Podcasts

Making Data Simple: Will machines take our jobs?

Post Comment
Making Data Simple: Will machines take our jobs?

Overview

In this episode of Making Data Simple we hear insights from IBM Machine Learning Hub data scientists Jorge A. Castañón and Óscar Lara-Yejas as they discuss what machine learning is and is not. They also answer the most controversial question today: Will machines take our jobs? Listen to find out!

Show Notes

00:25 Connect with Al Martin on Twitter (@amartin_v) and LinkedIn (linkedin.com/in/al-martin-ku)
02:10 Connect with Jorge A. Castañón on Twitter (@castanan) and LinkedIn (linkedin.com/in/jorgecasta)
02:45 Connect with Óscar Lara-Yejas on LinkedIn (linkedin.com/in/óscar-lara-yejas-36700058)
10:10 Tweet to Ginni Rometty @GinniRometty and read more about her here: https://www.ibm.com/ibm/ginni/
11:00 Listen to "Machine Learning" by Business Daily here: http://www.bbc.co.uk/programmes/p050db63
16:45 Watch Neo Yokio on Netflix https://www.netflix.com/browse
31:50 Find Michael Lewis "The Undoing Project: A Friendship That Changed Our Minds" here: http://amzn.to/2yr5Dgv
35:55 Find "The Greatest Show on Earth" by Richard Dawkins here: http://amzn.to/2wi4pmT
36:25 Find Ian Goodfellow's "Deep Learning" here: http://amzn.to/2hwZWKr

Hungry for more? Check out our previous podcast episodes of Making Data Simple:

Transcript

Making Data Simple: Episode 4 Transcript

Update: Will machines take our jobs?

Al Martin:  Hey, guys, welcome back to the series — Making Data Simple.  This is Al Martin here.  I'm the lucky guy that gets to host this.  Today's topic is going to be, Machine Learning and we'll probably dive down into Machine Learning Hub. 

Most of you that are listening probably know what machine learning is but my simple definition — and I will let my guests that I introduce in just a moment — give a better definition.  But my simple definition is machine learning gives computers the ability to learn without being explicitly programmed. 

And the cool thing about it is — at least for me — being the IBMer — is I believe that term was actually coined by Arthur Samuel in 1959.  So we've been doing it a while.  And it's really about prediction making in my mind but, again, I'll let the experts define it with use of computers, models, algorithms and that kind of thing.

With that — without further ado — I'll introduce my guest.  I've got two today so this will be a little different than what we've done in the past.  Jorge  Castañón; and Óscar Lara-Yejas.  I apologize in advance guys.  I'll let you guys introduce yourselves.  But they lead the machine learning hub here at IBM and they know much more about machine learning than I.  So they're going to straighten me out today.

And the cool thing about this, guys, is I've done this so I get to learn, right?  I can put all my ignorance aside and I can ask you guys questions and get smart.  That's the deal.  So Jorge let me start with you.  Could you quickly introduce yourself?  And you might say your name again just so we get it right for once.  And then your background on machine learning and then I'll turn it over to you just the same, Óscar.  

Jorge A. Castañón:  Sure Al.  Thanks for having us.  My name is Jorge A. Castañón and I've been at IBM for the last 3 years.  We started with the machine learning codes early this year in February.  And before joining IBM I did my PhD in Machine Learning numerical optimization at First University.  And also so this year would have been working on the ML Hub and, of course, Óscar and I were - could live here at the ML Hub. 

Al Martin:  Fantastic.  Óscar, a little introduction from yourself.  

Óscar Lara-Yejas:  So my name is Óscar Lara-Yejas.  And — by the way you pronounced it just perfectly.  You know it can be tricky sometimes.  But that being said I've been at IBM for a little over five years.  Matter of fact, last week was my 5th year anniversary at IBM and... 

Jorge A. Castañón:  Congratulations. 

Óscar Lara-Yejas:  Thank you.  Thank you so much.  If you work here the Silicon Valley level is very pretty — it is beautiful nature around here.  And initially I joined as part of the big data team working on some big projects for the last year in machine learning and analytics and now I'm part of the machine learning hub.  Very excited, very happy to be a data scientist and be part of this initiative. 

Before that, I did my PhD in Computer Science and Engineering at the University of South Florida in Tampa.  Go Bulls by the way.  And I'm really excited to be here sharing some our experience and little knowledge of I’d say you’re overselling us.  Happy to be here and sharing some core thoughts with you.

Al Martin:  Fantastic.  I guess we've established that the only person without a PhD at the table here is myself.  I'm feeling a little bit inferior at this point in time.  That leads me to my next question then. H ow you go about being an expert in machine learning?  Maybe we have some students out there listening.  How do you get on that path?  How do you stay on that path?  What inspires you to be on that path?  One of you - just anyone - Óscar, why don't you take this one first?  How you jumped into that path.

Óscar Lara-Yejas:  Absolutely.  I will say the first thing is to have passion or an interest in something.  Once you have the passion you can accomplish pretty much anything.  So in my specific case it just fascinates me how you can get inside from data.  To me that's what machine learning actually is.  You have data and our customers have tones and tones of data.  Many of them — they already have –Db2 products, you know, all sorts of data.  But then data usually from this - you can get inside out of them.  And that's what specifically fascinates me about machine learning and the fact that you can learn stuff from your data, right?  And you said, that you learn to make predictions or to make decisions in the present and the future.

So I would say, of course, that taking formal training in terms of taking classes at the University it could - that data mining class, I took (unintelligible) optimization. Taking classes is, of course, very useful but one thing, machine learning is something that you only learn when you do it, right? 

When - you can know so many equations, you can know so many more complicating algorithms but there's some kind of common sense when you approach every single problem then you can only develop by doing it.  By, you know, working with the data, writing the algorithms and making things happen. 

It's a very hands-on kind of field and that'll be my take on it. 

Al Martin:  So Jorge, when did you know that machine learning was your career path  and, you know, hell, you've got a long career ahead of you but in terms of where you are now, how did you know that this was the place you were going be and what got you here? 

Jorge A. Castañón:  Yes, good question. I think at the beginning or maybe half way of the PhD I started to work on machine learning related problems with medical imaging.  So I think there's - that's where I started to see how important it is to use historical information to build machines, to build algorithms, to give predictions.  So it was very important for me to realize that I can use historical information, for example, brain images to train the mothers to be better and have the patient less time on the machine just because you have seen the history of (brains) before. 

If you haven't seen that history then you will give the patient more time and your condition is going to be less accurate and so on.  So that's where I, you know, when there's - that concept of having or of learning from examples — and I really like your definition of machine learning because that's what it is, right?   So you take the history to take examples?   And you provide that machine to learn rather than coding it explicitly.  I think we’re completely aligned with that definition. 

Al Martin:  So let's hit that one more time.  Just so, what is machine learning?  If you could repeat that one last time just so we're clear.  And then what is it not - if you can answer that question. 

 Jorge A. Castañón:  Sure.  Machine learning is the ability to learn from example.   Right?  So it's always examples that you  need - examples of data in the case of supervised machine learning.  For example do you always need this associated information with the label of the correct thing that you are trying to build around.  Right?  So if you don't have a examples, if you don't have historical data - then you cannot do any machine learning.  What is not?  That's a question that we have to answer all the time with the machine learning hub because a lot of our customers and colleagues come to us and they just ask - they just feel that they can solve every single problem with machine learning.

So the first conversation that we have with people is like this is what machine learning is.  This is what an example looks like.  This is what we need to actually start to do a machine learning model.                              

Óscar Lara-Yejas:  What is not — I'll say — it's not the Holy Grail.  That's what people say when they approach us.  You brew the label into machine learning and then magically you get all the information and answers.  That's what it's not and we have to do a lot of education and a lot awareness on that.  

Al Martin:  Sounds like to me that there's — like anything — particularly new technologies.  There's a lot of information out there.  When I look at this, is I kind of - I think we're in an unprecedented — this is an overused term just like the next one I’m going to use  -- but we're in an unprecedented time or era where we often say data or overuse the term that says data is the new natural resource.  

And I think there's a tendency for myself as well to say, yeah, whatever.  I mean we've heard that before.  Blah, blah, blah.  But when you start thinking about it more deeply you realize I mean when we really think about it that everything we're doing today, darn near everything is leaving a digital footprint — from credit, to mobile, to online, to purchasing, and we're feeding all this data into machines like machine learning.  And I think that creates a lot of worry and fear — which then promotes the information that you're talking about out there in the industry.  

The first one that I always find funny is that the first fear that I hear   about is machines are going to take our jobs.  And we've got to do something about it.  What are we going to do?  And I think - I give credit to our Chairman, Ginni Rometty and her definition is augmented intelligence, not artificial intelligence.  Because I think, you know, we fit side by side with that machine learning to make it useful. 

But the interesting thing to me and this is what I want to get your feedback on is when I don't hear as much about - there are concerns that I would have and that is around ethics.  I mean by example on one hand we need data, you know. The more medical data by example we feed into some of these algorithms the better we're going to be able to find cures. 

But then on the other hand, you're thinking, well, do I want to have my personal medical data out there for, you know, you don't know where its going to go, who's going to consume it.  And then it goes beyond that in that - literally this morning I was listening to a thing about the European Union and they proposed the idea of e-persons where you have electronic persons where you'll have a robot that is responsible and, you know, it can even be paying its own taxes at some point in time.  And for me to think a robot is liable is ridiculous.  I mean, again, it goes back to the automated intelligence.  Only the human is responsible, whatever, the output of, you know, machine learning or otherwise. 

If that's not enough I'll tell you one more thing and then I'll turn it over to you guys.  I want to hear your take on ethics.  My other worry is that if anything machine learning could make us lazy in that just like you said — you said we think it's the Holy Grail or there's many people that come in and think it's the Holy Grail. 

It' s kind of like my back up camera or the backup camera they have on cars now.  I'm always worried about my children.  You know, they just rely on their back up camera.  It's got to work, it's got work.  And they forget, you know, where we used to look at mirrors.  We used to turn our head to make sure nothing's there.  We're going to continue to rely on machine learning the same way and I have to admit that my back up camera the other day was covered up with mud and it was going off and sure as heck wasn't my kid, it was me that backed into the car behind me.  So I'm as guilty as anyone else. 

So talk to me a little bit about - and I'll start with you Óscar about the ethics.  What do you see in terms of automated intelligence, what it means, what do you think we're going to have to be more concerned about today versus the fact that they're just going to take our jobs?

Oscar Lara-Yejas:  Absolutely.  And I completely agree with you.  The way I see machine learning and artificial intelligence is the tools that help humans to become better at their jobs rather than, you know, something that perhaps has jeopardized people's, you know, work.  So just to give you an example, right, like Watson is doing a lot of stuff with healthcare.  And understanding a ton of research papers to try to come up with potential cures for cancer and many other diseases, right?  And something that a doctor would take a doctor, you know, days or weeks to understand and process --  machine learning algorithm may do it much more quickly, therefore, making the doctor much more productive to make a decision and to help diagnose it and so forth, right.  I also heard that the, you know, there's a system that helps musicians to compose music so that’s called Watson Beat and one thing was ordered and actually went to his concert last week and he’s using this system to write music himself.

And I don't believe that we're going to reach a point where, you know, composers that are going to be kicked out of a job by robots but on the other hand it's more like, you know, a composer may use or may, you know, leverage inspiration  from this system on different musical components that may be more productive in the composition of music relation kind of, you know, process, right?

And I somehow associate this era with the Industrial Revolution, right?  Before the Industrial Revolution happened there was a lot of people doing manual labor with very heavy lifting.  In order to build stuff and transportation and many other things.  And at that point, what happened was that many people "lose their jobs" but it allows, you know, society to advance technologically and those people who lost their jobs at the time, they had to be reassigned to other jobs, right? 

So I guess that's a lot of training. There may be certain jobs, for example, you know, it is found that the - for example, the drivers are, you know, the number one job in the US, right. You may argue that we have just driving cars, right?  Those people are going to be displaced out of their jobs.

And that's, of course, of concern.  My thinking is that would be, you know, many people would need to perhaps learn new skills in order to be able to, you know, perform other types of jobs.  So I guess it's two different things.  One, in terms of augmenting excuse me, and improving the skills of people such as of the case in the doctors, such in the case of musicians, and so forth. 

And on the other hand, I would say some people, yes, they may need to learn new skills in order to perform different kinds of jobs that are no longer going to be available for us.  They are going to be totally automated by machines.

Al Martin:  If anything, I agree with you.  I think we are in the era of continuous learning.  I mean you don't just do one thing and then stay with it for 50 years.  Maybe like our grandparents did.  You have to learn new skills and continue to, you know, almost in agile fashion do continuous learning. 

So, Jorge what do you think the biggest challenge is going to be moving forward in the era of augmented intelligence and machine learning.  What do you think the real concern is that we should be addressing? 

Jorge A. Castañón:  Yes, so I see machine learning - an era of having machines helping you rather than replacing you.  So I'm in line with you guys.  And it's funny that I just saw a TV series from Netflix.  It's six episodes.  It's very new.  It's called Neo Yokio.  It's very fun.  But I want to talk about is that the main character of this cartoon, it's like anime cartoon.  And this guy has a robot with him all the time.  And that robot is assisting the main character all the time on everything that he's doing and he's not doing, right?

And at some point of the TV series the robot opens and it turns out that there's a human inside the robot.  And they the guy is so surprised.  Like how come I didn't know you were there?  And then the person is then like, yes, so I've been here all the time.  The robot is only filtering and it turns out that this human is actually not nice.  Right?  It has no filter.  Like, they don't - they're kind of like - they're not friends with each other but the robot and the guy are still like super friends and it's very funny.  I think that's the way we're going to go, right? 

But I think, we need to actually need to teach these machines correctly, right?  So if we're filtering things going back to ethics, right?  What to filter?  What not to filter?  Right?  As a data scientist I think you can - we have a big responsibility, right?  Sometimes some of the things that we're working on are so difficult that it's our responsibility to be super transparent and, you know, in terms of ethics, maybe you understand things that not everyone is understanding.

So you have the responsibility to, you know, share that information in a very clear way like crystal clear, like what are their limitations, what are - what is this algorithm for?  And so, right, so that's the big thing that I think in terms of ethic as a data scientist.  

Al Martin:  Fantastic.  That's a good example.  I'll have to check that out.  Hey, so let's turn this in to machine learning hubs.  Because I know that's where you spend a lot of your time right now.  Not so much looking for a sales pitch but what is it?  Why do you think it's necessary?  And what kind of outcomes are you able to drive within these "hubs?"  Can you tell us a bit on that Jorge? 

Jorge A. Castañón:  Yes, so what we do at the machine learning hub is try to close a big gap that we have right now.  And that's the fact that there's not enough data scientists.  Right?  So there's - irrespective to - in 2018 we're expecting to have 18 - sorry half a million data science roles.  And we need it, right, so - and we will only have 200K data scientists out there. 

So 40% only and I think that's only the US but it can translate to all the world.  And so in the machine learning hub what we're trying is to close a gap, right?  So maybe there are some companies approaching us.  They have data.  They have a lot of passion to get value out of their data.  They don't know how.  Right? 

So they come with us - the come to us.  We spend those two or three days with them and we help them achieve one single model.  One simple use case.  And then after that maybe they can take it from there.  They can build more models.  They can be more precise.  They can incorporate more data.  But I think the main mission is to close that gap and to be more transparent to us.  How data science is applied.  

Al Martin:  So describe to me some of the common issues you find as clients come in, they say, "Look I need help.  I need help with machine learning."  They come into the hub.  What do they usually come with is the question I would have and what are the common problems you see and how are they reconciled?

Jorge A. Castañón:  Absolutely.  So as we mentioned in the very beginning, one of the most difficult problems is to understand what machine learning can do and what it cannot do.  Right?  And in  many case, you know, people come to us with big dreams and one of the most common challenges is to nail down what is it that the machine learning problem is going to look like. 

 Identify the problem and then identify what data is going to be used to solve such problems.  In many cases there are issues with data access, for example, the data is very sensitive, for example, in this use case we're working on with a healthcare company.  They have patient records, right.  So we have issues getting the data put in - get out of the main frame because the sensitivity of the data and we actually in that particular case we don't - the customer is coming to us.  We have essentially travel to a site because the data couldn't get out of theirs.  Data access is an issue for sure. 

Understanding or trying to nail down a machine learning problem is also an issue.  Data curation is also an issue.  In many cases, the data is simply not clean and this thinks that we take for granted in the relational world, you know, such as having a data scheme available, we take for granted that in the relational world — such as, you know, having the data scheme available, having a primary case — having (unintelligible). 

You know, machine learning, use cases, you don't have that.  And oftentimes you get a datacenter. You will know what the comments are.  You know where the data types are.  You know, a lot of missing information and missing values.  Instead of having just one single data source, which is, you know, most machine learning now takes us into one specific dataset instead of having just one, you usually can come there.  Tons and tons of files you have to somehow reconcile, put together a team and all those things.  Right? 

And those become challenges as even before you being able to train your machine learning model you need to, you know, do a lot of data planning and data preparation.  And, of course, understanding also some of the nuances of the - itself is - it may be challenging as well. 

Machine learning generalists.  But we're not experts in finance, we are not expert in healthcare we are not experts in biology. In many cases we have to learn from all these domains in order to be able to build a machine learning model.  And it responds because the more you learn about machine learning you end up learning about all this specific domains, which is beautiful.

Al Martin:  Here's something that confuses me about this and maybe Jorge you can help me.  I get what you're saying. Well, I thought I heard you say, one of the biggest challenges you face is actually getting at the (desperate) sources of data, making sense of that data, cleansing that data, ingesting that data, and go on and on. 

But I got to imagine if I'm a client and I'm coming into this hub, I bet their thinking of the opposite end and maybe I'm wrong here.  You guys tell me in terms of what your experience is and what I mean by that is they probably have what they feel is a handle on their data. 

 Though you probably - they probably have to convey that to you and convince you of that.  So I bet - I just would assume that they would have come in from a client perspective and they’re worried about how am I going to find the right programming language to do this - the machine learning?  What language should I use?  How do I train these models?  How do I base these models off of?  Am I right in that or are they - do a lot of the clients come in and are concerned about everything top to bottom?   

Óscar Lara-Yejas:  I think that's a very interesting topic that you bring.  It's - we have both I think. Sometimes the customer, they already have the data that they want to use so that they already have even have the idea.  You may need to clarify or a (comparison) and so on.  And so in those cases we can really start like right in the middle or with a lot of, you know, advantage.  We can use our time much better. 

But, yes, sometimes, the customers come and its - they are just starting.  Maybe they have just a little bit of data, right?  So the best thing to understand is that they data limitations can only take you, you know, you have, I mean going back to the machine learning definition.  You learn from example.  You learn from history.  You don't have history in the data, you would not learn that stuff. 

So if the data is a limitation then you need to think about, okay, if you don't have a good presentation of your data in this dataset then you cannot expect that this model will work for any case, right? 

So sometimes it's very - it - we need to spend a lot of time at the beginning to just make clear that the model that they are - that we're building is just based on these datasets and that it's very important that there's another data source that you need to connect or that there's more data that you need to use covering many years, for example. 

Then it's always good that they go back and actually give the model that is more comprehensive.  So - but definitely we have customers that have very different focal points like - so it's pretty open.  And it's always a challenge even though we have customers in different parts of the pipeline.

Al Martin:  Is face-to-face absolutely required with these or can you do them remotely or is there a preference?  What's your thoughts on that?  

Jorge A. Castañón:  I think face-to-face is a must at least at the beginning.  I think there's three reasons because of that.  I think trust is the first one.  Because you will not just give the data to anyone.  Right?  I think you need to - my second point is to build a relationship, right?  So you build a relationship and you start to trust people, right?  Maybe just trust IBM and with everything in time with (MBA) and all that, then, of course. You are legally okay, but you're going to kind of like get naked in front of people when you share the data, right?

Something like that.  You really need to like feel comfortable with someone else to getting to that point. 

I think trust and - it's actually two.  I think trust and building the relationship goes in the same bullet and the second one is clarity.  Sometimes customers come and you need to clarify what the goal is today or what is achievable in those two days and to be in line. 

Sometimes if you're not face-to-face it's very easy to not be here and maybe you go and solve a problem that is not the one that they need a solution for.  Right so clarity is one and the other one is trust and building relationships.  You will definitely trust someone more that you can see and with only legal implications on the table, of course.  But, yes, so it just having that face-to- face will help you build the relationship and they will be more comfortable sharing the data.  

Al Martin:  I think it goes back to the concern with data that I had.  I think those are some of the bigger issues is people willing to share their data.  The ethics around the data.  You know, relying too much on machine learning.  Again, it keeps coming back to that for me.  So let me ask you this as we kind of wrap up and like with any topic it seems like, you know, we may have to bring you back and do a deeper dive on some more questions on this because we could go on forever. 

But, Óscar have you - could you describe a wow moment or something - I mean something, by example, an outcome that was just growing as a result of, you know, the machine learning hub or your work with a client.  You know, give me something that you just uncovered that is just like brilliant.  

Óscar Lara-Yejas:  So one of the very first engagements that we had was with a healthcare company.  And that's the one I was actually telling you that we have to go inside because of the hub, you know, because the data being so sensitive, patient records and whatnot.  And it was very rewarding because of different reasons.  I would say the first one is, you know, for us, data scientists and in being journalists as well. 

We get to work in different domains like, you know, financial services, you know, manufacturing all kinds of work.  But somehow healthcare, it gives you this kind of extra reward that you are somehow helping or contributing to society by improving people's health with your work.  Right?  So that was actually very interesting for us.   In that specific use case, we were given data from patients, medications on file, also from patient blood tests and some of the different kinds of medical tests.

And we were able to create a model to predict which patients were at the highest risk — not compliant with their medications or were more prone to get certain diseases and that kind of stuff.  So - and it was very interesting not only because of the fact that we weren't able to build an actual model from the machine learning, you know, technical perspective.  But also because of the fact that, you know,  it's rewarding to feel like you were contributing to improving people's health. 

And also we were working with a medical doctor.  Who was advising us all the time and giving us great feedback on the topic and it was also a very good collaboration with him just from the onset — very  nice individual and very eager to, you know, use machine learning to improve people's health.  Just as simple as that. 

Yes, and I think a very, wow, moment was when this doctor came and saw the results and saw the slides and saw the tradition.  And in his head, he was always making sense and he would transmit that to us so it would be like a very very, you know, interactive session and we would get the results pretty fast, like (intradate).  It was really amazing.  

Al Martin:  You know, I read a book.  I can't remember the book and I - well I can't remember the book.  Very interesting. It was talking about doctors assessment of the human brain and it showed - they went through many many experts in the field.  They would assess the brains for, you know, malignant or tumors and essentially they couldn't - sometimes they wouldn't even come up with the same results themselves the second time they went through and looked through the information.  Given a case for machine learning to help guide them because it's going to be, you know very objective, just very precise, statistical in measure, et cetera. 

 Anyway, let me ask this and then I want to go to a lightening round.  Any last word that you guys have on machine learning that you want this audience to understand and know that you feel you wouldn't be right if you didn't say a couple of comments on it before we leave.

Jorge A. Castañón: It's very good to get that, you know, introduction course of machine learning before trying to apply it.  The conversations are much better after, you know, quality-wise know what machine learning can reach.  It's a very important thing.  

Al Martin:  Any last words you have Óscar?  

Óscar Lara-Yejas:  Yes, I would say even if you're not an engineer or a computer scientist or as a musician or somebody in science you need to know what machine learning is.  Because machine learning can help you improve your business.  It can help you improve your daily work if you are in health, you are in science, any domain.  I believe that companies that are going to survive and become competitive are the ones that leverage machine learning.  Because the main thing is the machine learning can do things much much faster and in some cases do more accurately than humans.

So the question is going to be more productive.  If you know what machine learning is and if you are eager to apply it.  Everybody you are going to see a lot of engineers, computer scientists.  You learn about machine learning. You learn the possibilities that we have and how we can help you in your day to day.  This is no sci-fi anymore.  This is real and it's happening as of today. 

Al Martin:  Fantastic.  Thank you very for that.  So before I finish I want to go into what I call lightening round.  It's a little bit personal, it's a little about business, in other words, so the audience gets to know you.  They are like 10-second type of answer.  Like boom, boom, boom, boom, boom.  Because we have two of you, Óscar I'll go to you and then I'll go to Jorge and then you guys can answer real quick.  So these are a couple of questions I just quickly jotted down that I wanted to get from you.  Number one is the biggest challenge you currently have with data? 

Jorge A. Castañón:  Cleaning.  

((Crosstalk)) 

Jorge A. Castañón:  Cleaning data to scale in a more automatic way because if  not, all your time goes there.  Okay, that's my major challenge. 

Al Martin:  Got it.  You got anything different Óscar or is that your same one too.  

Óscar Lara-Yejas:  I agree with that.  I will say data security and data access as well because of the same things you mentioned.  Data access, security and curation is definitely a big one. 

Al Martin:  All right.  Óscar got another one for you. What are you learning about now?  What's your learning vehicle?  Do you like podcasts?  Where do you go to get your information?  

Óscar Lara-Yejas:  I'm not much of a podcast person to be honest with you.  I like to read books that are online.  I read a lot of Wikipedia too.  In terms of, you know, machine learning, I was recently learning about on a supervised learning with categorical variables. I was learning about K Prototypes and K Modes and stuff like that. In terms of innovative stuff I was learning about the theory of evolution by natural selection.  I was reading about a book by (Richard Dawkins) The Greatest Show on Earth and all that.  

Al Martin:  The first part of that was PhD stuff. I think it just went over my head.  But another one on evolution, kind of like evolution and safety.  If you haven't read that one, that's a great one,. What about you, Jorge? 

Jorge A. Castañón:  Yeah.  I usually use media and gitHUB.  Those are my like primary sources for learning.  I'm reading a book on Deep learnings, too, which I got from one of my interns in the summer. 

Al Martin:  You have the title? 

Jorge A. Castañón:  It's called Deep Learning. 

Al Martin:  Oh, that's pretty easy.  All right, fair enough.  Who is it by?  I'm sorry.  We'll put it in the show notes.  Let's move on.  Two more questions.  We'll start with you, Jorge, this time.  If you want to become - if I wanted to become an ML expert, what's the fastest way to go - where should I go with the fastest way to get there?  

Jorge A. Castañón:  Oh, the fastest.  So I think you need three things.  I don't know how fast you can get them.  One is the math background.  Just to have a very math background, linear algebra.  That's one.  The second is programming, right.  You need to be like very happy programming on python, MATlab…R  Whatever script and language.  But to be very proficient.  

And also they are applied.  So be able to actually, you know, be creative and get some datasets and imagine what to do and what to try.  Those three things.  And I think you can get them in school or outside school — course books.  

Al Martin:  Okay, fair enough.  What about you Óscar?  Same answer or you've got a different one? 

Óscar Lara-Yejas:  I agree with what Jorge just mentioned. Especially the third part because it only happens in knowledge but having this - I call it common sense --  which I mean really it's not so common I have come to realize.  So this kind of creativity and this becomes like the artistic part of machine learning and data centers.  So the science is, of course, this math and this algebra and also some math problems.  If you encounter and (unintelligible) but also this artistic that comes from Black Magic.  Just coming up with the right features, kind of our artistic component (unintelligible) to make a lot of difference in the response you get.  

Al Martin:  Awesome. Thank you.  The easiest question of the day.  Where can the team reach you guys?  So Óscar where can you be reached and where can, if you want to get information on the machine learning where do you go? 

Óscar Lara-Yejas:  Absolutely.  So you can shoot me an e-mail.  My e-mail is - I guess you can put it on the note.  It's (obye) actually it's kind of complicated to spell.  

Al Martin:  Okay.  We'll get it in the notes

Óscar Lara-Yejas:  ...site as well.  We are located in San Jose, California the heart of Silicon Valley.  We also do have some other machine learning options in Germany, Toronto, Beijing  and China.                             

Al Martin:  And same question for you Jorge.  Any social media outlets that you would reference? 

Jorge A. Castañón:  I'm on Twitter.  My channel is (@castanan).  

Al Martin:  We'll get that in the show notes as well.  Is that your preferred channel?  

Jorge A. Castañón:  Yes, for sure.  Or LinkedIn as well. 

Al Martin:  Look, I'm going to wrap up guys.  I thank you so much.  You gave a lot of information today.  Kate Nichols who does the production, we're going to have to get these guys back on and next time I'd like to go into more of the how and how machine learning works and the statistics behind it.  

We'll have to keep it just below PhD level knowledge so I can understand it but if you could do that, that would be great.  Meanwhile, guys, Óscar, Jorge, thank you very much.  Pleasure talking with you guys today.  Until next time I hope everybody enjoyed this.  Let's take a look at the show notes and let us know what you think.  Well done. Talk to you next time. 

END