Blaise Agüera y Arcas: How computers are learning to be creative
Blaise Agüera y Arcas: Cómo las computadoras aprenden a ser creativas
Blaise Agüera y Arcas works on machine learning at Google. Previously a Distinguished Engineer at Microsoft, he has worked on augmented reality, mapping, wearable computing and natural user interfaces. Full bio
Double-click the English transcript below to play the video.
trabaja en inteligencia artificial;
that works on machine intelligence;
of making computers and devices
para fabricar computadoras y dispositivos
que hacen los cerebros.
that brains do.
interested in real brains
mucho el cerebro real,
las cosas que hace nuestro cerebro
in the things that our brains do
to the performance of computers.
al rendimiento de las computadoras.
has been perception,
ha sido la percepción,
las cosas que hay en el mundo,
out there in the world --
en conceptos en la mente.
para nuestro propio cerebro,
for example, that our team makes,
de percepción computacional
on Google Photos to become searchable,
es la creatividad:
en algo que hay en el mundo.
out there into the world.
our work on machine perception
trabajo en percepción computacional
with the world of machine creativity
con el mundo de la creatividad
had a penetrating insight
tenía una visión aguda
between perception and creativity.
entre la percepción y la creatividad.
has a statue inside of it,
una estatua en su interior,
is to discover it."
es descubrirla".
Michelangelo was getting at
is an act of imagination
es un acto de imaginación
and perceiving and imagining,
la percepción y la imaginación,
with a brief bit of history
con un poquito de historia
el corazón o los intestinos,
the heart or the intestines,
about a brain by just looking at it,
de un cerebro con solo mirarlo,
que exploraron cerebros dieron
cosa todo tipo de nombres de fantasía,
of this thing all kinds of fanciful names,
que significa "pequeño camarón".
doesn't tell us very much
eso no nos dice mucho
sucede en el interior.
developed some kind of insight
desarrolló una visión
Santiago Ramón y Cajal,
Santiago Ramón y Cajal,
y tinciones especiales
or render in very high contrast
en un contraste muy alto
their morphologies.
that he made of neurons
que hizo de las neuronas
of different sorts of cells,
de diferentes tipos de células,
was quite new at this point.
era bastante nueva en este momento.
estas arborizaciones,
very, very long distances --
muy largas distancias,
a los cables.
para algunos en el siglo XIX;
to some people in the 19th century;
were just getting underway.
la electricidad se estaban iniciando.
of Ramón y Cajal's, like this one,
de Ramón y Cajal, como éste,
en cierto modo, insuperables.
that Ramón y Cajal started.
el trabajo que empezó Ramón y Cajal.
de nuestros colaboradores
de Neurociencia.
of Neuroscience.
trozos de tejido cerebral.
is about one cubic millimeter in size,
es de 1 mm cúbico de tamaño,
un trozo muy pequeño.
very small piece of it here.
de aproximadamente una micra.
son las mitocondrias
tiny block of tissue.
of hair is about 100 microns.
de pelo es de unas 100 micras.
mucho más pequeño
much, much smaller
electron microscopy slices,
de microscopía electrónica de serie,
in 3D of neurons that look like these.
de las neuronas con este aspecto.
que las de Ramón y Cajal.
style as Ramón y Cajal.
be able to see anything here.
no se podría ver nada aquí.
one neuron to another.
conectadas una a otra.
un poco a su tiempo,
ahead of his time,
en la comprensión del cerebro
a lo largo de las siguientes décadas.
over the next few decades.
usan electricidad
la tecnología avanzó lo suficiente
was advanced enough
experiments on live neurons
eléctricos reales con neuronas vivas
when computers were being invented,
se desarrollaban las computadoras
of modeling the brain --
as Alan Turing called it,
como decía Alan Turing,
looked at Ramón y Cajal's drawing
miraron el dibujo de Ramón y Cajal
imagery that comes from the eye.
las imágenes que provienen del ojo.
un diagrama de circuito.
like a circuit diagram.
diagrama de circuito de McCulloch y Pitts
in McCulloch and Pitts's circuit diagram
of computational elements
una serie de elementos computacionales
al siguiente en cascada,
one to the next in a cascade,
visual information would need to do.
para procesar la información visual.
de ver con nuestro cerebro.
for us to do with our brains.
que para una computadora
that for a computer,
just a few years ago.
hace pocos años.
de la computación clásica
this task is easy to do.
entre los píxeles,
and the word "bird,"
y la palabra "pájaro"
de neuronas conectadas entre sí
connected to each other
inside our visual cortices,
en nuestras cortezas visuales,
to have the capability
en la computadora.
on the computer.
that actually looks like.
about as a first layer of neurons,
como una primera capa de neuronas,
how it works in the eye --
como funciona el ojo,
after another layer of neurons,
y luego a otra capa de neuronas,
of different weights.
por sinapsis de diferentes pesos.
of all of those synapses.
de todas esas sinapsis.
properties of this network.
computacionales de esta red.
or a small group of neurons
those three things --
in the neural network,
las sinapsis en la red neuronal,
estas sinapsis en la red neuronal.
these synapses in the neural network.
is just a simple formula,
que esto es solo una fórmula simple,
going on there, of course,
por supuesto,
of mathematical operations.
de operaciones matemáticas.
si uno tiene una ecuación,
that if you have one equation,
by knowing the other two things.
conociendo las otras dos.
that the picture of a bird is a bird,
de un pájaro es un pájaro,
and w and x are known.
y W y X las conocidas.
you know the pixels.
y también los píxeles.
a relatively straightforward problem.
un problema relativamente sencillo.
and you're done.
una red neuronal artificial
haciendo exactamente eso.
doing exactly that.
on a mobile phone,
en un teléfono móvil,
amazing in its own right,
sorprendente en sí mismo,
billions and trillions of operations
de operaciones por segundo.
picture of a bird,
de un pájaro una tras otra.
"Yes, it's a bird,"
"Sí, es un pájaro"
with a network of this sort.
de pájaros con una red de este tipo.
y la Y es la desconocida.
and the y is the unknown.
la parte más difícil, por supuesto,
difficult part, of course,
do we figure out the w,
podemos averiguar la W,
un modelo de este tipo?
of solving for w,
de despejar W,
with the simple equation
cómo hacer eso: 6 = 2 x W,
it's the inverse to multiplication,
por ser la inversa de la multiplicación,
algo de mentira aquí.
very non-linear operation;
nada lineal;
to solve the equation
una manera de resolver la ecuación
es bastante sencilla.
is fairly straightforward.
a little algebra trick,
un pequeño truco de álgebra,
to the right-hand side of the equation.
el lado derecho de la ecuación.
about it as an error.
de la manera correcta,
for w the right way,
conjeturas para minimizar el error,
to minimize the error,
computers are very good at.
una aproximación inicial:
El error es 4.
sort of play Marco Polo,
una especie de Marco Polo,
successive approximations to w.
aproximaciones sucesivas a W.
but after about a dozen steps,
pero tras una docena de pasos,
which is close enough.
que es lo suficientemente aproximado.
a lot of known x's and known y's
through an iterative process.
de un proceso iterativo.
that we do our own learning.
lo hacemos en nuestro propio aprendizaje.
this is not a bird."
esto no es un pájaro".
for those neural connections.
para esas conexiones neuronales.
x and w fixed to solve for y;
y la W para resolver Y;
que es mucho más difícil,
para el entrenamiento.
Alex Mordvintsev, on our team,
de nuestro equipo,
with what happens if we try solving for x,
qué sucede si intentamos resolver X,
that you've trained on birds,
entrenada en aves,
the same error-minimization procedure,
de minimización de errores,
trained to recognize birds,
para reconocer aves,
generated entirely by a neural network
en su totalidad por una red neuronal
rather than solving for y,
en lugar de resolver Y,
by Mike Tyka in our group,
por Mike Tyka en nuestro grupo,
of William Kentridge's artworks,
de arte de William Kentridge,
over the space of different animals,
de diferentes animales,
to recognize and distinguish
para reconocer y distinguir
morph from one animal to another.
de un animal a la Escher.
have tried reducing
de solo dos dimensiones,
out of the space of all things
fuera del espacio de todas las cosas
over that entire surface,
sobre toda la superficie,
you make a kind of map --
se hace una especie de mapa,
the network knows how to recognize.
que la red sabe reconocer.
"armadillo" is right in that spot.
el armadillo está justo aquí.
of networks as well.
con otros tipos de redes.
to recognize faces,
para reconocer caras,
in a y that says, "me,"
psychedelic picture of me
como un cuadro surrealista,
multiple points of view at once
de vista a la vez
to get rid of the ambiguity
para descartar la ambigüedad
or another pose,
another kind of lighting.
this sort of reconstruction,
of different points of view,
de diferentes puntos de vista,
his own face as a guide image
su propia cara imagen como guía
to reconstruct my own face.
para reconstruir mi propia cara.
that optimization process.
el proceso de optimización.
more like a coherent face,
más parecido a una cara coherente,
with a blank canvas
con un lienzo en blanco
that is itself already some other image.
que en sí es ya una imagen.
esta pequeña demostración.
that is designed to categorize
man-made structures, animals ...
hechas por humanos, animales...
with just a picture of clouds,
con una imagen de las nubes,
what it sees in the clouds.
qué se ve en las nubes.
you spend looking at this,
will see in the clouds.
en las nubes.
to hallucinate into this,
para alucinar,
zooms hallucinates, zooms.
la alucina, la amplía...
of the network, I suppose,
de estado de fuga de la red, supongo,
is eating its own tail.
What do I think I see next?"
¿Qué pienso que veré ahora?"
called "Higher Education" --
Seattle llamada "Educación Superior",
marijuana was legalized.
la marihuana fuera legalizada.
terminar rápidamente
is not constrained.
que esta tecnología no está limitada.
because they're really fun to look at.
porque son muy divertidos.
puramente visual.
Ross Goodwin,
a camera that takes a picture,
a una cámara tomando una foto,
writes a poem using neural networks,
escribe un poema usando redes neuronales,
has been trained
ha sido entrenada
are very intimately connected.
están conectadas muy íntimamente.
things in the world,
cosas diferentes en el mundo,
para generar nuevas cosas.
Michelangelo really did see
realmente vio
any being, any alien
cualquier ser, cualquier alienígena
perceptual acts of that sort
actos de ese tipo
machinery that's used in both cases.
se usa la misma maquinaria.
and creativity are by no means
y la creatividad no son absolutamente
that can do exactly these sorts of things.
que pueden hacer exactamente estas cosas.
the brain is computational.
el cerebro es computacional.
in designing intelligent machinery.
de diseño de máquinas inteligentes.
máquinas inteligentes.
empezando a cumplir
of those early pioneers,
de aquellos pioneros,
is not just about accounting
no es solo contabilidad
siguiendo el modelo de nuestra mente.
we modeled them after our minds.
to understand our own minds better
mejor nuestra propia mente
ABOUT THE SPEAKER
Blaise Agüera y Arcas - Software architectBlaise Agüera y Arcas works on machine learning at Google. Previously a Distinguished Engineer at Microsoft, he has worked on augmented reality, mapping, wearable computing and natural user interfaces.
Why you should listen
Blaise Agüera y Arcas is principal scientist at Google, where he leads a team working on machine intelligence for mobile devices. His group works extensively with deep neural nets for machine perception and distributed learning, and it also investigates so-called "connectomics" research, assessing maps of connections within the brain.
Agüera y Arcas' background is as multidimensional as the visions he helps create. In the 1990s, he authored patents on both video compression and 3D visualization techniques, and in 2001, he made an influential computational discovery that cast doubt on Gutenberg's role as the father of movable type.
He also created Seadragon (acquired by Microsoft in 2006), the visualization technology that gives Photosynth its amazingly smooth digital rendering and zoom capabilities. Photosynth itself is a vastly powerful piece of software capable of taking a wide variety of images, analyzing them for similarities, and grafting them together into an interactive three-dimensional space. This seamless patchwork of images can be viewed via multiple angles and magnifications, allowing us to look around corners or “fly” in for a (much) closer look. Simply put, it could utterly transform the way we experience digital images.
He joined Microsoft when Seadragon was acquired by Live Labs in 2006. Shortly after the acquisition of Seadragon, Agüera y Arcas directed his team in a collaboration with Microsoft Research and the University of Washington, leading to the first public previews of Photosynth several months later. His TED Talk on Seadragon and Photosynth in 2007 is rated one of TED's "most jaw-dropping." He returned to TED in 2010 to demo Bing’s augmented reality maps.
Fun fact: According to the author, Agüera y Arcas is the inspiration for the character Elgin in the 2012 best-selling novel Where'd You Go, Bernadette?
Blaise Agüera y Arcas | Speaker | TED.com