Camels, Code & Lab Coats: How AI Is Advancing Science and Medicine

man: There’s been many
innovations over the years that drive science TubeRushr OTO. If you go back in history, TubeRushr OTO many scientific insights
actually derived from new tools Tube Rushr OTO that were able to
measure Tube Rushr OTO new things. AI is very, very good at
finding new paths that haven’t been seen before. It’s almost an enhancement on
our ability to sense the world. man: The main benefit of
machine learning is the ability to learn from
lots and lots of data. man: We can show a computer a
lot of examples, and rather than tell a model
how to evaluate that data can learn actually how to
interpret it and figure out what to do next. woman: Here at Google, we’ve
been using machine-learning technologies
in all of our products, things like Search, Translate,
Google Photos, and the Assistant. man: People are realizing that
learning from examples is a very powerful tool.

Man: So researchers in the
health care community were looking at some of the
work that Google was doing in artificial intelligence and
actually reached out to us and said, “Is there a way that
you could apply these same kind of technologies
in health care?” Doctors today have more
artificial intelligence working for them on their
smartphone for their personal use than
they have working for them in the clinical context.

Woman: And as doctors, we’ve
just gotten more information kind of shoved at us in terms
of records, in terms of images. woman: Think about all the
different type of cells that you have. man: 15,000, 20,000 different
diagnosis codes. man: Just a cubic millimeter of
tissue, it’s like taking,
you know, a billion photos. man: And if you print it out,
it would probably be about as tall as a ten-story
building. man: Really any kind of data
that people can generate on a massive scale is really an
area where machine learning can

[upbeat music] man: Now there’s an opportunity
to use all our digital technologies that have been
developed at Google to really try to help doctors. man: Let’s take some of these
tools that are great for analyzing videos and
YouTube and apply them to problems that matter to
science. man: One of the things we’ve
been working on in medical imaging is in the
area of pathology. Traditionally, pathologists
take some tissue sample, and they look around in a sea
of cells looking for the kind of needle
in a haystack cancerous tissue. man: We know that the earlier
you detect the cancer the greater your chance of
curing it and the greater your chance of
curing it without chemotherapy or

So the biggest challenges are
speed and accuracy of diagnosis. So far we’ve trained models for
breast cancer and prostate cancer. woman: These technologies can
actually identify suspicious areas to direct the
doctor’s attention. man: So one of the things we
wanted to do was get this work into the hands of as
many people as possible. man: And so we developed
something called the augmented reality
microscope, where you can actually see
machine learning assistance overlaid in real time as
you’re looking through the microscope. man: These units can be
attached onto any existing microscope greatly
reducing the cost We’re really excited to bring
machine learning to parts of the world with
limited access. [light vibraphone music] man: One of the tools some
biologists use is, they will dye cells with
different colors to highlight certain important
features that make sense to them. man: The problem is, you have
to kill the cells in order to color them. man: So we said, “Well, if we
can do this virtually in a computer, we can preserve
the cell in its natural state.” man: One of the things that you
can do on your pixel phone right now is given a selfie,
you can predict the depth, and you can do interesting
visual effects.

So we thought, “Hey, can we
take this same technology and apply it in a biological
context” and essentially use machine
learning to predict the stain. We weren’t sure whether this
would work or not, and it ended up working so
well. We’re very hopeful that this
technology can be used to just have the computer
generate the pretty pictures that people know how to
interpret. [orchestral music] man: The brain is probably the
most complex physical object in the known
universe. We know that there are these
basic units called neurons, and they’re connected in many
different ways. What’s shocking is how little
is known about what those patterns of
connectivity look like and what that means for how the
brain works. The problem is very hard, because the connections are too
large to analyze. For example, a fly brain has
100,000 neurons, whereas a human brain has 100
billion neurons. So fortunately, at Google,
you know, there’s been already a lot of
work put into dealing with data sets of that size.

Machine learning and the
computer vision technology that we’ve developed has been
designed to accurately trace the wiring
of the brain in 3-D. Prior to that technology, it would have taken thousands
of years to basically finish mapping the
fly brain. Now you can do it within a year
or two. It’s a warm-up to understanding
larger and more complex brains, hopefully human brains. We’re hoping that mapping a
brain could potentially help us
understand a lot of the neurodegenerative
disorders– for example, Schizophrenia or
Parkinson’s– then we’re gonna be able to
design better therapies that might improve those
conditions. [delicate piano music] man: There are tremendous
inequities in the way that health care is
distributed across the globe. Any disease or outcome is
predicted as much by your zip code as it
is by your biology. man: So what can we do with
AI to bring the expertise to where no expertise exists? woman: One of the complications
of diabetes is diabetic retinopathy.

It causes blindness, and it’s diagnosed by seeing
little lesions in the eye. But in India, there is a
shortage of eye doctors. And as a result, about half the
patients suffer some form of vision loss. This disease is completely
preventable. This shouldn’t really be
happening. So we were able to train a
model to read these images and match board-certified
ophthalmologists. We’re now figuring out how to
deploy this into the clinic. In India, people who did not
have access now have access. [light piano music] man: A lot of science are is driven by a hypothesis that someone has. But some of the biggest
breakthroughs in science come from surprises, things no
one expected to have happened. man: One of the really exciting
things about deep learning is where you can just give it
the raw data, and it finds the important

That’s interacting with
scientists at a very different level than
saying, “Well, let me show you
something you’ve never seen before.” man: We had one research
project, we were looking at human
retinas. But what surprised us was
machine learning started seeing things that people
didn’t know was possible. woman: Turns out that a
deep-learning model is actually able to identify
things that have nothing to do with
your eyes, like your cardiovascular health
and your metabolic profile. These are things that if you
had asked experts, “What do you think we’ll find?” They would have said nothing. man: The important thing about
this is, it’s a visual biomarker that we
did not know existed before.

Man: Those unexpected things
may lead to a whole new idea, a whole new approach, a whole
new hypothesis for how to attack the problem
you’re trying to address. man: There are real patients
suffering today, and if we’re not doing
everything we can with all the technology at our
disposal to help those patients, then
what are we doing? man: I think that 10 or 15
years from now, the use of machine learning in
health care is just going to be how health
care is one woman: I think this is how
we’re gonna find new discoveries.

I think this is how we’re gonna
find ways to care of more people. man: The more we can accelerate
that basic work and give all these scientists
new tools, we will as humans really
benefit from the new discoveries that
one day end up in our doctors’ offices..

OTOs Upsells


Leave a Reply

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