Start small, easy and start simple and start playful as well. So of course what you want to do is move at your own pace. You want to start off with something simple, you want to start off with python, go over C, go over Swift, just make sure whatever you do, you’re doing at your own pace. You’re not just following a course but you are also learning on your own by creating or building your own examples and your own applications and you will learn a lot.
Al Martin: So you actually make mistakes from time to time?
Tanmay Bakshi: Yes, of course
Al Martin: So I know you come from a family of educators and teachers. So my question is I went out to your YouTube channel, pretty impressive. How do you set that up to make it so simple where you can jump in and people are consumed by actually being engaged with that channel?
Tanmay Bakshi: Of course. Something that I believe. Is before I record a YouTube channel, or video is understand the technology inside and out. Like if I am recording a video on how to use the (Unintelligible) J algorithm, it were to create (Unintelligible) facial segmentation and images. If I am going to do that, I want to build the examples myself. I will understand the actual system, I will understand how everything works.
Then I will say, okay, what are the complexities and what can be made simple? What doesn’t need to be explained? What needs to be made simpler? How can I phrase this so that some one who doesn’t know much about machine learning can understand this. Once I have kind of boiled that down, then it comes to the actual video recording, and finding an example that is interesting enough and useful enough at the same time to have people to be able to relate to it.
For example, if I want to take a picture of you and do a face swap with me, I would have to have some way to find out pixel for pixel where is your face. Now it’s not just about facial recognition, I want to know that pixel for pixel so I can do the proper face swap. And so, what I can do with a pix to pix algorithm is gather data even if it’s just like 200 examples. I can take a picture of your face, get it in a (Unintelligible), extract the box, feed that into the neural network and it will give me a pixeled map of where your face is, the imagine.
Al Martin: I wouldn’t recommend the face swap. I will take yours but I don’t recommend mine.
So, have a daily routine. I mean any leader has a daily routine. I get up. I usually work out. I, you know, have some water when I work out. Have coffee. I read. I typically write. You got to have some kind of routine when you get up every day to keep going, right. To keep energy going. What is it?
Tanmay Bakshi: Well, in terms of a routine, quote on quote routine. I couldn’t say that I have a strict, I do this, this, this and then I do this and then this. No, that’s not how it works for me. I kind of adapt to what I want to do, what I need to do, what’s coming up.
For example, for the past three days. I knew the past three days would be filled with me just doing machine learning and right after that the podcast. So what I would do is say okay, I want to work on this, this and this today because I have got this coming up. Just for example, but I do have to have some kind of routine though because I am home schooled, so when I wake up that’s kind of the first thing that we’re going to do.
Once that’s done it’s not like I’m going to work on this project and this project and then I’m done. It’s more like, cognitive story is coming up, I am going to spend all my time on the cognitive story. I have a meeting with a person regarding the mental health project, I am going to work on the mental health. Of course, then it depends on the importance and what I need to do.
So it depends on what I am working on, my schedule just kind of flex- is really what I want to do and then of course travelling over here for the podcast and then I have THINK and everything. So it is fun, not necessarily a strict routine but I generally have an idea of how my day looks.
Al Martin: Okay. You mentioned IBM THINK, that’s the IBM conference.
Tanmay Bakshi: Yes.
Al Martin: How many presentations do you have at that conference?
Tanmay Bakshi: Well, around one every day. Four presentations at IBM THINK and one before THINK as well. So it’s going to be really fun. One of them is on the cognitive story, I talk as well about a few projects that I am working in the field of healthcare and I hope to see you- some of you there.
Al Martin: I have two that I thought it was a lot. I thought that was stressful.
Tanmay Bakshi: No it’s not stressful it’s fun.
Al Martin: Good for you. Hobbies. Do you do anything outside of coding?
Tanmay Bakshi: Well, since you added that condition- Coding is my hobby. I like biking, it’s a great outdoor sport only when the Canadian weather allows for it. I like table tennis, my favourite indoor sport. One more thing that I recently got into is cricket. The reason I started getting into cricket is because a few months ago, an ex cricketer coach Jeremy Snape emailed me and he said he wanted to interview me.
So of course, being that I am in Toronto, how do we work that out? Well, I was actually packing to leave for the Watson Summit in Denmark and so what happened is we thought hey, why not just meet up in the middle. He flies to Denmark, I fly to Denmark and just before my presentation he took an interview of me and during the interview he taught me some cricket and so yeah it was really fun.
Al Martin: It’s like optimizing your time. So what’s next? I mean what’s in your future here? It could be immediate future or anything. The question’s open-ended.
Tanmay Bakshi: Sure. Well what I say is three things. First, is love doing AI and (Unintelligible), love researching AI, learning new kinds of AI, generally working on machine learning technology you know like Watson, Tenserflow all these kinds of libraries that allow me to develop machine learning quickly.
A part from that though I don’t like to just implement AI, but take the (Unintelligible) that I implement and put it to use into fields like healthcare. Of course healthcare is my main focus but I do other fields like entertainment, finance, security you know all these kinds of things.
But from there though, not only do I put that to use, but I take that and make it open source and make it available for the rest of the community and I make it easy for them to learn about those technologies in the first place. That’s really what I love to do and what I expect to continue doing in the future. That’s – there are no explicit plans for exactly what I want to do but this is -
Al Martin: Have you thought about university?
Tanmay Bakshi: Well, not specifically yet but there are some ideas, I live the research that the University of Toronto 53.20, Ryerson 53.21, Stanford 53.23, MIT 53.24, California 53.25, Berkley 53.26 they’re all doing great stuff. But I am not sure just yet.
Al Martin: So for people that follow you, is there a suggestion that you have if they want to see what you’re up to, that kind of stuff, without giving too much information.
Tanmay Bakshi: Yeah. I do have twitter that you can go to @TajyMany, I have got a YouTube channel, Tanmay Bakshi, LinkedIn, Facebook all under my name but if you would like to reach out to me with any sort of question, my email is the best way to reach out to me.
Al Martin: Fantastic. You’re fantastic. Kate, do we have time to open up to a few questions?
Fatima Sirhindi: Yup, we have two from the Facebook live. So the first one is Tanmay can you please talk about three cool new jobs that teens can start preparing for now?
Tanmay Bakshi: Sure, what I would say is first of all the technology. You should be preparing for coding and preparing for using machine learning. One more, I won’t be able to name a specific job but the field that there will be lots of job is healthcare. One more that I would like to say is education. Education, Healthcare and Technology, where I believe there are going to be lots of jobs opening up due to the fact that AI is going to be impacting them so much.
Fatima Sirhindi: And the second one- Thank you for that- The second one is going back to your emoji conversation, Tanmay what is your favourite emoji?
Tanmay Bakshi: Well, I don’t really use emojis that often. So I wouldn’t be able to comment on that but I will take a look in the emoji section and let you know soon.
Al Martin: You haven’t had that one before have you?
Tanmay Bakshi: No I actually haven’t. It’s the first time someone has asked me that, little- not sure what to say there unfortunately.
Al Martin: We got one in the audience here- we are getting the microphone over.
Mark Ryan: Thanks very much. This is really inspiring. Thanks very much for coming. It’s really, really inspiring. You mentioned some research. Is there any particular area or aspect of machine learning research that you think will really make a breakthrough? Something you look at and say, this is really exciting.
Tanmay Bakshi: Sure, one of them I believe that is number one is natural language processing. Now the reason I say natural language processing. The reason I say natural language is because if you think about it, 80 percent of the data we generate or use is unstructured. And that becomes, sort of with our technology, in a way becomes, you could say dark data.
We can’t use that data at all. It is just sitting there untapped. The reason is because our computers can’t make sense of them, there’s no point. For example, all that cancer literature that’s coming out, that’s a gold mine of information but no cancer expert or radiologist can actually go through that, and understand and make sense of it or apply all of that information.
I believe natural language processing is a field where we can understand all the data that we generate and then apply it in different kinds of fields as well. Apart of natural language other things are also important such as vision. Vision is so important because pictures worth more than 1000 words, of course, you know that.
So with pictures we have so much information but to a computer that’s just an array of pixels. How can we make the computer understand what’s in that picture, what’s in that video, understand what is in that surveillance video or understand all these kinds of videos and pictures and be able to derive insights from them and allow us humans to work with this data and work efficiently.
Just for example, as a radiologist I am training through all these different documents and the thing is, it will take me a few minutes or a few hours to find out exactly what the problem is here, where is the tumor here. With AI, we can do that so much more quickly and so much more accurately.
For example, Watson (Unintelligible) in Australia, every 6 hours one person will die from melanoma, that is a lot of people and it is one of the deadliest cancers in the world. But the thing is, if you catch it early enough, you got a 98 percent chance of surviving. And so if it detected early enough you have one of the easiest to solve cases.
Now the problem is, it can easily be confused with other kinds of diseases, in fact a professional oncologist can detect this kind of cancer with his naked eye with about 60 percent accuracy and with tools, around 80 percent accuracy. Watson gets 91 percent accuracy. So imagine how many people’s lives we can save just by having Watson do that vision processing.
Then of course audio. Audio is very important How do you and I talk? (Unintelligible) We don’t talk –we wouldn’t be texting each other if you were across the room. But as we’re talking to each other, how can a computer understand this audio or this very natural way of communicating without speech detection text to speech. Of course machine learning being able to understand and synthesize audio. So these are the three main fields I think machine learning is going to be used for in the future. In the near future and far future. Thank You.
Fatima Sirhindi: We have another Facebook question. Tanmay, how do you compare AI with Anaytics.
Tanmay Bakshi: So you see, analytics is what comes before AI. That is what I believe. Analytics is not something that you can say alright, this is something separate. They are you could say, two parts of the same field. The reason I say that in fact, I talked about this in India when I was (Unintelligible) Data Science, IBM Partner and so, really the point is Data Science, Analytics and AI are inseparable.
Unless you do the Data Science to actually get the data and actually clean up and understand that data, and until you use analytics to properly get a thorough understanding of what is within that data and what that data encodes, you cant really use AI. Like with the cognitive story, we’re not just saying EEG brainwaves, neural network. No, what we’re doing is EEG and neural networks manually and once we’ve gotten those analytics and the data that we collect. We say alright, we have done the analytics, we understand the data, now how can we apply that knowledge to create a better, more fine tuned AI model. That’s really what I believe, they’re like separate parts of the same field and are inseparable and they come one after another.
Fatima Sirhindi: Is there a way AI can help the media sales industry?
Tanmay Bakshi: Absolutely. This is a bit funny because funny example. What I was able to do is actually a twitter pole. I created this twitter pole and I asked people, what do you prefer, IOS or Android? I got a few responses as you can imagine, quite a few people on IOS, quiet a few people on Android and so what I did is I actually asked Watson personality insights to find the personality of the people that liked IOS and Android and I fed that into one of my neural networks.
I was able to, with a good amount of accuracy detect, if this person likes IOS or Android and so yes, AI can be used in sales and in most cases its not just used for bad tracking. But also for good cases as well. You as a human might actually legitimately be looking for a product unable to fund it. Through AI we can do targeting advertising, not only to the benefit of the retailor like amazon or whatever else you may have but also to the benefit of the end user. Hey I was looking for that and AI helped me find it. Its not like you were looking for shoes, you get shoes, I am looking for this kind of shoe, this very specific, natural language shoes and now AI realizes that, finds that kind of shoes and advertises that specifically to me.
Generally yes it would be interesting to see how AI ties into you know the cognitive, and advertising Watson has got. They are using these kinds of things, its being lots of different fields and sales for example is really going to be impacted by it a lot.
Kate Nichols: Yes we got one more here.
Woman 1: Hi. You are inspirational and overwhelming all at the same time. So a personal question. If your father had not been a software engineer, let’s say he was a plumber, what do you think you’d be doing?
Tanmay Bakshi: Yes, Well I mean you’re right. What really got me fascinated by computers was the fact that my dad was one of the astonishing parts but it was not just that. It was also because of the fact that generally I could see my name pop up on a screen somehow and I had no idea how and it was like magic.
I think it was a combination of the fact that my dad used to do programming and programming in general would be so fascinating to someone five years old, that combined allowed me to get into that field.
Also one more thing to add to that, when you’re five years old, you’re more open to learning about anything than when you’re older. Just the fact that you have got more (unintelligible) your brain can adapt to whatever you give it. Just because of that, it became easier for me to understand that concept than it would for me for example now if I started learning how to program. It would actually be harder for me to learn how to program is I started at 14 than It would than when I was five. Because at five, you’re so much more open to and can quickly learn more concepts.
In fact its actually known that if you were to for example take a two year old and show them two monkeys, even though we have humans have lost the ability to distinguish between these animal faces, a two year old can do that perfectly because they’re born with billions of neurons in their brain that they’ll never need. My brain just goes oh I don’t need that and just puts it away.
However, if you keep showing them monkey faces, their brain stimulates those neurons and that ability stays for the rest of their life. So similarly at five years old, I had this interest towards computers and the brain power to actually adapt to that understanding and grasp that concept and then the openness to learn that in the first place.
So generally, I think it was a good combination of everything that allowed me to get into that field. That is what I really believe.
Al Martin: Kate how much time do we have? We got to end it there, that was a good ending right there.
Hey I have got to say you have exceeded expectations, I appreciate you being here with us, you can see that everyone is appreciative as well. Your family, your sister is here, it’s a family sport so thank you very much. Have a nice day.
Tanmay Bakshi: Thank You, glad to be here.