The Future of Cognitive Computing



[ MUSIC ] [ APPLAUSE ] KELLY: So, let me begin by…with a little
bit of a word of caution or a way to think about this, and I urge you to try to
keep this in mind throughout the day. It's very easy as we get into areas of
cognitive computing and artificial intelligence to rapidly drop down and start to talk
about deep machine learning and algorithms and some really exciting technologies. And for those of you that know me, you know,
nobody loves technology more than I do. And it's a fascinating, fascinating area,
and lots of that will be talked about today. But I think the thing that we must
keep in mind is towards what end? Towards what end? What is it that we're really trying to do here? And I would argue that it's
all about the outcomes. It's all about changing the world and
changing industries and seeing things and getting insights into things that we never
have been able to get our arms around before. And so, I would urge you throughout the day as
we, again we talk about the various technologies and opportunities to keep
thinking about what applications, what can we do with this technology? How can we impact society and the human state
in ways that we've never been able to do before? So, I sort of put it in the flip context, if
you will, what is the price of not knowing? We had a discussion last
night at the Churchill Club and someone asked me, John,
what is Watson worth? What is the market value of Watson? And I said, I don't know. It's really…I mean, I could
calculate it based on IBM's business but I think it's much, much broader than that. You know, what is the price of
not knowing the cure for a cancer? What is the price, the downstream cost
of that patient towards end of life? What is the cost of not discovering alternate
energy, not drilling into the proper areas? All I know, as I said last night,
it's in the billions and trillions. You know, take healthcare as an example. It's a seven or eight trillion
dollar industry worldwide; three and a half trillion in the United States. And every estimate says that 30
or 40 percent of that is waste and inefficiency or bad outcomes in the system. So, just healthcare in the United
States is a trillion dollar waste — or, cost of not knowing. So, huge opportunity to apply
these new technologies. Now what are we really after here
and what is sort of fueling this? We all know all these things about big data, the amounts of data that's
being generated, et cetera. Another comment or question from last
night — and a caution flag, again — is it's very tempting also in addition to
talking about algorithms, it's very tempting to talk about what we're trying to do here
as replicating what the human brain does. That is not at all what this is about. When we started out in the early days
— as Guru will know with Watson — we were not trying to do what previous AI
researchers had done to mimic the brain. We were doing something very simple. We were trying to build a system that could
deal with this massive amount of data, because our human intelligence was
not scaling the way data is scaling. In a sense, there's a new Moore's Law and
that new Moore's Law is in the data space. And much of that data is dark or
invisible to our current computer systems. And that was the aha! moment or the awakening for us in IBM Research. We said, we need an entirely new type
of computer system to deal with this. Today, it's estimated that 80 percent
of the world's data is dark — meaning that we as humans in our current
computer systems cannot make sense of that data. It's either noisy or in formats
that can't be read. And furthermore, that by 2020,
that's going to exceed 90 percent. Very interesting numbers because if you…for
those of you what are…know anything about physics or astrophysics, there's such
a thing called dark matter in the universe. There's a set of matter and energy in the
universe that has not been observed directly but we observe the effects of it
in terms of gravity and things. Our optical telescopes and other our
telescopes cannot see that matter. This is the equivalent in the data space. Ninety percent of it we can't
get our arms around. Think about the solutions that are in that
data if we can get to it because we're in a sense just seeing a little slice of
the world when beyond that little slice, there's great opportunity and
probably great danger for humanity if we can't get our arms
around the rest of that space. So, in my mind, this is not a journey
to reproduce what the human mind does. Yes, we'll be inspired by what the human
mind can do, but that is not the objective. The objective is to analyze and garner
insights from that massive amounts of data. If we don't, that 93 percent will just keep
growing and we'll be getting such a minute view of what's going on in the world that we'll
really be in a very, very rough place. So, think about the industries
that could be effected by this. And I don't make these things up. We are working with every one of these
industries that I'm going to show you. Oil and gas. Huge opportunity. The industry spends billions
of dollar per oil rig. Often, drilling is in the wrong place. They miss it. Often the pumping from the wells of
the reservoirs is too much, too little, not optimized, a huge opportunity. Tens of thousands of sensors on these platforms. More than current analytic
capability can deal with. Huge opportunity to impact
the oil and gas industry. Retail. We talked a little bit
before about the massive amounts of data that's coming through social media. Think of the content and the insights that
can be garnered from that for retailers. As many of you know, we have a partnership
with Twitter where we get the big hose, if you will from Twitter of all tweets. Incredible insights into buying patterns,
preferences, where society is moving — insights that can be leveraged across
literally every form of commerce. The Internet of Things, again, I think
is one of the great next frontiers. Signal processing. We started with Watson in natural
language processing which let of course to the infamous Jeopardy! match, and today it's unmatched
in natural language capability. We've moved it to images and vision. But think about signal processing. Machine to machine data will dominate
the data scene in just a few years. And that is noisy, unstructured
and really a perfect application for cognitive computing whether it's connected
appliances or an inner city where you're dealing with security issues, where you're
dealing with traffic management issues. All of the things that are going on in the city. A perfect environment for a cognitive system. Security, another area that you
might not think of as a natural, but security is no longer
about building firewalls. Security now is about deep behavioral
analysis of people and systems — a perfect application of cognitive
systems to measure and predict behaviors and abnormalities and react to them in realtime. Energy and utilities. We already have instrumented many of
the meters in many, many countries, but the data now again remains dark. Very little is being done. It's a huge opportunity as we try to
integrate renewables in and start the feedback and altered behaviors of consumers. And one of the biggest, of
course, is healthcare. As I mentioned before, an enormous industry,
ripe for not only digital disruption — which was sort of what happened
with electronic medical records — but cognitive disruption as we
really bring in new forms of insight. And this is one of the industries where we
ever really doubled down our bets not only through electronic medical records, patient
population, healthcare, but also medical images. And I'll talk a little bit
more about that in a minute. But think about it, a million
gigabytes per person in our lives. Immense amount of information, and
in that data is really the secret to our own health and well-being. Transportation, another Internet of Things. We started with, you know, processors in cars; we're now talking about self-driving
or assisted cars. These devices will need to be cognitive. They will need to make realtime on-the-fly
decisions about the environment based on learning about the environment
and the driver behavior. So, every single industry
now is being swamped in data. Every industry is trying to find a way to
get at and access to that 80 to 90 percent of the dark data and get
insights differentiated. And I think that is really what's going
on here in terms of the turning point in our industry and all of these industries. So, think about this in terms of, we
are at an incredible inflection point. We are no longer just sort of
incrementally improving our IT equipment. The first era of computing was
simply a set of tabulating machines. Largely mechanical. We put data in through punch cards or
some other means of setting switches. We programmed it by telling the
machine what to do, and away we went and we automated basic human
tasks such as arithmetic. The second era of computing — which began
in the late forties and early fifties — was a real turning point where
we went to programmable systems. And the point in time where this really started
was where we had enough memory in the computer that we could put the programming
from the punch cards into the computer and let the system run itself
with no external programming. And again, it was because we had enough memory
in the system to put those instruction sets into the memory; and of course, away we went. And everything since the late forties
to today has been programmable. But again, as we looked at
this a number of years ago in IBM Research, we said, you know what? We're going to run out of programmers. There's no way we are going to be able
to program and keep up with the scale and speed and exponential growth of data. We had better take a different path. And that was what really got us on this cognitive computing
AI path in a major, major way. And we believe that we are at one of these very
unique points in time that only occurs every 40 or 50 years in this industry where we are
creating entirely new computer systems that do entirely different
things than the last era. This era will be more different from
the programming than programming was from the tabulating era in my belief. Think about also what happened with
that, you know, infamous System/360 which was…IBM had done small
numbers of programmable systems before that 360, but the 360 was a platform. It was a platform where we
mass produced the systems. We separated the hardware and software and
we create a platform that became the platform that transformed banking, airlines,
transportation, literally every industry in the world and remains sort of the
backbone of all enterprise transactions. It's important in this new era
to also think about a platform. Not a discrete tool to address one problem, but a platform that will
transform a number of industries. And that is how we are thinking about this
in IBM and what we're trying to achieve with our cognitive system and with Watson. So, it all began, it doesn't seem
possible, but it was almost five years ago, five years in February that this
infamous match occurred between Watson — the first really cognitive
platform — and human beings. Our goal that day was not to just win a game
show; in fact, it was a pretty close match for those of you that study that game. But we wanted to demonstrate that
we were going through a transition. We were going through a transition from this
programmable era into this cognitive era. And I've been amazed, frankly,
what happened since February 2011. This whole field has exploded. It has stirred the imagination of academia,
it stirred the imagination of industry, which I think is fantastic — because again,
this is not about one company or one capability; this is about creating a
whole new era of computing. Lots of stories I could tell about that match. I'll just tell you one fun one. I talked to Ken and Brad about,
how do you do what you do? I mean, these two human beings are amazing. Amazing. I asked them, well,
how do you know so much? Did you study? And both of them independently said, no,
everything I see and hear, I never forget. I remember it. I said, okay. That's pretty interesting. And then I said, well, when
you're asked a question, what reasoning process do you go through? Because I know how my brain works. You know, I'll start to think about
alternatives, I'll do lookups in my head. Both of them independently said, I don't know. The answer is just instantly in my head. So, they have complete memory and instant
recall of everything they've ever seen. Incredible capability. To beat these two humans, that system had
to be right 85 to 95 percent of time in two and a half seconds or it was lights out. That is a really tough problem in open domain. The reason we won is that we took
an entirely different approach — not a rules-based approach but an open
approach, machine learning, deep learning and very sophisticated natural
language processing. A completely different approach to the problem. And I think that's what's
required across the board. So, where do we need to go with this
then as humans, because we're often…one of the positives of that match was it really
set the stage for cognitive computing. It captured people's imagination, but
it also set up man versus machine, and that was not the intent at all. Where do we go? It's not man versus machine. Every study has shown that man and
machine will beat either man or machine. And that I think is a really key point because
of the different capabilities that we each have. So, we as humans have a number of capabilities which I'm not sure we'll ever be
able to really get a machine to do. Now, I hesitate to say that
because I said that before and we've gone on to get machines to do that. But some of these things relative to
intuition, compassion, moral values, unless they can really be quantified, I don't
ever see a system being able to do that. On the other hand, these massive
systems and immense capability. They'll have total recall. They'll be the Jenner and Rudder of computing. They'll have instant recall of
everything, source to all knowledge. Large scale capability for fact checking; and
with deep learning and the capability to start to reason, we can really get into discovery. So, I think the opportunity,
again, here is man and machine. And we are seeing this in every discipline,
in every industry that we go into with Watson. And the pattern seems to be that we as humans
have sort of a normal distribution of capability for whatever it is we're talking about. It could be simply a call center operator
or literally an oncologist, a cancer doctor. And the distribution is not surprising to
technical people, a normal distribution. What we're finding in this man plus machine
is that we can move that distribution. We can take the best oncologist at Memorial
Sloan-Kettering and make them even better. We can take the mean of the distribution
and move it and we can take those that are on the tail end of the distribution and
move them up to be as good as anyone else in the world by introducing
this man and machine. And we're seeing this repeatedly
in the financial sector and across other industries of the world. So, the secret is then how do we get
this synergy between man and machine? Now, since that Jeopardy! match, this field has lit up. Just lit up. And that's of course what brings
colloquium like this together. Lots of people are working on this. Lots of people are trying to do
image processing and finding pictures of cats on the Internet and whatnot. Lots of people are trying
to optimize buying behavior. Lots of people are trying to
do signal processing of voice or making voice recognition smarter. But each of these is really a point solution
to improve some sort of one-dimensional aspect of a business model or something
that they're trying to achieve. And it's wonderful. It's great. But it is "a" tool — a hammer
or a screwdriver from a toolkit. Very few, and really, other than ourselves
at IBM, I don't know of anyone that's trying to build a whole toolkit and a whole
platform equivalent to what we did with the System/360 back in 1964. We're not trying to just poke
at these individual problems; we're trying to build an entire
platform of capability for all industries with this cognitive computing capability. So, to do that, we took that large machine
that was Watson that won that Jeopardy! match which was about half as big as this stage, and it was so heavy it probably
would have fallen through the stage, consumed
85,000 watts of power. We took the Watson capability out of
that machine, brought it to our cloud, decomposed what was at the time one system
— a question and answering system — that had basically five technologies
understand it. It had other bits and pieces,
but these were the big ones. We took that capability, brought it
to our cloud so that it could scale. And then we proceeded to offer that
as a service but not just that. Build out a suite of services on the
Watson cloud that are composable assets. So, in a sense, you as a developer
can go in and pick and choose and construct a mini Watson for
a solution for your problem. And this has been incredibly successful. We have held on one extreme we've held
hackathons where…I still say kids, but you know, young people in 12 hours are
composing meaningful solutions with those assets on the Watson cloud in a day or two. Very, very powerful sort of time to market. Now, this was a fundamental decision that
we made in IBM right after that Jeopardy! match. We could have started
just selling Watson boxes and we could have sold a lot of Watson boxes. But we decided that, no, we were
going to make this cloud based and as a service composable for all industries. And that has been what has guided us
over the past few years as we build this. And as you can see, what
we have a couple of dozen or so of these services available
today, this will grow. This has become the platform for our ecosystem. We have hundreds of partners now
building with this capability. Dozens of universities engaged in this,
hundreds of universities engaged in educating and how to in a sense program
or assemble with this language and to develop the underlying
skills to use this. And as you can see, our plan is to develop
not just dozens but hundreds of these services on the Watson cloud as fast as we can. We have a pipeline of these services
and I think you're going to find that this is a very rich environment. So, when you step back, then you say
okay, we're building this platform. We're building all these capabilities. What is the essence of what we're trying to do? What is the essence of this
cognitive capability? First is learning at scale — learning at scale
in data, learning at scale in those solutions. It's about reasoning or developing insights from the data whether it's
natural language or images. It's reasoning over that
with purpose, with purpose. With a goal. Whether it's a radiologist
or a financial advisor. Someone that is reasoning over data to get
an insight with a purpose to take an action. And then finally, to interact with
humans because as I said, in the end, it's this magic of man and
machine that I think is going to produce really leapfrog
capabilities going forward. So, what gets me excited and has kept me helping
to drive this project forward is to really start to think about how we can rethink
what's possible with this technology. It's no longer about just automating or
programming these systems for ERP or back office or mobile phones and connections. The possibilities with this are immense. Medical imaging. We in IBM are absolutely convinced that with
Watson capability, with image analytics, with machine learning over these images,
we can change the course of healthcare. Vast majority, on the order of two-thirds of
health information is contained in images, x-rays, MRIs, CAT scans, et cetera. We know that the diagnosis associated with these
images by humans is not what it needs to be. Think about a radiologist who sits in a room
and looks at thousands of these images a day. Obviously fatigue sets in and other
human, natural human issues set in. So, we are in the process of buying a company
Merge Healthcare, 30 billion medical images. We are going to train Watson
on those medical images. And not only are we going to do the analytics
on the image, we're going to bring the learnings from the electronic medical records because
Watson can read the medical records… Bring that information together, bring insights
from previous patient and previous outcomes all in one place with a physician to make
a decision on treatment in minutes. Very, very powerful capability. I think this is going to change
the course of healthcare. It's going to change outcomes and it's going
to take a lot of waste out of the system. Seismology, as I mentioned
earlier, we're working with a number of the world's largest oil and gas companies. They are building cognitive
environments for decision making. Decision making around where
do we drill the next well? Do I bid on this piece of land to try
to get the oil reserves underneath? What's going to happen to
those reserves over time? These are very complex decisions
that need to be made, in many cases, in very short periods of time. I think that we're going to help transform what
has also become a very intensive data industry. Education is another one that
I'm particularly excited about. In a sense, there's a direct analogy
between education and healthcare. Today in healthcare, I'm over
simplifying, but we are still diagnosing and treating to the average, by and large. You're still going to get
prescription or treatment that is to the doc's best understanding
what people sort of like you on average he or she has seen in the past. Education is very much the same. We put our children in the class and by and
large, the teachers teach to the average and they try to adjust a little bit
for children on the two extremes. Think about having a Watson engage in the
education system with the individual student where Watson can observe the learning
patterns and decide, you know what? This child is not learning properly
or this child is having no problem with these concepts cannot learn this
concept, and intervening at that moment. Think about a pre-K'er. We know that between the ages of
two and three, the number of words that child learns is a direct
correlation to their ultimate potential. Think about Watson intervening and doubling or tripling the vocabulary
of a two- or three-year old. An accelerating learning. Huge opportunity to change the education
system and the outcomes that we see. Think about genomics. Another area of healthcare. Genomic data will probably
eventually swamp image data. Everyone involved in genomics will tell
you more data than I know what to do with. I not only cannot deal with the hundreds of
mutations, I cannot deal with the thousands of pathways that may be causing
that tumor to manifest itself. Think about using a Watson to do…to
understand what do those mutations mean and what is the right drug or
cocktail to use for that cancer. Working with a number of leading edge genomic
institutions around the United States and Canada to explore what Watson can do in this area. Frankly, it's the only way we're
going to deal with genomic data. So, let me ends then by reflecting on some
words that I found from Thomas Watson, Jr. As you all probably know,
Thomas Watson was our founder. Tom Watson, Jr., ran the company during the time
of the 360 and a lot of our explosive growth. And I thought that this quote was quite
interesting, because he describes systems that are not going to, quote-unquote, rob
man of his initiative but are going to start to take away some of the…what he refers to as
menial tasks, mental menial tasks and free us up for creativity and other things. And in a sense, that's what
happened during the programmable era. We took what we wanted the machine to do. We put it into memory. We made the machine to do what we wanted and
we went off and did more creative processes. I don't know what Mr. Watson would think
about what we could do with Watson itself. But think about the fact that it's
no longer about displacing work. Machines displace manual labor. Programmable systems displaced the
menial, quote-unquote, mental processes. We're now talking about man
and machine tackling problems that were inconceivable just a few years
ago, whether it's education, healthcare, energy sources, on and on and on. Some of the world's biggest problems I believe
are going to be solved by these technologies. So, again, I'll sort of end where I started. I urge you throughout the course of the
day, enjoy, participate in the discussion around the technology, around what's
going on in artificial intelligence. But don't fall into the trap of, we're trying
to reproduce a human brain and don't fall into just the deep technology trap. Think about the applications of this technology. Whether you're going to create a business,
whether you want to apply this into your field or your discipline, there is not an industry or
discipline that won't be completely transformed by this technology over the next decade. [ MUSIC ]

11 Comments

  1. FourBloodMoons said:

    yes but where is the ibm 5100

    June 27, 2019
    Reply
  2. Audrey Mciver said:

    I don’t use Twitter anymore

    June 27, 2019
    Reply
  3. Audrey Mciver said:

    There is probably a lot of solutions in my data and I’m willing to let you analyze it, as I was effected by hackers

    June 27, 2019
    Reply
  4. Audrey Mciver said:

    I really want to find a cure for cancer. I’m not sure how to combine my data to make this happen

    June 27, 2019
    Reply
  5. Audrey Mciver said:

    Amen

    June 27, 2019
    Reply
  6. Namit Kewat said:

    Anyone relates it with samaritan from "person of interest"?

    June 27, 2019
    Reply
  7. Supriya Desai, Desai Transformation said:

    Fantastic way of thinking about the value of technology – especially in the case of cognitive computing technology like Watson: What is the price of not knowing? In the case of the United States healthcare system, it's possible that the price of not knowing may be in the neighborhood of $1 trillion. What could we do that money instead that could have a positive impact on society? What others entrenched problems have plagued humanity that we might be able to solve? What solutions work best to keep girls in school in developing nations? What factors most strongly correlate to corruption in government and how early can we detect them? That kind of world-changing, ethically-minded, large-scale opportunism is, in my mind and as a lay person, the real power of this brave new world of cognitive computing.

    June 27, 2019
    Reply
  8. Supriya Desai, Desai Transformation said:

    Fantastic way of thinking about the value of technology – especially in the case of cognitive computing technology like Watson: What is the price of not knowing? In the case of the United States healthcare system, it's possible that the price of not knowing may be in the neighborhood of $1 trillion. What could we do that money instead that could have a positive impact on society? What others entrenched problems have plagued humanity that we might be able to solve? What solutions work best to keep girls in school in developing nations? What factors most strongly correlate to corruption in government and how early can we detect them? That kind of world-changing, ethically-minded, large-scale opportunism is, in my mind and as a lay person, the real power of this brave new world of cognitive computing.

    June 27, 2019
    Reply
  9. Mayank Vora said:

    Just mind blowing… amazing vision Dr. John Kelly

    June 27, 2019
    Reply
  10. Raydnt said:

    Who else came here from Steins; Gate?

    June 27, 2019
    Reply
  11. Brad Haaf said:

    But is there enough computing power in the world to give everyone a cognitive assistant ? Or is it a 1 billion out of 8 by 2030 kinda thing ?

    June 27, 2019
    Reply

Leave a Reply

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