machines to perform tasks humans have been doing and in order to do those
been doing and in order to do those
been doing and in order to do those tasks we have utilized our knowledge
tasks we have utilized our knowledge
tasks we have utilized our knowledge about the world that we have accumulated
about the world that we have accumulated
about the world that we have accumulated over our entire lifespan right now when
over our entire lifespan right now when
over our entire lifespan right now when we were at the pool and that collection
we were at the pool and that collection
we were at the pool and that collection of knowledge about the world is going to
of knowledge about the world is going to
of knowledge about the world is going to continue to be done so on the screen you
continue to be done so on the screen you
continue to be done so on the screen you see a robot
see a robot
see a robot Alecto bot at CMU whose task is to
Alecto bot at CMU whose task is to
Alecto bot at CMU whose task is to basically go around and help people in
basically go around and help people in
basically go around and help people in the office environment so you could
the office environment so you could
the office environment so you could potentially ask over to go and collect
potentially ask over to go and collect
potentially ask over to go and collect the freedom that they are fired or let
the freedom that they are fired or let
the freedom that they are fired or let go in coffee
go in coffee
go in coffee so in order to do this on trivial tasks
so in order to do this on trivial tasks
so in order to do this on trivial tasks the machine the robot in this particular
the machine the robot in this particular
the machine the robot in this particular case each to have a shared view of the
case each to have a shared view of the
case each to have a shared view of the work with you and me but unfortunately
work with you and me but unfortunately
work with you and me but unfortunately these machines right now don’t share
these machines right now don’t share
these machines right now don’t share that same worldview which is really
that same worldview which is really
that same worldview which is really important to do these intelligent
important to do these intelligent
important to do these intelligent decision-making tasks very well so that
decision-making tasks very well so that
decision-making tasks very well so that brings us to the main thesis of my talk
brings us to the main thesis of my talk
brings us to the main thesis of my talk which is that background knowledge is
which is that background knowledge is
which is that background knowledge is really key to intelligent decision right
really key to intelligent decision right
really key to intelligent decision right and we as I mentioned very recently
and we as I mentioned very recently
and we as I mentioned very recently asking machines to do those same tasks
asking machines to do those same tasks
asking machines to do those same tasks as you and I do from which we benefit
as you and I do from which we benefit
as you and I do from which we benefit from having the word of it but machines
from having the word of it but machines
from having the word of it but machines don’t have
don’t have
don’t have so – just a drive of the point let’s say
so – just a drive of the point let’s say
so – just a drive of the point let’s say I don’t have time to go into need ten
I don’t have time to go into need ten
I don’t have time to go into need ten different websites every morning but
different websites every morning but
different websites every morning but took bet that today’s news I want to be
took bet that today’s news I want to be
took bet that today’s news I want to be the system which can summarize the news
the system which can summarize the news
the system which can summarize the news world and one subsystem goes and finds a
world and one subsystem goes and finds a
world and one subsystem goes and finds a sentence it says that stay from stocks
sentence it says that stay from stocks
sentence it says that stay from stocks double the Hamid merger happen for us as
double the Hamid merger happen for us as
double the Hamid merger happen for us as humans to interpret this particular
humans to interpret this particular
humans to interpret this particular sentence we need to know that there’s a
sentence we need to know that there’s a
sentence we need to know that there’s a concept of company and instances of
concept of company and instances of
concept of company and instances of companies are register which is state
companies are register which is state
companies are register which is state department’s Berkshire Hathaway there’s
department’s Berkshire Hathaway there’s
department’s Berkshire Hathaway there’s something called a stock market where
something called a stock market where
something called a stock market where company stocks are traded and there are
company stocks are traded and there are
company stocks are traded and there are prices associated with them but none of
prices associated with them but none of
prices associated with them but none of those additional knowledge which is
those additional knowledge which is
those additional knowledge which is necessary to understand this particular
necessary to understand this particular
necessary to understand this particular sentence is mentioned the sentence
sentence is mentioned the sentence
sentence is mentioned the sentence itself right so we have built our
itself right so we have built our
itself right so we have built our knowledge above the world about this
knowledge above the world about this
knowledge above the world about this company’s concepts as stock prices and
company’s concepts as stock prices and
company’s concepts as stock prices and all and we’re using all of that
all and we’re using all of that
all and we’re using all of that knowledge to make sense of this
knowledge to make sense of this
knowledge to make sense of this particular sentence right but machines
particular sentence right but machines
particular sentence right but machines as I mentioned earlier don’t have that
as I mentioned earlier don’t have that
as I mentioned earlier don’t have that same understanding and share of the
same understanding and share of the
same understanding and share of the world so even if you do all of that bag
world so even if you do all of that bag
world so even if you do all of that bag of knowledge why there might be
of knowledge why there might be
of knowledge why there might be correlation between the stock prices of
correlation between the stock prices of
correlation between the stock prices of these two companies that’s not
these two companies that’s not
these two companies that’s not immediately clear but if I should do
immediately clear but if I should do
immediately clear but if I should do some structured knowledge which could
some structured knowledge which could
some structured knowledge which could support this do this particular sentence
support this do this particular sentence
support this do this particular sentence which says that Berkshire Hathaway which
which says that Berkshire Hathaway which
which says that Berkshire Hathaway which is mentioned in this particular sentence
is mentioned in this particular sentence
is mentioned in this particular sentence has a subsidiary of a Heinz Heinz
has a subsidiary of a Heinz Heinz
has a subsidiary of a Heinz Heinz is been insured by State Bar which is
is been insured by State Bar which is
is been insured by State Bar which is also mentioned in the sentence right so
also mentioned in the sentence right so
also mentioned in the sentence right so one possible explanation of why there
one possible explanation of why there
one possible explanation of why there might be a correlation between the stock
might be a correlation between the stock
might be a correlation between the stock prices of these two companies is that
prices of these two companies is that
prices of these two companies is that the market may think that the losses of
the market may think that the losses of
the market may think that the losses of Berkshire Hathaway
Berkshire Hathaway
Berkshire Hathaway is because of its subsidiary Heinz and
is because of its subsidiary Heinz and
is because of its subsidiary Heinz and the losses of the Heinz Company are
the losses of the Heinz Company are
the losses of the Heinz Company are going to ultimately get consumed by the
going to ultimately get consumed by the
going to ultimately get consumed by the insurance company State Farm
insurance company State Farm
insurance company State Farm right so now because we have this kind
right so now because we have this kind
right so now because we have this kind of structured knowledge our
of structured knowledge our
of structured knowledge our interpretation of the sentence
interpretation of the sentence
interpretation of the sentence even as humans are what deeper now
even as humans are what deeper now
even as humans are what deeper now compared to say 30 seconds before when
compared to say 30 seconds before when
compared to say 30 seconds before when we didn’t have this better and the
we didn’t have this better and the
we didn’t have this better and the metric knowledge that we see which is
metric knowledge that we see which is
metric knowledge that we see which is organized as a multi relational graph
organized as a multi relational graph
organized as a multi relational graph here is fragment of a much larger graph
here is fragment of a much larger graph
here is fragment of a much larger graph that we call is the knowledge so that’s
that we call is the knowledge so that’s
that we call is the knowledge so that’s the main branch of the idea here is that
the main branch of the idea here is that
the main branch of the idea here is that if you are able to organize our
if you are able to organize our
if you are able to organize our knowledge about the world in the form of
knowledge about the world in the form of
knowledge about the world in the form of substructure depository in this
substructure depository in this
substructure depository in this particular case a knowledge graph then
particular case a knowledge graph then
particular case a knowledge graph then we are able to be available to
we are able to be available to
we are able to be available to intelligent decision making AI and
intelligent decision making AI and
intelligent decision making AI and machine living idioms then their
machine living idioms then their
machine living idioms then their performance is going to improve
performance is going to improve
performance is going to improve significantly so now if you agree with
significantly so now if you agree with
significantly so now if you agree with me on that particular point then the
me on that particular point then the
me on that particular point then the question is how are we going to build
question is how are we going to build
question is how are we going to build this large knowledge posit trees are
this large knowledge posit trees are
this large knowledge posit trees are large vegetables so one solution would
large vegetables so one solution would
large vegetables so one solution would be that we actually ask humans and type
be that we actually ask humans and type
be that we actually ask humans and type them by being obvious knowledge but
them by being obvious knowledge but
them by being obvious knowledge but that’s not really going to be scalable
that’s not really going to be scalable
that’s not really going to be scalable because there’s a lot of knowledge that
because there’s a lot of knowledge that
because there’s a lot of knowledge that has already been produced and then this
has already been produced and then this
has already been produced and then this alone if we say
alone if we say
alone if we say a constantly evolving process as well so
a constantly evolving process as well so
a constantly evolving process as well so wrap them crowdsourcing are typing on
wrap them crowdsourcing are typing on
wrap them crowdsourcing are typing on this knowledge ourselves our goal is to
this knowledge ourselves our goal is to
this knowledge ourselves our goal is to see how we can leverage already existing
see how we can leverage already existing
see how we can leverage already existing so we as humans we have been documenting
so we as humans we have been documenting
so we as humans we have been documenting our knowledge through emails tweets
our knowledge through emails tweets
our knowledge through emails tweets chats about and with a group of the
chats about and with a group of the
chats about and with a group of the internet and the world wide web there’s
internet and the world wide web there’s
internet and the world wide web there’s an explosion of that type of data that’s
an explosion of that type of data that’s
an explosion of that type of data that’s readily available out there another so
readily available out there another so
readily available out there another so the question is how we can leverage all
the question is how we can leverage all
the question is how we can leverage all of that data and extract from this
of that data and extract from this
of that data and extract from this fuelie unstructured data which is
fuelie unstructured data which is
fuelie unstructured data which is written in natural language how we can
written in natural language how we can
written in natural language how we can extract the type of knowledge graphs
extract the type of knowledge graphs
extract the type of knowledge graphs then we can see in the previously so
then we can see in the previously so
then we can see in the previously so that’s that’s the main mode that you’re
that’s that’s the main mode that you’re
that’s that’s the main mode that you’re interested in doing so whenever I say we
interested in doing so whenever I say we
interested in doing so whenever I say we need to extract and harvest this
need to extract and harvest this
need to extract and harvest this knowledge from webseal unstructured
knowledge from webseal unstructured
knowledge from webseal unstructured natural language data
natural language data
natural language data what will it be so let’s say out of
what will it be so let’s say out of
what will it be so let’s say out of those millions and billions of documents
those millions and billions of documents
those millions and billions of documents let’s consider two of them and one of
let’s consider two of them and one of
let’s consider two of them and one of them says Luke Ravenstahl is the current
them says Luke Ravenstahl is the current
them says Luke Ravenstahl is the current mayor of Pittsburgh and the other one
mayor of Pittsburgh and the other one
mayor of Pittsburgh and the other one says after the death of so any one of us
says after the death of so any one of us
says after the death of so any one of us reading is too stupid with we can infer
reading is too stupid with we can infer
reading is too stupid with we can infer that we are talking about two people
that we are talking about two people
that we are talking about two people here which is Luke remains solid ball
here which is Luke remains solid ball
here which is Luke remains solid ball under there’s a location make Pittsburgh
under there’s a location make Pittsburgh
under there’s a location make Pittsburgh and then the mayor of relationship
and then the mayor of relationship
and then the mayor of relationship between these two entities and this
between these two entities and this
between these two entities and this particular Commission right
particular Commission right
particular Commission right and there’s also employing aspect that
and there’s also employing aspect that
and there’s also employing aspect that because both of them were not Mary the
because both of them were not Mary the
because both of them were not Mary the city so whenever I’m talking
city so whenever I’m talking
city so whenever I’m talking harvesting we are talking about how to
harvesting we are talking about how to
harvesting we are talking about how to build machines which are good enough
build machines which are good enough
build machines which are good enough readers which can read this kind of
readers which can read this kind of
readers which can read this kind of textual data at scale and is able to
textual data at scale and is able to
textual data at scale and is able to recognize what kind of entities are
recognize what kind of entities are
recognize what kind of entities are mentioned in the data what kind of
mentioned in the data what kind of
mentioned in the data what kind of relationships exist among those entities
relationships exist among those entities
relationships exist among those entities and how those relationships all evolved
and how those relationships all evolved
and how those relationships all evolved over time and the framework of that I
over time and the framework of that I
over time and the framework of that I have shown this progressive so by the
have shown this progressive so by the
have shown this progressive so by the way this is also not that abstract
way this is also not that abstract
way this is also not that abstract Academy pursue this idea of building
Academy pursue this idea of building
Academy pursue this idea of building knowledge graphs that became available
knowledge graphs that became available
knowledge graphs that became available for you propositioning in fact right
for you propositioning in fact right
for you propositioning in fact right from around 2012 or so whenever you
from around 2012 or so whenever you
from around 2012 or so whenever you search in google you may have see this
search in google you may have see this
search in google you may have see this panel panel on the right side so in this
panel panel on the right side so in this
panel panel on the right side so in this case I search for exception and you’ve
case I search for exception and you’ve
case I search for exception and you’ve seen how we see this knowledge battle as
seen how we see this knowledge battle as
seen how we see this knowledge battle as it’s about maybe we quickly summarizes a
it’s about maybe we quickly summarizes a
it’s about maybe we quickly summarizes a lot of things about that nature so in
lot of things about that nature so in
lot of things about that nature so in this particular piece I have no thought
this particular piece I have no thought
this particular piece I have no thought about who the founders of Accenture are
about who the founders of Accenture are
about who the founders of Accenture are like what their head battery is what is
like what their head battery is what is
like what their head battery is what is a stock simple things like that so the
a stock simple things like that so the
a stock simple things like that so the benefit is that now I don’t have to go
benefit is that now I don’t have to go
benefit is that now I don’t have to go and read ten different web pages to get
and read ten different web pages to get
and read ten different web pages to get the same summary information about
the same summary information about
the same summary information about Accenture and it includes by web search
Accenture and it includes by web search
Accenture and it includes by web search experience and probably all of us are
experience and probably all of us are
experience and probably all of us are beneficiaries of this kind of memory and
beneficiaries of this kind of memory and
beneficiaries of this kind of memory and is that these knowledge panels are
is that these knowledge panels are
is that these knowledge panels are internally being powered by a knowledge
internally being powered by a knowledge
internally being powered by a knowledge graph of the form that we have seen
graph of the form that we have seen
graph of the form that we have seen before so you and I are already
before so you and I are already
before so you and I are already beneficiaries of this technology but
beneficiaries of this technology but
beneficiaries of this technology but it’s the solve problem so Claire is the
it’s the solve problem so Claire is the
it’s the solve problem so Claire is the lady happens to be one of the cofounders
lady happens to be one of the cofounders
lady happens to be one of the cofounders of Accenture so when I met
of Accenture so when I met
of Accenture so when I met that is really I didn’t find any
that is really I didn’t find any
that is really I didn’t find any chocolate metal as it is an exception
chocolate metal as it is an exception
chocolate metal as it is an exception right so that’s a good start there is a
right so that’s a good start there is a
right so that’s a good start there is a long way to go in terms of making it
long way to go in terms of making it
long way to go in terms of making it comprehensive knowledge so done net
comprehensive knowledge so done net
comprehensive knowledge so done net project at CMU when I was also part of
project at CMU when I was also part of
project at CMU when I was also part of that we continue to do that pilot
that we continue to do that pilot
that we continue to do that pilot research at I RAC is an attempt to read
research at I RAC is an attempt to read
research at I RAC is an attempt to read basically documents at web speed and
basically documents at web speed and
basically documents at web speed and extract and organize the knowledge
extract and organize the knowledge
extract and organize the knowledge that’s trapped inside the unstructured
that’s trapped inside the unstructured
that’s trapped inside the unstructured people so in particular the nervous
people so in particular the nervous
people so in particular the nervous system has been running continuously for
system has been running continuously for
system has been running continuously for more than seven years now January 12
more than seven years now January 12
more than seven years now January 12 happens to be next birthday you’re
happens to be next birthday you’re
happens to be next birthday you’re welcome to come and celebrate the
welcome to come and celebrate the
welcome to come and celebrate the birthday in our lab or in Pittsburgh
[Music]
[Music] extracting knowledge about the world of
extracting knowledge about the world of
extracting knowledge about the world of art and structured data it’s also able
art and structured data it’s also able
art and structured data it’s also able to do reasoning over the next rack
to do reasoning over the next rack
to do reasoning over the next rack technology for example yeah if we if Mel
technology for example yeah if we if Mel
technology for example yeah if we if Mel knows that sheikah is a place for Team
knows that sheikah is a place for Team
knows that sheikah is a place for Team India Team India place for cricket
India Team India place for cricket
India Team India place for cricket then then we’ll be able to infer that
then then we’ll be able to infer that
then then we’ll be able to infer that she had actually plays for free even
she had actually plays for free even
she had actually plays for free even though that particular fact may not have
though that particular fact may not have
though that particular fact may not have been present upfront and additionally
been present upfront and additionally
been present upfront and additionally it’s also able to discover new predicate
it’s also able to discover new predicate
it’s also able to discover new predicate or new types of knowledge that it thinks
or new types of knowledge that it thinks
or new types of knowledge that it thinks is Ness
is Ness
is Ness to learn about the particular to be so
to learn about the particular to be so
to learn about the particular to be so just to give an example at one point
just to give an example at one point
just to give an example at one point they knew a lot about rivers and then it
they knew a lot about rivers and then it
they knew a lot about rivers and then it also do a lot about cities so then it
also do a lot about cities so then it
also do a lot about cities so then it saw that by analyzing data that people
saw that by analyzing data that people
saw that by analyzing data that people mentioned reverses it is it close
mentioned reverses it is it close
mentioned reverses it is it close proximity it’s much more probability
proximity it’s much more probability
proximity it’s much more probability than random chance so if even with this
than random chance so if even with this
than random chance so if even with this new predicate or new type of knowledge
new predicate or new type of knowledge
new predicate or new type of knowledge small River flows through City so that
small River flows through City so that
small River flows through City so that becomes a new learning task for now and
becomes a new learning task for now and
becomes a new learning task for now and then that job will be to enter five
then that job will be to enter five
then that job will be to enter five additional instances of that new type of
additional instances of that new type of
additional instances of that new type of knowledge and basically densified
knowledge and basically densified
knowledge and basically densified populate part of the knowledge graph
populate part of the knowledge graph
populate part of the knowledge graph that it has been so this is a horse of
that it has been so this is a horse of
that it has been so this is a horse of work of a much larger group that I
work of a much larger group that I
work of a much larger group that I presented there so what you see here on
presented there so what you see here on
presented there so what you see here on screen is basically a fragment of the
screen is basically a fragment of the
screen is basically a fragment of the knowledge graph that Nell has
knowledge graph that Nell has
knowledge graph that Nell has accumulated by reading over this more
accumulated by reading over this more
accumulated by reading over this more than seven years now where the nodes are
than seven years now where the nodes are
than seven years now where the nodes are activities of interest like the Toronto
activities of interest like the Toronto
activities of interest like the Toronto Maple Leafs and type edges other
Maple Leafs and type edges other
Maple Leafs and type edges other relationships that basically connect to
relationships that basically connect to
relationships that basically connect to entities which says Toronto people names
entities which says Toronto people names
entities which says Toronto people names for Thomas sector Agra so you can go and
for Thomas sector Agra so you can go and
for Thomas sector Agra so you can go and browse this whole knowledge graph and a
browse this whole knowledge graph and a
browse this whole knowledge graph and a URL rtw
URL rtw
URL rtw a Nazi a VW and those of you who are on
a Nazi a VW and those of you who are on
a Nazi a VW and those of you who are on Twitter you can also actually going
Twitter you can also actually going
Twitter you can also actually going hello Mel a biscuit Elsie mmm
hello Mel a biscuit Elsie mmm
hello Mel a biscuit Elsie mmm so that’s really and what we have
so that’s really and what we have
so that’s really and what we have observed is there actually a few
observed is there actually a few
observed is there actually a few thousand human followers who follow them
thousand human followers who follow them
thousand human followers who follow them on Twitter and they’ll also tweet from
on Twitter and they’ll also tweet from
on Twitter and they’ll also tweet from time to time
time to time
time to time facts are beliefs that it thinks is
facts are beliefs that it thinks is
facts are beliefs that it thinks is interesting and not always completely
interesting and not always completely
interesting and not always completely correct but it cries interest and what
correct but it cries interest and what
correct but it cries interest and what we also have is that we have feedback
we also have is that we have feedback
we also have is that we have feedback coming from these humans so here you
coming from these humans so here you
coming from these humans so here you have a situation where you have an AI
have a situation where you have an AI
have a situation where you have an AI agent which is reading these facts on
agent which is reading these facts on
agent which is reading these facts on its own tweeting about them and humans
its own tweeting about them and humans
its own tweeting about them and humans are living free so it’s a real
are living free so it’s a real
are living free so it’s a real interesting open question is how we can
interesting open question is how we can
interesting open question is how we can leverage this table interested
leverage this table interested
leverage this table interested interchange of ideas between the natives
interchange of ideas between the natives
interchange of ideas between the natives and few thousand humans to build better
and few thousand humans to build better
and few thousand humans to build better images all together
images all together
images all together so how has Nell perform over time so you
so how has Nell perform over time so you
so how has Nell perform over time so you saw the fragment of the knowledge graph
saw the fragment of the knowledge graph
saw the fragment of the knowledge graph that has automatically extracted but
that has automatically extracted but
that has automatically extracted but some quantitative numbers so in this
some quantitative numbers so in this
some quantitative numbers so in this particular plot on x-axis you basically
particular plot on x-axis you basically
particular plot on x-axis you basically have a number of iterations so that’s a
have a number of iterations so that’s a
have a number of iterations so that’s a proxy for time and the y-axis in this
proxy for time and the y-axis in this
proxy for time and the y-axis in this first one you see how many believes
first one you see how many believes
first one you see how many believes there has accumulated so the general
there has accumulated so the general
there has accumulated so the general trend you see that over time there has
trend you see that over time there has
trend you see that over time there has been able to extract more and more beads
been able to extract more and more beads
been able to extract more and more beads by reading these documents from the web
by reading these documents from the web
by reading these documents from the web on a continuous basis so right now
on a continuous basis so right now
on a continuous basis so right now Mendes cross thousand iterations and
Mendes cross thousand iterations and
Mendes cross thousand iterations and there’s an accumulated more than hundred
there’s an accumulated more than hundred
there’s an accumulated more than hundred million believes of
million believes of
million believes of and by medieval I mean basically facts
and by medieval I mean basically facts
and by medieval I mean basically facts of the form that MIT wanna build is
of the form that MIT wanna build is
of the form that MIT wanna build is located in one specific one edge from
located in one specific one edge from
located in one specific one edge from the ground so society is one aspect but
the ground so society is one aspect but
the ground so society is one aspect but also in terms of quality when we found
also in terms of quality when we found
also in terms of quality when we found when we did the evaluation we found that
when we did the evaluation we found that
when we did the evaluation we found that over time that has been able to
over time that has been able to
over time that has been able to basically hold down on the quality and
basically hold down on the quality and
basically hold down on the quality and opti baby in terms of performance so on
opti baby in terms of performance so on
opti baby in terms of performance so on average then tends to be about eighty to
average then tends to be about eighty to
average then tends to be about eighty to eighty-five percent back so here you
eighty-five percent back so here you
eighty-five percent back so here you have an area in Egypt which has been
have an area in Egypt which has been
have an area in Egypt which has been running continuously pretty much self
running continuously pretty much self
running continuously pretty much self supervised without external inputs for
supervised without external inputs for
supervised without external inputs for more than seven years now this being
more than seven years now this being
more than seven years now this being able to read documents and extract
able to read documents and extract
able to read documents and extract believes which is across more than one
believes which is across more than one
believes which is across more than one hundred million mark and has not
hundred million mark and has not
hundred million mark and has not degraded performance so as far as I know
degraded performance so as far as I know
degraded performance so as far as I know is probably one of the first instances
is probably one of the first instances
is probably one of the first instances having this kind of sensor provides
having this kind of sensor provides
having this kind of sensor provides agents which is able to run for extended
agents which is able to run for extended
agents which is able to run for extended time crisis and is able to learn
time crisis and is able to learn
time crisis and is able to learn effectively as as much as that has done
effectively as as much as that has done
effectively as as much as that has done so which is really interesting so now
so which is really interesting so now
so which is really interesting so now our research kind of goes beyond this
our research kind of goes beyond this
our research kind of goes beyond this and also in some elections to
and also in some elections to
and also in some elections to neuroscience that I want to quickly
neuroscience that I want to quickly
neuroscience that I want to quickly Alistair so if you think about this text
Alistair so if you think about this text
Alistair so if you think about this text at the top means that who’s met with
at the top means that who’s met with
at the top means that who’s met with technology we are trying to organize so
technology we are trying to organize so
technology we are trying to organize so these are usually not machine generated
these are usually not machine generated
these are usually not machine generated right so these are written by humans so
right so these are written by humans so
right so these are written by humans so there is some processing
there is some processing
there is some processing happening inside
happening inside
happening inside and that process is generating all the
and that process is generating all the
and that process is generating all the data on our four data format for us and
data on our four data format for us and
data on our four data format for us and we are trying to model the knowledge of
we are trying to model the knowledge of
we are trying to model the knowledge of the data but with advances in fMRI and
the data but with advances in fMRI and
the data but with advances in fMRI and brain imaging techniques we also have a
brain imaging techniques we also have a
brain imaging techniques we also have a defect peek into that same method
defect peek into that same method
defect peek into that same method phenomenon of what’s going on in South
phenomenon of what’s going on in South
phenomenon of what’s going on in South Africa so rather looking at so we have
Africa so rather looking at so we have
Africa so rather looking at so we have one common source and we have two data
one common source and we have two data
one common source and we have two data sources basically two different views
sources basically two different views
sources basically two different views onto the same common flower so rather
onto the same common flower so rather
onto the same common flower so rather looking at each one of them in isolation
looking at each one of them in isolation
looking at each one of them in isolation our rule list of how we can basically
our rule list of how we can basically
our rule list of how we can basically connect and exploit both modalities of
connect and exploit both modalities of
connect and exploit both modalities of text and renovated to get a better
text and renovated to get a better
text and renovated to get a better handle into how and what is happening
handle into how and what is happening
handle into how and what is happening inside our heads so we are we have also
inside our heads so we are we have also
inside our heads so we are we have also done research in that direction and we
done research in that direction and we
done research in that direction and we have found that when you combine these
have found that when you combine these
have found that when you combine these two modalities you are able to answer
two modalities you are able to answer
two modalities you are able to answer question there will be really hard
question there will be really hard
question there will be really hard answer if we’re just looking at one
answer if we’re just looking at one
answer if we’re just looking at one modality or the other so there’s of
modality or the other so there’s of
modality or the other so there’s of course work that needs to happen but
course work that needs to happen but
course work that needs to happen but that said early promising direction so
that said early promising direction so
that said early promising direction so with the convergence of data computation
with the convergence of data computation
with the convergence of data computation and advances in statistical machine
and advances in statistical machine
and advances in statistical machine learning techniques we have I think for
learning techniques we have I think for
learning techniques we have I think for the first time in human history a real
the first time in human history a real
the first time in human history a real opportunity to go from big debts to big
opportunity to go from big debts to big
opportunity to go from big debts to big knowledge a text is just one of the
knowledge a text is just one of the
knowledge a text is just one of the modalities that
modalities that
modalities that the star but their provincial government
the star but their provincial government
the star but their provincial government that leaders like speech if vision but
that leaders like speech if vision but
that leaders like speech if vision but the point here is that we have an
the point here is that we have an
the point here is that we have an opportunity to business basically equity
opportunity to business basically equity
opportunity to business basically equity in all the knowledge of human knowledge
in all the knowledge of human knowledge
in all the knowledge of human knowledge into one structured form and make it
into one structured form and make it
into one structured form and make it available to AI machine learning agents
available to AI machine learning agents
available to AI machine learning agents and overcome the politic of email which
and overcome the politic of email which
and overcome the politic of email which has played AI agents and nowhere in
has played AI agents and nowhere in
has played AI agents and nowhere in history BC this was possible so this is
history BC this was possible so this is
history BC this was possible so this is really exciting times which also keeps
really exciting times which also keeps
really exciting times which also keeps us motivated to work with the problem
us motivated to work with the problem
us motivated to work with the problem and true to our research which spans the
and true to our research which spans the
and true to our research which spans the areas of machine learning natural
areas of machine learning natural
areas of machine learning natural language processing a large content
language processing a large content
language processing a large content analysis we want to organize this kind
analysis we want to organize this kind
analysis we want to organize this kind of structured knowledge graphs make it
of structured knowledge graphs make it
of structured knowledge graphs make it available to AI and machine learning
available to AI and machine learning
available to AI and machine learning agents and hopefully that results in
agents and hopefully that results in
agents and hopefully that results in better position maybe so this is a
better position maybe so this is a
better position maybe so this is a really deep area which is in its infancy
really deep area which is in its infancy
really deep area which is in its infancy and my six-year goal is that at least
and my six-year goal is that at least
and my six-year goal is that at least some of you will pick up the challenges
some of you will pick up the challenges
some of you will pick up the challenges in this area and contribute tremendously
in this area and contribute tremendously
in this area and contribute tremendously towards building better they are agents
towards building better they are agents
towards building better they are agents with the thank you for attention
with the thank you for attention
with the thank you for attention [Applause]
”
Be First to Comment