Blaise Agüera y Arcas: How computers are learning to be creative
Blaise Agüera i Arcas: Com els ordinador aprenen a ser creatius
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.
that works on machine intelligence;
encarregat d'IA.
of making computers and devices
la disciplina d'enginyeria
imitin processos mentals.
that brains do.
interested in real brains
en els cervells reals,
in the things that our brains do
en els processos cerebrals
to the performance of computers.
als processos dels ordinadors.
has been perception,
és la percepció.
out there in the world --
coses del món exterior,
a la nostra ment.
per als nostres cervells
en el cas dels ordinadors.
for example, that our team makes,
que fem a l'equip
on Google Photos to become searchable,
buscar les fotos a Google Photos,
out there into the world.
una cosa que hi ha al món.
our work on machine perception
en la percepció mecànica
with the world of machine creativity
amb el món de la creativitat mecànica,
had a penetrating insight
una profunda comprensió
between perception and creativity.
entre percepció i creativitat.
has a statue inside of it,
hi ha una escultura,
is to discover it."
Michelangelo was getting at
en Miquel Àngel es referia
is an act of imagination
és un acte d'imaginació,
and perceiving and imagining,
percep i imagina és,
with a brief bit of history
amb un xic d'història,
the heart or the intestines,
com el cor o els intestins,
about a brain by just looking at it,
amb només mirar-lo,
que observaren el cervell,
of this thing all kinds of fanciful names,
tota mena de noms originals,
doesn't tell us very much
no ens en diuen pas gaire
developed some kind of insight
va desenvolupar algun tipus de coneixement
Santiago Ramón y Cajal,
Santiago Ramón y Cajal,
i colorants especials
or render in very high contrast
per tal d'aconseguir un contrast molt alt
their morphologies.
les seves morfologies.
that he made of neurons
en són el resultat,
of different sorts of cells,
de tipus de cèl·lules que hi ha,
was quite new at this point.
era innovadora en aquell moment.
very, very long distances --
molt i molt lluny,
to some people in the 19th century;
per a alguna gent del segle XIX,
were just getting underway.
la revolució elèctrica i dels cables.
of Ramón y Cajal's, like this one,
d'en Ramón y Cajal, com aquest,
són inmillorables.
that Ramón y Cajal started.
que ell va començar.
dels nostres col·laboradors,
of Neuroscience.
de teixits cerebrals.
is about one cubic millimeter in size,
té una mida d'un mil·límetre cúbic,
very small piece of it here.
és una micra, si fa no fa.
són les mitocòndries.
tiny block of tissue.
bloc de teixit.
of hair is about 100 microns.
té unes 100 micres.
much, much smaller
quelcom molt més petit
electron microscopy slices,
de porcions de microscòpia d'electrons,
in 3D of neurons that look like these.
de neurones en 3D. Són així.
style as Ramón y Cajal.
a les de l'estil de Ramón y Cajal.
certes neurones
be able to see anything here.
no podríem veure res.
one neuron to another.
entre una neurona i una altra.
ahead of his time,
un avançat al seu temps.
a la comprensió del cervell
over the next few decades.
durant les dècades següents.
que les neurones usen electricitat.
was advanced enough
la tecnologia havia avançat prou
experiments on live neurons
amb neurones vives,
when computers were being invented,
s'inventaren els primers ordinadors,
of modeling the brain --
el cervell humà,
as Alan Turing called it,
com va dir l'Alan Turing,
looked at Ramón y Cajal's drawing
observaren els dibuixos d'en Ramon y Cajal
imagery that comes from the eye.
que reben els ulls.
like a circuit diagram.
un esquema de connexions.
in McCulloch and Pitts's circuit diagram
d'en McCulloch i en Pitt
of computational elements
com una serie d'elements computacionals
one to the next in a cascade,
en una cascada
visual information would need to do.
de processador d'informació visual.
for us to do with our brains.
per als nostres cervells.
that for a computer,
en el cas d'un ordinador,
just a few years ago.
només fa uns anys.
this task is easy to do.
és fàcil de fer.
and the word "bird,"
i la paraula "ocell",
connected to each other
un conjunt de neurones interconnectades
inside our visual cortices,
dins el nostre còrtex visual,
to have the capability
a ser capaços de fer,
on the computer.
a l'ordinador.
that actually looks like.
about as a first layer of neurons,
com una primera capa de neurones,
how it works in the eye --
en el cas de l'ull,
after another layer of neurons,
de neurones,
of different weights.
de diferents pesos.
of all of those synapses.
de totes aquestes sinapsis.
properties of this network.
les propietats computacionals de la xarxa.
or a small group of neurons
o un petit grup de neurones
those three things --
in the neural network,
a la xarxa neuronal
these synapses in the neural network.
a la xarxa neuronal.
is just a simple formula,
com una fórmula senzilla:
going on there, of course,
per descomptat,
of mathematical operations.
molt complicades.
that if you have one equation,
by knowing the other two things.
per mitjà de conèixer les altres dues.
that the picture of a bird is a bird,
la imatge de l'ocell és un ocell,
and w and x are known.
i 'w' i 'x' són valors coneguts.
you know the pixels.
sabem les píxels.
a relatively straightforward problem.
d'un problema relativament senzill.
and you're done.
estaria resolt.
una xarxa neuronal artificial,
doing exactly that.
seguint aquesta idea.
on a mobile phone,
des d'un telèfon mòbil,
amazing in its own right,
prou sorprenent;
billions and trillions of operations
milers de milions i bilions d'operacions,
picture of a bird,
imatges d'ocells.
"Yes, it's a bird,"
i dient "Sí, és un ocell",
with a network of this sort.
per mitjà d'aquesta xarxa.
and the y is the unknown.
i 'y' és la incògnita.
difficult part, of course,
la part més difícil,
do we figure out the w,
of solving for w,
de resoldre el valor de 'w',
with the simple equation
it's the inverse to multiplication,
perquè és l'oposat a la multiplicació,
very non-linear operation;
una operació no lineal,
to solve the equation
la manera de resoldre l'equació
is fairly straightforward.
és bastant directa.
a little algebra trick,
un petit truc d'àlgebra.
to the right-hand side of the equation.
about it as an error.
for w the right way,
correctament,
to minimize the error,
per tal de minimitzar l'error,
computers are very good at.
en aquest tipus de coses.
L'error és 4.
sort of play Marco Polo,
a Marco Polo,
successive approximations to w.
fa aproximacions successives a 'w'.
but after about a dozen steps,
però desprès d'uns 12 passos,
which is close enough.
que està prou a prop.
a lot of known x's and known y's
molts valors coneguts 'x' i 'y'
through an iterative process.
mitjançant un procés de repetició.
that we do our own learning.
en la que aprenem nosaltres mateixos.
this is not a bird."
"Això és un ocell, això no ho és".
for those neural connections.
resolent les seves connexions neurals.
x and w fixed to solve for y;
'x' i 'w' per tal de resoldre 'y'
Alex Mordvintsev, on our team,
del nostre equip,
with what happens if we try solving for x,
què passa en intentar resoldre 'x',
that you've trained on birds,
entrenada en ocells,
the same error-minimization procedure,
de minimitzar l'error,
trained to recognize birds,
en reconèixer ocells,
generated entirely by a neural network
es genera totalment per la xarxa neural
rather than solving for y,
by Mike Tyka in our group,
del nostre grup.
of William Kentridge's artworks,
l'obra d'en William Kentridge,
over the space of different animals,
en l'espai de diferents animals,
to recognize and distinguish
que reconeix i distingeix
morph from one animal to another.
d'animals transformant-se'n altres.
have tried reducing
out of the space of all things
fora de l'espai de totes les coses
over that entire surface,
sobre tota aquesta superfície
you make a kind of map --
es genera una mena de mapa;
the network knows how to recognize.
que la xarxa pot reconèixer.
"armadillo" is right in that spot.
com l'armadillo just en aquest punt.
of networks as well.
altra mena de xarxes.
to recognize faces,
dissenyada per a reconèixer cares.
in a y that says, "me,"
on hi posa "jo"
psychedelic picture of me
i psicodèlica foto meva,
multiple points of view at once
múltiples punts de vista alhora
to get rid of the ambiguity
per a obviar l'ambigüitat
or another pose,
amb un gest o un altre,
another kind of lighting.
o una altra de diferent.
this sort of reconstruction,
algun tipus d'imatge de guia,
of different points of view,
de diferents punts de vista,
his own face as a guide image
usa la seva cara com a imatge de guia
to reconstruct my own face.
de reconstruir la meva cara.
that optimization process.
aquest procés d'optimització.
more like a coherent face,
quelcom semblant a una cara,
with a blank canvas
d'un llenç en blanc
that is itself already some other image.
és a dir, amb una altra imatge.
that is designed to categorize
per tal de categoritzar
man-made structures, animals ...
estructures fetes per l'home, animals...
with just a picture of clouds,
fotografia de núvols
what it sees in the clouds.
què hi veu als núvols.
you spend looking at this,
observant-ho,
will see in the clouds.
als núvols.
to hallucinate into this,
de reconeixement facial per al·lucinar,
zooms hallucinates, zooms.
al·lucina, fa zoom...
of the network, I suppose,
de fuga dissociativa a la xarxa,
is eating its own tail.
What do I think I see next?"
que ho vaig ensenyar en públic
called "Higher Education" --
a Seattle anomenat "Higher Education",
marijuana was legalized.
de legalitzar la marihuana.
is not constrained.
tecnologia no és només això.
because they're really fun to look at.
perquè són divertits,
de tecnologia visual.
en Ross Goodwin,
a camera that takes a picture,
fent fotografies amb una càmera
writes a poem using neural networks,
en faci un poema amb una xarxa neural;
has been trained
s'ha preparat
del segle XX.
are very intimately connected.
estan lligades molt íntimament.
discriminar
things in the world,
del món;
per tal de poder generar.
Michelangelo really did see
en Miquel Àngel era capaç de veure
any being, any alien
qualsevol ésser, qualsevol marcià
perceptual acts of that sort
d'actes de percepció,
machinery that's used in both cases.
el mateix mecanisme en ambdós casos.
and creativity are by no means
la percepció com la creativitat no són
that can do exactly these sorts of things.
capaços de dur a terme aquestes tasques.
the brain is computational.
ja que el cervell és computacional.
in designing intelligent machinery.
de disseny de màquines intel·ligents.
a la idea
of those early pioneers,
d'aquells primers pioners,
is not just about accounting
no és només comptabilitat,
we modeled them after our minds.
el model de les nostres ments.
to understand our own minds better
d'entendre'ns millor a nosaltres mateixos
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