Pleasure to have you on the
podcast today. Thank you. Just for people who don’t know, Dr. Alistair
Johnson, he holds a PhD from the University of Oxford and is currently a
research scientist at MIT. He is also the youngest person to ever graduate from
McMaster University until recently until two years ago until two years ago beaten
by a year beaten by a year right and he’s named Forbes 30 under 30 under the
healthcare category, am I right? That’s correct. Dr. Johnson analyzed
data from millions of patients to identify novel treatments and create
highly accurate artificial intelligence algorithms. He is an expert in
machine learning and AI. We’re ecstatic to have him on our show today.
Yeah, thank you for having me. So just a warm-up question, where are you situated right now? So yeah, I’m at MIT
right now doing research mainly focusing around healthcare. I’ve always
had strong interest in health care, working with hospital data from around
America mostly we have a quite a deep
collaboration with the Beth Israel Deaconess Medical Center so we will
collect data from that hospital, analyze it, try to predict deterioration try to
essentially catch things which are going wrong before before they go wrong. Okay,
so when you say collect data what does that mean?
So all hospitals nowadays there’s a lot of history to it I guess but the main
driver in America was an act of Congress which gave strong incentives for
hospitals in terms of money in order to go digital, so all the hospitals went
digital and started collecting all their data on computers and then people like
myself and many other researchers realized well you have all this data
collected on all these patients and we can certainly use it to learn
something. So initially there was a lot of epidemiology done understanding
different treatments different outcomes for those treatments and now more and
more the systems are getting better the data that’s being collected is being
collected more regularly you know with wearables and that sorts
of devices and now you’re getting to the stage where you can say okay well, maybe
we can use this data to sort of predict the future which is essentially
what we do now. Okay, so the wearable I’m wearing right now is being-
So wearables are a good explanation, good way to understand the increased
data collection. You don’t have them in hospitals really yet, a lot of people are
doing pilot projects but it’s going towards that. It’s more along the
lines of at least where I work a lot in the intensive care unit and in
that environment you have lots of devices monitoring the patient
like a wearable does but you better accuracy and that sort of thing,
better sensors that sort of thing. All right, so we can talk about this
stuff later, so I want people to know little bit more
about you so tell us about your background. Here’s a question, who
did you hang out with in high school? Oh that’s taking me back, so I actually hung out with somebody named Carl a lot.
We would play Euchre at lunch sometimes, we would play with paper clips and we
would pretend to sort of have wagers but they actually banned us from playing
with paperclips because it could be a surrogate for money, I don’t know why
they thought 11 year olds and this was after we all went all-in with
like our thousand paper clips and they still were very serious about this
gambling policy so, it was a little bit silly but well- But you said you were
11 at that time? Yeah, so I was living, actually my nickname was 12, I guess
that’s the year I made the most friends.
I kind of embrace it, I think I was and I continue to be quite
self-deprecating so I had when we all bought our sort of class sweaters I got
12 stitched onto the other arm. Sort of embraced it a bit but it was good fun, I mean it was kind of like my calling card like my shtick is that I’m a little bit
younger than everyone so always making jokes when people you know are talking
about their driving test I’m like oh what’s a car, you know? And so
I moved around quite a lot I actually was in Florida for a little bit of my
high school education, then I moved to Grimsby Ontario, and went to Blessed
Trinity, graduated there, went to McMaster like you said a little bit earlier than
some. How old were you? I was 15 and I wasn’t really sure
what I wanted to do. So happily, their engineering class is the same in the
first year, you just take physics, you take chemistry, you take all those
classes, and then second year on where do you specialize, and it’s somewhere in
the first year that you actually have to decide what you want to do. So somewhere along there, they had the new degree of biomedical engineering which I thought
sounded kind of cool. I honestly didn’t have strong idea of what it was but
it definitely sounded interesting. I knew I wasn’t interested in civil engineering, say. Right, but you always have like engineering tendencies. Yeah, definitely I
was always going to be an engineer. My dad and I have been playing with
computers since I was a kid so kind of fixing things, problem solving, I would
play these games, I actually looked it up the other day, Museum Madness was a game I love to play where you’re this kid in a museum and you have to solve all these
logic puzzles to prevent the aliens from taking over the museum or something
silly but it was like an education game that I was
introduced to and I think a communications technology class came out in 1994 and I actually wanted to introduce it to my niece because I
thought it was such a great game, you know? Problem solving, so yeah, I was always
going to be an engineer for sure. Right so when you got to McMaster you kind of just found your way into a biomedical
engineering. Yeah, I would phrase it that way I def- it was a very
new program, it sounded interesting I mean I wasn’t sure what to
make of it and I think biomedical is actually quite broad if you think about
it. If you think about what I do which is almost purely digital and sort of
digital signal processing versus making a prosthesis which I think it’s a much
more tangible more common belief of what biomedical engineering is. It’s kind of like
Alita, the recent movie where they made that arm? Yeah something like that
I think you know when you say biomedical engineer, I think it’s someone
who’s building those and I definitely don’t. So what kind of what
kind of Biomedical Engineering, you said digital- Yeah so my undergrad had covered
quite a bit of ground and so there was a bit of modeling of
systems, of biomedical systems, how they work. There were some Medical Physics so
how does an MRI scanner work, how does a CT scan work, and there are a
few sort of basic medicine classes so Anatomy that sort of things so it kind
of covered a broad spectrum. Happily, it was the second year so I
didn’t quite get the totally new curriculum where they weren’t exactly
sure what they were doing by the second year, you know they had tried it and I
remember the first year had to take Ecology, we didn’t and happily
because I don’t think I would have done well in ecology. A little bit out there, so yeah I kind of fell into it. I always was kind of
interested in human body and medicine and how that system kind of works and I
guess that’s been a driver throughout in terms of what I chose to do. And
so how did that feed into you ending up in Oxford? Yeah, so I was
finishing my degree, the third year, fourth year of my
degree, I think were the most interesting for me. I think especially the fourth
year, so in the first few years it was very much a general I would say mostly
an electrical engineering degree with some medicine thrown in there and then
in the fourth year you really sort of specialized in those courses like I said, Medical Physics, and I got really interested in that and
then I wasn’t really sure what I wanted to do, I didn’t feel like joining the
workforce and my dad and I had a conversation and he said oh why don’t
you sort of continue, why don’t you get a PhD, why don’t you try this program at
Oxford that I’ve been reading about so we talked about it. I think actually,
initially I was thinking I would do the master’s program there and
mainly because it was the same idea, my fourth year they were teaching sort of a
broad spectrum of biomedical engineering type research and I figured I would get
a little bit more exposure and then I would find something I’m really
interested in and I run off with it and so I was originally going to do the
master’s program but they suggested, and my dad suggested well you could
also try the Ph.D. program as well, and I got into that, which was great.
Was it like a crazy process of applying to this? No, I mean I don’t think it was
quite, it’s like any other university, you have to get the transcripts, you have to
get the letters. Happily, I had some nice professors who gave me some
letters of recommendation even though you know, now that I look back at it with
a bit more experience, I realized I was, you know one of the
undergrads who comes to you and is like oh, could I have a recommendation and
they were in the class, so very grateful for them for helping. I also did
a summer. But you always stood out as
being like- I think it was obvious because I was very interested in all the
classes. I think I was still a little bit shy so, and I lived off campus so I
wouldn’t necessarily sort of arrange meetings outside of class but within
class and within office hours, I think I definitely stood out and so- I
mean like you start high school when you were
11 yeah and then when you got to McMaster
you were like what 15? Yeah, so in their classes I would have been 17, 18 so
I think they would have noticed that for sure. Hard for me to really say that and
be objective, but I do remember having a lot of interesting discussions
with my professors and so I think they picked up on that. I remember
that one of the professors I asked for a recommendation from and I finished the final exam in an hour and a
half when it was three hours, so I think he probably remembered that as well and
just remember. Wait, you finished it in an hour and a half when that suppose to be three
hours long? Yeah. So you were whiz kid of the class. Yeah, I can’t say that, I don’t want to- Objectively speaking,
would you be the whiz kid? I nearly got perfect in my last
year, nearly I was one grade off- So you were the whiz kid. Yeah, so unfortunately, I
would have gotten a very nice scholarship if I had gone perfect but-
Really? What’s the difference? Oh, if you get all twelve,
McMaster has an unusual grading system where everything is out of 12 so if you
get all 12s I think you get a special scholarship and I got one in 11,
I think, in a communications technology class that, still not so great
at how radio is working, well I kind of know. Right, but everything
else was 12. Yeah, and like I said it was a
fun year. I think it really helped that Wait, so socially, how did that work? You were, what, 16, I mean you entered a big mess when
you’re 15, and that’s when everyone starts drinking and partying, so
what about your social life at that point? I remember sitting in class,
my second year, and two people were walking by and I know that they
weren’t really friends initially because one was sort of a laid-back cool person
the other was quite a studious academic and they were talking and the sort of
laid-back kind of cool kid invited him to a party, and I think invite him to kegger, and the guy was like, what’s a kegger? Like oh you’ve never been to it? And he’s like yeah I kind of had a lame first
year and so I’m kind of like that studious kid except probably for the
entire… No, it wasn’t that, I don’t think it
was because people excluded me or anything like that because they
thought I was a different age. I honestly was just not that extroverted. I was
quite quiet, I kept to myself mostly, I lived off campus so I would go home, I
would take an hour-and-a-half bus home either at 6 or 8 p.m. depending on the
day, pretty much every day and so, socially, I was fine but I
definitely didn’t party so much, I didn’t really get into that lifestyle.
Right, until Wolfson. Until yeah. So you went to
Oxford and you would be how old at that time? I was 19 going on 20 when I
joined Oxford. And so you entered Wolfson College, which is the graduate
college, that’s where we met. Yeah, great college. Yeah, and so how did
that transition go, from going from like McMaster living off campus being timid,
not really know what a kegger is, and when I went there, you were running
the bars. Yeah, I would say a lot of personal development happened for me at
Wolfson and at Oxford and that’s why Oxford in general has got a very special
place in my heart, and for that. So the first year, I remember moving, I’m
surprised actually thinking back to it because I wasn’t scared or you know
afraid or nervous or anything like that. You just went in and did it. And yeah, I kind of went for it. I remember being on the plane and talking to the flight attendant and in
that moment I think I was like, oh wow, yeah, I guess this is kind of a big deal
and I got a little bit nervous. What made you feel it was a big deal,
after talking to this flight attendant? Mainly because she was saying like, oh
you’re so young, you’re going to school, that’s so nice, and I was like, oh yeah, I
guess I am going to, and she’s like, oh, what was it like leaving your family? I
was like, oh, it was okay, I guess it’s someone who I didn’t
really know talking to me about it and having to explain it to them, I guess you
sort of realized when you say it out loud, oh I guess it’s a big deal to sort of, you know, that was the first time I had left home. A lot of people leave home for their undergraduate but I was at home the entire time and so
that was a big deal, that was the first time I ever lived alone. It was not a
massive transition because, as you know, they have accommodation within the
college, so the college system I think, was very beneficial for me, and I’ll just
explain it, essentially you have like a residency but it’s more than a residency,
it’s like a community and sort of living in the residency in an accommodation where you had a lot of amenities the
laundry was right there. I think it made the transition quite easy. You know,
I wasn’t exactly moving to some seventh floor apartment in New York where I
would have to go down this street to do the laundry. Which is what you are doing in Boston now. Yeah, that’s important but it was definitely a
smoother transition right away, and I had to do some adaptation you know, a
little bit of shame to admit I didn’t do my laundry regularly, my mother helped a
lot with that. How about the cooking though, I remember we had a
conversation about the stuff you cooked. Yeah, so I tried to save some money as
well, so I had that in mind, so I actually got into baking a lot, which was
interesting. I would bake my own bread and because it was really cheap and as a
side benefit it was tastier than the store-bought bread or at least I insist
it was. I guess it depends on the batch, so I started baking quite a lot and somewhere around my second year I definitely got into
cooking and kind of having fun in the kitchen trying different recipes. I met
my, now wife in my second year and so I would really go all out on a special occasion, like the month anniversary or the year anniversary, I
would cook like a four or five-course meal. I think the best I’ve ever done was
on one of our, and it sort of became expected, because I kept doing it, and so, but I like the surprise factor so once we, you know, I met her in town and we went for a walk in one of the colleges and
sort of very casual, very relaxed, and got home around 6:00 or 7:00 and when we got
home I had a four-course meal which was kind of ready but I’ve had everything kind of
cold, like a cold soup, that kind of stuff and so I surprised her
again, and she did. Because she was expecting- Yeah, like how would I, we just spent the whole afternoon hanging out. Gotcha, we’re having a cold meal. Yeah, exactly, well you know, and I heated up the main, you have to have a hot main. Right, you got to have something hot. It was fun.
So, when you got to Oxford, you were still 18, 19 and you were there
for how long? So I was there for five years. I want to say, I did four years for
the PhD, and then just to make this timing a bit easier, I did a research
assistant position while Penny, my wife, was finishing her
PhD so I was there for five years, and the transition was, like I said, it was
interesting, I think, what definitely held to me was the first year of that PhD
program. So PhDs in Oxford are usually three years, but I entered a
special program where the first year was sort of like a taught course and
they essentially spend one to two weeks going over an area of research that’s
done in the department, so they will spend a week or two, I think the
first one was on nanoparticles, and there was one on tissue engineering, and so you
sort of have a 9 to 5 schedule of being taught or working through a workshop
on this topic and that definitely helped me, I had a lot of
structure moving in, and I think PhDs in England in general, they are three
years, they sort of just drop you right into research and I think I would have struggled at it.
So none of the coursework. Yeah, I think I would have struggled with that
for sure, so having this sort of structure being introduced to all the
topics, having a week or two to see, like do I like this, would I want to do a PhD on this, that was very helpful. It made the transition a lot easier.
How was socially, like personally, because you know, it is your first time leaving,
and how did you integrate into Wolfson because everyone
there is a graduate college everyone’s older, so once again you were 18, 19,
but people there are like, 20s? That’s true, so definitely and that’s helpful right,
but I see it from the other angle, most people there are around the same age or
at least the same life stage, they’re trying to get a bit more
education and they’re very excited to be, you know somewhere new, somewhere as
beautiful as Oxford, excited to meet new people, talk about their interest, because
you don’t just have engineers in Wolfson. In one college,
you’ll have literature majors, you’ll have philosophy majors, international
development majors, so you get that really kind of interesting mix. In my PhD I was definitely still the youngest for a
long time but it moved away from defining me to just being kind of like
an interesting quirk, like one of my interesting quirk, everybody sort of had
something interesting, you know, I was talking to someone who had never seen snow
before and we had snow that first year, and it sort of took a
backseat which was nice little, kind of cute, my first year, my roommate
Nisha who actually, I’m still really good
friends with, I went to her wedding a few years ago which was amazing, she
sort of was a little bit like a big sister to me
I guess which was also very helpful. Like took you under her wing?
Yeah, so that still happened but I think throughout the PhD less and
less and like I said it was kind of like a coming of age in a way. Right. So tell me about this Forbes 30 under
30 accolade that you have, I mean, that’s a pretty big deal. Yeah, I guess so,
I was almost blindsided by it to be honest. So, it happened very quickly that I
received an email, they said, oh you’ve been nominated, we’re going
to have an election and just to let you know and then they said okay
well you’ve been selected, can you send us a 5,000 by 5,000 resolution photo in
two days. Wow, it was just like they selected you out of the blue. It was very
clear. Who nominated you? I have no idea and I’m so curious to find out I really want
to know because I’m definitely an atypical choice, I think, so I look
through the rest of the- Other candidates? Alumni? They all sort of are in startups. Very
entrepreneurial, doing that sort of thing right, and I think I was the
only academic, I was the only one who was in the academic system so I’m kind of, at
the very least, from that, I’m a very interesting nomination. I was very flattered, I’m very happy that it happened because I
you know I have my area of research which I push and it’s sort of like
validation that that’s a legitimate area you always have this
sneaking suspicion that you’re gonna work on this for five years and you know
you picked the wrong horse in the race. So that was very nice. Right, because that’s always a problem, you do research, you
throw five years and then you realize, someone develop a better method
or like, oh it’s obsolete already, or you know, your approach was the wrong
approach. That happens a lot, that happens all the time. Right, because in the sciences, it’s the worse thing, one of the terrible things that
could happen in science is like the hypothesis I’ve been working on is, oh you’re just false, it’s false. Yeah and that happens. Right and that happens and that’s life. Yeah, so receiving the award was wonderful. Right, so how old are you right
now? I am turning 30 this year. You are turning 30 so you’re officially- Getting to the big three. Right, so your research, what you’re pushing in terms of your research, how does machine learning or AI, when I said AI
you told me it is like, it’s machine learning, what’s the difference? I guess when you’re talking among scientists, you call it machine
learning, and when you’re making slides you call it AI, so that’s a little
bit flippant. So essentially, AI is kind of a rebranded
term so it becomes a little bit confusing. AI used to have a very
specific meaning of artificial intelligence like the movie AI.
Right, the Spielberg kid. Yeah the little kid talking. Or I, Robot, Will Smith. Or world classic, Asimov sort of stories. In all of these reasoning beings which were
created artificially, I guess- What’s an Asimov story?
I know what that is but can you explain.
Oh iRobot was based off a few books by Isaac Asimov and he coined the Three Laws
of Robotics which are very famous, eventually, he added a fourth because he
started to believe that the the three weren’t sufficient and you know I won’t
spoil the story but definitely should read some of those stories if
you haven’t and so AI used to mean this sort of creating intelligence
essentially. It’s in the name and I don’t know where it changed, but
essentially there’s a huge resurgence in neural network research. So neural
networks were this model for making a classification that you would
want, so say you have a picture and you ask the question, is there a cat
in the picture? A neural network was a type of approach of telling you if
there’s a cat in the picture. Very famously, there was a TV show which
made an app to classify if it’s a hot dog or not.
So that’s another case of a neural network and for the longest time neural
networks were one of the tools you had in your toolbox. There were neural networks, there were support vector machines, and they sort of, I had a lot of
excitement around them. The interesting thing about them is they have a
biological basis, they’re sort of trying to emulate the way the brain
works in terms of, you have these little neurons, and these neurons are
interacting in some way, and if sort of enough of the neurons in your brain fire,
then you know you feel something or you decide something right, or at least
that’s kind of the loose logic. Right, so these neural networks you’re talking about are all digital, are on the computer. Yeah, so
they actually were called artificial neural networks for the longest time. This was when? So they started around, I would say 1960s, Rosenblatt and then there’s some work, I think they
call them multi-layer perceptrons, they like to change the name just to confuse
you every so often. It’s basically rebranding. Yeah maybe, I
think artificial neural network is a pretty good name because it’s really, we based this model in biology but it’s it’s not biological so as an
example in biology, you have action potentials which go between neurons and
so whether an action potential goes or not, that’s the neuron
firing, but also the amount of time it takes is also a factor and we
don’t consider any of those factors in artificial neural networks
and so they are certainly a simplification. Alright, so a lot of the stuff is just going over in my head, can you speak in plain English?
Yeah, so essentially we had this artificial neural network model, I say we,
this clearly predates me, and there was a lot of
excitement around them and then this excitement receded in part because
someone showed that there was this really simple toy problem that they
couldn’t solve, that they just fundamentally couldn’t solve,
and there were caveats to it but most people didn’t pick up on the caveats and
I think you know the combination of a lot of excitement like we’re emulating
the brain and then we like there’s this really basic problem
that they can’t solve had people sort of pull away from that field and then
very recently 2010, 2012 some researchers combined
oddly enough graphics cards with these neural networks because they realized
that graphics cards for computers are being built to make these
amazing displays and it turns out the way that they do that computation is
very similar to the way the neural network does the computation. And I don’t think it was obvious to too many people before but essentially it
turned out that the key to get these neural networks to work or some device
that could operate quickly and a lot of data. So pretty
much the hardware at that time wasn’t able to support this kind of- Yeah, and the
data, we didn’t have anywhere near as much data back then. Right, and now with the GPUs that we have, which is like, I know what a GPU is
because it’s used to process video games or movies or whatnot right, and
that’s what is now powering the ANNs. Absolutely, so there’s these three factors which came together very nicely and there’s a very
famous sort of task that computer science researchers and computer vision
researchers focused on and the first application of the model was at
50% lower error or something astronomical, like people were not doing
well at this task, and then out of nowhere this team comes out with this
neural network approach and it’s incredible and it’s really just
taken off since then and so I think the combination is, now I’m realizing this is
a very roundabout answer to the question, I think it was the combination of the
fact that we use artificial neural networks to sort of do some really
amazing things which feel like only humans can do. I think people have said,
oh well, this looks like artificial intelligence and so now what sort of
happened is now, machine learning and artificial
intelligence are kind of interchangeable and people who are really focused on
that sort of I, Robot, you know, life-like robot they call that general artificial
intelligence. Okay, and what’s the other one? Just artificial intelligence. So if we’re
talking artificial intelligence, we’re just talking about teaching a
machine to do something. But if we’re talking general artificial
intelligence You are talking about terminator stuff. Yeah, we’re talking about like teaching a machine to reason, maybe have
intelligence, have some you know concept of what a world is. So here’s a
question, I mean are we going to be replaced by machines anytime soon? Let’s get straight
to it. Yeah, so probably next year January, no. So it’s an interesting question, so I’ll
explain a bit about my background. Recently I’ve been working a lot with
radiology and radiology data, so that’s, if you get an x-ray, if you get an MRI, if
you get a CT scan, you know these big machines which essentially take pictures
of your body, they generate a lot of images and there’s a medical profession
which is dedicated to interpreting these images. So they look at the image, they
look at you, and they understand how old you are, maybe what your symptoms are, and they say okay well given this person and given this image I think the diagnosis
is this. So very common one is you get a chest x-ray, you have a cough, you are
feverish, a radiologist will look at the chest x-ray, look at your history, and say
okay, well this looks like pneumonia, because they can make that
interpretation and so there’s a lot of fear and discussion in
that community about the job being taken by AI, and I think that
it’s interesting and I cannot say for sure your job won’t exist in the
future, but I can say that looking at the history, you know we tend to adapt, so
there’s only one job which was replaced by computers by transistors and that was
the job of a computer, like the person who actually computed, their job no
longer exists, but the rest of the jobs, they
still exist in some degree, maybe there’s fewer of them but then entirely new
careers opened up, so I think it really depends on the
career. There’s this interesting paradox with robots where AI or whatever,
you know they can do millions of mathematical calculations per second,
but picking up a pencil is really difficult for them. There’s
this weird sort of, but it’s easy for us, so it can be very difficult to reason
about what kind of job is replaced. If your job is solely, you know, crunching
numbers, then yeah, that that’s not a good sign, right, because computers are very
good at that but if you have to move or something, you know, it’s a different
story. So lately you’ve seen those Boston robotics right, they put these videos out, like robotic dog, or even this robot
that’s able to like open the door. True, deal with dishes or whatnot, I mean. But
interestingly, I think a lot of their technology isn’t actually based in this
sort of neural network approach that people are afraid of. I think a lot of
their approach is sort of classic robotics and their little demos look
good but I don’t think they are anywhere near having a robot that can
they can deploy in your house to do things, so I think it really depends on the idea that you know these AI is just gonna
come in and replace all the jobs is definitely over the top.
One of the most famous researcher said, a few years ago, oh, they should stop
training these radiologists because their job isn’t going to exist and he’s
actually sort of gone back from that opinion now I think because people
realize there’s a lot more than just processing data and giving you an output
which is what our models can do. They can process data and give you a
classification, tell you it’s a cat, tell you it’s a dog. That’s really useful but
that’s not the whole story. A radiologist might read an image but the
radiologist also looks at patient, reads the history, and these sorts of
reasoning, I don’t think we have a good way of actually coming up with that,
so you know, I wouldn’t recommend you know taking a job as an
Excel spreadsheet But take driverless car, that’s something that we have been working on for five years,
a decade now, there aren’t driverless cars yet because
it’s a hard problem, it is a hard problem, even though
most of us can do it, and so if it’s that hard to come up with driverless cars
when you have Tesla collecting tons of data from all these Tesla cars
around the world then you sort of start to realize like, okay well it’s not going
to be as easy as you know snapping your fingers. Right, so in terms of like
job security or you know if you’re doing like Excel spreadsheet stuff or you’re
like driving Uber or Lyft and we’re kind of in the not so good situation,
but if you’re like doing a little bit slightly more complex stuff that
requires interpretation, emotional work then it’s much harder. Yeah, so here are the things that, here are the take
away messages, you know, one, don’t be driving something. There’s so much capital put behind driverless cars, it will happen. There’s
tons of benefits, it’s a hard problem but it’s clear that eventually driverless
cars will be safer than humans, and it will save us a lot of time. And a driver will
be a luxury by that point. Yeah, it will have a lot of interesting effects,
if you have a driverless car then maybe living outside of the city is
nicer than inside the city. You get rid of a city problem. So, we’ll see. There’s a lot of
speculation area but at the very least, you know, you don’t want to be a cab
driver. I actually think, another one, which
is a bit surprising is anything that involves manipulating pixels.
Graphic artists, something like that. Like Photoshop stuff? Photoshop stuff. I
think those jobs will be fewer and far between because a graphic art and sort
of you know manipulating pictures is a very, you know it’s a challenging task but it’s a domain where we have a lot of
data, and we have a lot of pictures on the internet, we have a lot of art, we have all that stuff and all the computer science researchers are using
that to build fancier models and its really incredible. Just this week, a group
released a model which can make Mona Lisa speak. Wow. So you can
see a video of Mona Lisa speaking, essentially, and it looks very accurate.
So I think any job that involves manipulating pixels, honestly,
also, manipulating sound waves, is probably suspect, so voice, voice acting,
is probably going to be done a lot by AI in the future. And that’s why we
see all this, like you know Carrie Fisher, you know she’s still acting, even though she’s gone, she’s still in the Star Wars. That kind of stuff? Yeah. It will only get better. So definitely, I feel like graphic art is
not a good career to go into because there is a model where you can just
color green in the background and then do a giant white blotch and then you hit
a button and it turns it into a mountain in the forest, just like that. Wow. There’s an
actual model where it’s just green, white, boom, it turns into mountain. Yeah, you just paint the colors and it says yep, I’ll do scenery, I’ll do mountains, I’ll do the
sky, so those, that’s not a career you want to focus on. Right, so what people are saying, oh, don’t worry about AI, as
long as you enter art stuff, you’re fine. I don’t think that’s true,
at all. I think that’s where some of the biggest sort of
upsets will happen is in you know voice acting for video games, right? All of a sudden you can replace that with a machine and I think art for
a lot of a lot of different games, 3D, modeling, that sort of stuff, I think it’s
quite a difficult problem, but I think it’s a fully digital problem,
and I think those are the areas where our current models will excel, is if you
can keep everything fully digital. I think once you start
interacting with the real world and you run into issues where the model doesn’t
have all the information it needs, then you’re in trouble. So for example, in
in my field in healthcare, people are trying to build models to make decisions but the reality is at the current stage, the model can’t see everything. The
data isn’t accurate enough to really drive the decisions and so an
interesting example would be, the models I build would have your heart rate, your
blood pressure, your creatinine level, all these sort of lab measures, but it won’t
know the color of your skin, because that doesn’t exist and if you look at someone
and you see the color of their skin is off, you know they’re sick, but
the model is totally blind to that. So I think you’re sort of thinking of
jobs where everything is completely digital, and if that’s the case then
you’re a little bit worried. So if it’s something like dancing, you’re probably in a pretty safe position. Yeah, I mean, I think, maybe we will have a dog which dances, you know, a robot, right, but
I think those sorts of, actually, in general, the biomechanics of the
human body are very difficult to replicate. The knee, the hip, you know all
these things, anyone with a hip replacement will tell you it’s not
the same because they’re actually very very impressive and complicated systems
that we don’t fully understand, so I think a lot of that sort of tactile
activities, they wouldn’t be replaced and I actually don’t
think people would, well, maybe they would all want to watch robots rather than humans.
I think there’s something to watching another person who’s hone their skill
over in sort of a lifetime. I think there’s something to that at least
I would hope. That our robots simply won’t replace. Yeah, exactly. Nobody wants to
watch the Olympics where the darts are thrown by a robot, right?
Yeah, that’s fair enough. That’s true. If you’re gonna recommend people who are listening to this podcast to equip
themselves with skills for the future, what skills are we talking about, what
would you recommend people to prep themselves up for, so that they can
be able to flourish in the future? So, hopefully I’m not too biased, but I
definitely think a technical career focused on programming, or software
engineering, I think those are winning tickets. So coding? Yeah, so essentially computers are our general process objects and
they are sort of like, one of my professors made the equivalence that
they’re like very dumb file clerks, but they’re very fast. And I think the
ability to tell a computer what to do, to automate something, I think that’s key. I
think that’s a very useful skill, and there are people who can now automate, there are many of the activities in their home with computers and so I think
I think that really is the technology you want to focus on the sort
of career, it’s very common, it’s very in demand right now, that’s
what I want to say, a skilled tech worker, have a
very nice salary, very comfortable lifestyle, and I think it will continue
like that because I think there will always
be a need for someone to design the computers, design the systems, and that
sort of thing. So it sounds like machine learning or AI, it’s more like a
tool, and the tool is only as good as the person who wields it. Yeah, it’s interesting, there was a criticism by Lady Lovelace that
machines can never be truly intelligent because they can’t be creative, they can
only do what you tell them to. Is that true? It was a proposition and actually Alan
Turing in his landmark paper about can machines think, he
addresses it but it doesn’t fully answer it, and I think it’s still a reasonable
objection that at least with our current models, can they really surprise
you, I mean they can surprise you, but can they be creative, can they come
up with a new solution. Most of our models right now, you just give it a lot
of data and if it’s seen it before it will get
the answer right, but a problem we face a lot is that when you move to another
hospital which practices medicine slightly differently, the activity, it’s seeing different data and all of a sudden you have no idea whether
it’s going to perform well and 99% of the time it performs terribly because it
hasn’t seen that before so there is that question. How much of the error is human
made though, right? In terms of transferring the model, well, it’s just a very reasonable inference about a change of
practice, a change of data collection that any one of us could make, that the
machine can’t, so if all of a sudden instead of measuring heart rate before
blood pressure this next hospital measures heart rate after blood pressure.
Something as trivial as that can trip up the model because maybe it’s seen the
only times that they sort of flip that in the hospital was when things
were going really bad and they were rushed, so it’s going to think, oh you
know that’s a signal for things being really bad, when in reality, they just
practice it almost insignificantly differently, so I think that’s still kind of think that’s an objection to
AI in general to whether we can actually make you know systems which can reason
which you can have a model of the world and adapt that model. I don’t think we
have anything that does it near that. Obviously people are working towards it
but it’s a very open question. So what’s the most important point that you want
to make to us today? So there is this very good article that I highly
recommend everyone read, I don’t remember the exact title but it was roughly, There
is no fire alarm before general AI, for AI, and this article makes a very true
point about science in general which is that it tends to be incremental.
So science proceeds down a path that we expect, use, you have a method
that gets slightly better, and that’s most of scientific progress but there is some progress which comes
out of nowhere and there’s a breakthrough and you have a rapid advance that no one really would have predicted. There’s a
very poignant example of this in 1943 when Pauli who was the expert in physics
at the time said that he didn’t believe nuclear fission was possible and then in
1945, it clearly was very possible in a very dramatic fashion so even though we
have experts who are arguing about what will become of AI, you always have to be
a little bit skeptical because there’s always the possibility that you have a
radical breakthrough that none of us would have predicted and I think it’s
it’s a very good point to keep in mind. So you’re basically hedging. Yeah, you could say that, I mean- You only need one black swan to throw everything right off. Yeah, the money
is on that we don’t have any breakthroughs, that we keep having incremental progress,
you can sort of predict which fields will be most disrupted, but
there’s always the the possibility that researcher from a university
comes up with a totally new way of doing things and all of a sudden it’s a
lot more dramatic of a change. Here’s another question, last
question, I just thought about this one the spot. Corporations versus
universities, where is this research happening where
is it happening in the corporations or is having a very- It’s a very interesting
question, and it’s a change in the dynamic which I don’t think anybody
fully understands. A lot of this research actually requires a lot of
money, because it requires a lot of computation and it’s shifted into
these companies who have research arms. So Google had purchased DeepMind and it’s a big research organization
actually and you’re not sure if it’s a commercial venture or a research
organization but at the very least, at all the top quality top conferences
they are there you know publishing very high quality
articles and you’re also starting to see career paths emerge which are 80 percent
in a company and 20 percent in the university and with that 20 percent of
time you can do what you like in the university, you could mentor a
student, you could teach a course and you could sort of do, and that’s a very sort
of new development, I think research was traditionally in all the universities
and it shifted into these companies because you simply don’t have the
capital. Yeah and and they do and there’s a lot of advantages, there’s a
lot less, surprisingly, bureaucracy in these corporations if you’re on the
ground. A lot of what you do as a professor at a university is trying to
bring in the money, applying for the grants, and so it’s quite attractive to
join a company and not have to worry about that. Just give you this money
and go do your research. Yeah, but there is a question of incentives though.
Exactly, and the whole integrity behind the research for the sake of
research, right? Yeah, exactly so it’s unclear
whether this is a good change or a bad change and I think there’s a lot of good
quality research being produced by the companies today that wouldn’t exist if
if they didn’t try it, at the same time there are a lot of people
who feel a little bit left out or feel that it’s difficult to make progress
because they don’t have a vast army of computers available to them like these
researchers, so it’s hard to say, I would say I am not
entirely comfortable with companies becoming sort of the leaders of research
because of the incentive problem, because- They’re accountable to stockholders.
Right, yes, and they’ll push the research in the direction that you know,
is more likely to produce shareholder value and
that’s not necessarily the direction we want to go and the only reason why
neural networks, maybe it’s not the only reason, but a main reason that neural
networks are still prominent and still used is because researchers in Canada
were supported by CIFAR grants to continue the research when everybody
else said this isn’t interesting anymore, we’re done,
and they continued and they made a number of very important findings
but that was because they weren’t at the whims of some shareholders,
otherwise they would have stopped for sure. So that’s interesting, I can’t say I’m super happy with it but I have to
acknowledge that it’s been very beneficial to the field. Right, so the
future is it’s pretty much all uncharted territory. Yeah, it does seem
like it will just continue to move to corporations. Academia is outdated in a
lot of ways, the application process is outdated, the method of disseminating
information is outdated, the method of acquiring funding, rewards those
who are arrogant, I would say perhaps mislead you, perhaps are too confident in
the decision, will try and sell you, I don’t have to throw stones, I
know my laboratory is funded by a grant where in 2001 they were saying they were
going to have algorithms which you know made decisions for patient care in 2001
its 2019 we still don’t have those algorithms, so the sort of
grant mechanism does encourage you to over promise, I guess is the way I’ll phrase it. That’s clearly not for everybody and that pushes a lot of people out of academia
who would be brilliant researchers so I think it will continue to move towards
corporations and I don’t think we know what the meaning of that is yet and I
think we’ll learn soon. All right, thanks a lot for your
time today. I appreciate it. Thank you

Leave a Reply

Your email address will not be published. Required fields are marked *