0:00:19.386,0:00:21.246
In the movie "Interstellar,"

0:00:21.270,0:00:24.597
we get an up-close look[br]at a supermassive black hole.

0:00:24.621,0:00:26.764
Set against a backdrop of bright gas,

0:00:26.788,0:00:28.906
the black hole's massive[br]gravitational pull

0:00:28.930,0:00:30.365
bends light into a ring.

0:00:30.389,0:00:32.498
However, this isn't a real photograph,

0:00:32.522,0:00:34.308
but a computer graphic rendering --

0:00:34.332,0:00:37.722
an artistic interpretation[br]of what a black hole might look like.

0:00:38.351,0:00:39.517
A hundred years ago,

0:00:39.541,0:00:43.142
Albert Einstein first published[br]his theory of general relativity.

0:00:43.166,0:00:44.605
In the years since then,

0:00:44.629,0:00:47.602
scientists have provided[br]a lot of evidence in support of it.

0:00:47.626,0:00:50.710
But one thing predicted[br]from this theory, black holes,

0:00:50.734,0:00:53.084
still have not been directly observed.

0:00:53.108,0:00:56.314
Although we have some idea[br]as to what a black hole might look like,

0:00:56.338,0:00:59.117
we've never actually taken[br]a picture of one before.

0:00:59.141,0:01:01.320
However, you might be surprised to know

0:01:01.344,0:01:05.508
that we may be seeing our first picture[br]of a black hole in the next couple years.

0:01:05.532,0:01:09.490
Getting this first picture will come down[br]to an international team of scientists,

0:01:09.514,0:01:11.081
an Earth-sized telescope

0:01:11.105,0:01:13.937
and an algorithm that puts together[br]the final picture.

0:01:13.961,0:01:17.489
Although I won't be able to show you[br]a real picture of a black hole today,

0:01:17.513,0:01:20.424
I'd like to give you a brief glimpse[br]into the effort involved

0:01:20.448,0:01:22.061
in getting that first picture.

0:01:23.968,0:01:25.414
My name is Katie Bouman,

0:01:25.438,0:01:28.004
and I'm a PhD student at MIT.

0:01:28.028,0:01:30.055
I do research in a computer science lab

0:01:30.079,0:01:33.377
that works on making computers[br]see through images and video.

0:01:33.801,0:01:35.963
But although I'm not an astronomer,

0:01:35.987,0:01:37.272
today I'd like to show you

0:01:37.296,0:01:40.199
how I've been able to contribute[br]to this exciting project.

0:01:42.223,0:01:45.154
If you go out past[br]the bright city lights tonight,

0:01:45.178,0:01:47.614
you may just be lucky enough[br]to see a stunning view

0:01:47.638,0:01:49.131
of the Milky Way Galaxy.

0:01:49.655,0:01:52.117
And if you could zoom past[br]millions of stars,

0:01:52.141,0:01:55.896
26,000 light-years toward the heart[br]of the spiraling Milky Way,

0:01:55.920,0:01:59.441
we'd eventually reach[br]a cluster of stars right at the center.

0:01:59.465,0:02:02.671
Peering past all the galactic dust[br]with infrared telescopes,

0:02:02.695,0:02:06.562
astronomers have watched these stars[br]for over 16 years.

0:02:06.586,0:02:09.689
But it's what they don't see[br]that is the most spectacular.

0:02:10.199,0:02:13.265
These stars seem to orbit[br]an invisible object.

0:02:15.559,0:02:17.882
By tracking the paths of these stars,

0:02:17.906,0:02:19.200
astronomers have concluded

0:02:19.224,0:02:22.199
that the only thing small and heavy[br]enough to cause this motion

0:02:22.223,0:02:24.345
is a supermassive black hole --

0:02:24.369,0:02:28.547
an object so dense that it sucks up[br]anything that ventures too close --

0:02:28.571,0:02:30.065
even light.

0:02:30.089,0:02:33.150
But what happens if we were[br]to zoom in even further?

0:02:33.174,0:02:37.907
Is it possible to see something[br]that, by definition, is impossible to see?

0:02:39.509,0:02:42.753
Well, it turns out that if we were[br]to zoom in at radio wavelengths,

0:02:42.777,0:02:44.459
we'd expect to see a ring of light

0:02:44.483,0:02:46.894
caused by the gravitational[br]lensing of hot plasma

0:02:46.918,0:02:48.747
zipping around the black hole.

0:02:48.771,0:02:49.931
In other words,

0:02:49.955,0:02:53.126
the black hole casts a shadow[br]on this backdrop of bright material,

0:02:53.150,0:02:54.992
carving out a sphere of darkness.

0:02:55.446,0:02:58.785
This bright ring reveals[br]the black hole's event horizon,

0:02:58.809,0:03:01.209
where the gravitational pull[br]becomes so great

0:03:01.233,0:03:02.859
that not even light can escape.

0:03:04.793,0:03:07.652
Einstein's equations predict[br]the size and shape of this ring,

0:03:07.676,0:03:10.884
so taking a picture of it[br]wouldn't only be really cool,

0:03:10.908,0:03:13.526
it would also help to verify[br]that these equations hold

0:03:13.550,0:03:16.016
in the extreme conditions[br]around the black hole.

0:03:16.480,0:03:19.038
However, this black hole[br]is so far away from us,

0:03:19.062,0:03:22.160
that from Earth, this ring appears[br]incredibly small --

0:03:22.184,0:03:25.774
the same size to us as an orange[br]on the surface of the moon.

0:03:26.328,0:03:29.152
That makes taking a picture of it[br]extremely difficult.

0:03:30.215,0:03:31.517
Why is that?

0:03:32.082,0:03:35.270
Well, it all comes down[br]to a simple equation.

0:03:35.294,0:03:37.710
Due to a phenomenon called diffraction,

0:03:37.734,0:03:39.089
there are fundamental limits

0:03:39.113,0:03:41.783
to the smallest objects[br]that we can possibly see.

0:03:42.359,0:03:46.031
This governing equation says[br]that in order to see smaller and smaller,

0:03:46.055,0:03:48.642
we need to make our telescope[br]bigger and bigger.

0:03:48.666,0:03:51.735
But even with the most powerful[br]optical telescopes here on Earth,

0:03:51.759,0:03:54.178
we can't even get close[br]to the resolution necessary

0:03:54.202,0:03:56.400
to image on the surface of the moon.

0:03:56.424,0:04:00.041
In fact, here I show one of the highest[br]resolution images ever taken

0:04:00.065,0:04:01.462
of the moon from Earth.

0:04:01.486,0:04:04.043
It contains roughly 13,000 pixels,

0:04:04.067,0:04:08.117
and yet each pixel would contain[br]over 1.5 million oranges.

0:04:08.966,0:04:10.938
So how big of a telescope do we need

0:04:10.962,0:04:13.727
in order to see an orange[br]on the surface of the moon

0:04:13.751,0:04:15.965
and, by extension, our black hole?

0:04:15.989,0:04:18.329
Well, it turns out[br]that by crunching the numbers,

0:04:18.353,0:04:20.963
you can easily calculate[br]that we would need a telescope

0:04:20.987,0:04:22.380
the size of the entire Earth.

0:04:22.404,0:04:23.428
(Laughter)

0:04:23.452,0:04:25.571
If we could build[br]this Earth-sized telescope,

0:04:25.595,0:04:28.520
we could just start to make out[br]that distinctive ring of light

0:04:28.544,0:04:30.727
indicative of the black[br]hole's event horizon.

0:04:30.751,0:04:33.669
Although this picture wouldn't contain[br]all the detail we see

0:04:33.693,0:04:35.269
in computer graphic renderings,

0:04:35.293,0:04:37.702
it would allow us to safely get[br]our first glimpse

0:04:37.726,0:04:40.213
of the immediate environment[br]around a black hole.

0:04:40.807,0:04:42.420
However, as you can imagine,

0:04:42.444,0:04:46.068
building a single-dish telescope[br]the size of the Earth is impossible.

0:04:46.092,0:04:47.979
But in the famous words of Mick Jagger,

0:04:48.003,0:04:49.794
"You can't always get what you want,

0:04:49.818,0:04:52.005
but if you try sometimes,[br]you just might find

0:04:52.029,0:04:53.244
you get what you need."

0:04:53.268,0:04:55.732
And by connecting telescopes[br]from around the world,

0:04:55.756,0:04:59.294
an international collaboration[br]called the Event Horizon Telescope

0:04:59.318,0:05:02.427
is creating a computational telescope[br]the size of the Earth,

0:05:02.451,0:05:03.988
capable of resolving structure

0:05:04.012,0:05:06.211
on the scale of a black[br]hole's event horizon.

0:05:06.535,0:05:09.922
This network of telescopes is scheduled[br]to take its very first picture

0:05:09.946,0:05:11.761
of a black hole next year.

0:05:13.945,0:05:17.283
Each telescope in the worldwide[br]network works together.

0:05:17.307,0:05:20.019
Linked through the precise timing[br]of atomic clocks,

0:05:20.043,0:05:22.700
teams of researchers at each[br]of the sights freeze light

0:05:22.724,0:05:25.686
by collecting thousands[br]of terabytes of data.

0:05:25.710,0:05:30.727
This data is then processed in a lab[br]right here in Massachusetts.

0:05:32.631,0:05:34.425
So how does this even work?

0:05:34.449,0:05:37.852
Remember if we want to see the black hole[br]in the center of our galaxy,

0:05:37.876,0:05:40.858
we need to build this impossibly large[br]Earth-sized telescope?

0:05:40.882,0:05:43.114
For just a second,[br]let's pretend we could build

0:05:43.138,0:05:44.980
a telescope the size of the Earth.

0:05:45.004,0:05:47.459
This would be a little bit[br]like turning the Earth

0:05:47.483,0:05:49.230
into a giant spinning disco ball.

0:05:49.254,0:05:51.454
Each individual mirror would collect light

0:05:51.478,0:05:54.075
that we could then combine[br]together to make a picture.

0:05:54.099,0:05:56.760
However, now let's say[br]we remove most of those mirrors

0:05:56.784,0:05:58.756
so only a few remained.

0:05:58.780,0:06:01.657
We could still try to combine[br]this information together,

0:06:01.681,0:06:03.674
but now there are a lot of holes.

0:06:03.698,0:06:08.071
These remaining mirrors represent[br]the locations where we have telescopes.

0:06:08.095,0:06:12.174
This is an incredibly small number[br]of measurements to make a picture from.

0:06:12.198,0:06:16.036
But although we only collect light[br]at a few telescope locations,

0:06:16.060,0:06:19.483
as the Earth rotates, we get to see[br]other new measurements.

0:06:19.507,0:06:23.326
In other words, as the disco ball spins,[br]those mirrors change locations

0:06:23.350,0:06:26.249
and we get to observe[br]different parts of the image.

0:06:26.273,0:06:30.291
The imaging algorithms we develop[br]fill in the missing gaps of the disco ball

0:06:30.315,0:06:33.348
in order to reconstruct[br]the underlying black hole image.

0:06:33.372,0:06:36.008
If we had telescopes located[br]everywhere on the globe --

0:06:36.032,0:06:37.973
in other words, the entire disco ball --

0:06:37.997,0:06:39.381
this would be trivial.

0:06:39.405,0:06:42.727
However, we only see a few samples,[br]and for that reason,

0:06:42.751,0:06:45.139
there are an infinite number[br]of possible images

0:06:45.163,0:06:48.127
that are perfectly consistent[br]with our telescope measurements.

0:06:48.751,0:06:51.767
However, not all images are created equal.

0:06:52.209,0:06:56.667
Some of those images look more like[br]what we think of as images than others.

0:06:56.691,0:06:59.913
And so, my role in helping to take[br]the first image of a black hole

0:06:59.937,0:07:02.707
is to design algorithms that find[br]the most reasonable image

0:07:02.731,0:07:04.953
that also fits the telescope measurements.

0:07:06.487,0:07:10.429
Just as a forensic sketch artist[br]uses limited descriptions

0:07:10.453,0:07:13.967
to piece together a picture using[br]their knowledge of face structure,

0:07:13.991,0:07:17.306
the imaging algorithms I develop[br]use our limited telescope data

0:07:17.330,0:07:21.652
to guide us to a picture that also[br]looks like stuff in our universe.

0:07:22.176,0:07:25.827
Using these algorithms,[br]we're able to piece together pictures

0:07:25.851,0:07:28.031
from this sparse, noisy data.

0:07:28.055,0:07:32.584
So here I show a sample reconstruction[br]done using simulated data,

0:07:32.608,0:07:34.541
when we pretend to point our telescopes

0:07:34.565,0:07:37.150
to the black hole[br]in the center of our galaxy.

0:07:37.174,0:07:41.629
Although this is just a simulation,[br]reconstruction such as this give us hope

0:07:41.653,0:07:45.106
that we'll soon be able to reliably take[br]the first image of a black hole

0:07:45.130,0:07:47.725
and from it, determine[br]the size of its ring.

0:07:50.178,0:07:53.377
Although I'd love to go on[br]about all the details of this algorithm,

0:07:53.401,0:07:55.575
luckily for you, I don't have the time.

0:07:55.599,0:07:57.600
But I'd still like[br]to give you a brief idea

0:07:57.624,0:07:59.926
of how we define[br]what our universe looks like,

0:07:59.950,0:08:03.740
and how we use this to reconstruct[br]and verify our results.

0:08:05.180,0:08:07.676
Since there are an infinite number[br]of possible images

0:08:07.700,0:08:10.065
that perfectly explain[br]our telescope measurements,

0:08:10.089,0:08:12.734
we have to choose[br]between them in some way.

0:08:12.758,0:08:14.596
We do this by ranking the images

0:08:14.620,0:08:17.454
based upon how likely they are[br]to be the black hole image,

0:08:17.478,0:08:19.960
and then choosing the one[br]that's most likely.

0:08:19.984,0:08:22.378
So what do I mean by this exactly?

0:08:22.402,0:08:24.380
Let's say we were trying to make a model

0:08:24.404,0:08:27.587
that told us how likely an image[br]were to appear on Facebook.

0:08:27.611,0:08:29.312
We'd probably want the model to say

0:08:29.336,0:08:32.893
it's pretty unlikely that someone[br]would post this noise image on the left,

0:08:32.917,0:08:35.336
and pretty likely that someone[br]would post a selfie

0:08:35.360,0:08:36.694
like this one on the right.

0:08:36.718,0:08:38.357
The image in the middle is blurry,

0:08:38.381,0:08:41.020
so even though it's more likely[br]we'd see it on Facebook

0:08:41.044,0:08:42.404
compared to the noise image,

0:08:42.428,0:08:45.388
it's probably less likely we'd see it[br]compared to the selfie.

0:08:45.712,0:08:48.002
But when it comes to images[br]from the black hole,

0:08:48.026,0:08:51.528
we're posed with a real conundrum:[br]we've never seen a black hole before.

0:08:52.012,0:08:54.303
In that case, what is a likely[br]black hole image,

0:08:54.327,0:08:57.265
and what should we assume[br]about the structure of black holes?

0:08:57.789,0:09:00.421
We could try to use images[br]from simulations we've done,

0:09:00.445,0:09:02.975
like the image of the black hole[br]from "Interstellar,"

0:09:02.999,0:09:05.937
but if we did this,[br]it could cause some serious problems.

0:09:07.461,0:09:10.841
What would happen[br]if Einstein's theories didn't hold?

0:09:10.865,0:09:14.876
We'd still want to reconstruct[br]an accurate picture of what was going on.

0:09:14.900,0:09:18.271
If we bake Einstein's equations[br]too much into our algorithms,

0:09:18.295,0:09:21.050
we'll just end up seeing[br]what we expect to see.

0:09:21.074,0:09:23.350
In other words,[br]we want to leave the option open

0:09:23.374,0:09:26.297
for there being a giant elephant[br]at the center of our galaxy.

0:09:26.321,0:09:27.471
(Laughter)

0:09:27.942,0:09:30.931
Different types of images have[br]very distinct features.

0:09:30.955,0:09:34.503
We can easily tell the difference[br]between black hole simulation images

0:09:34.527,0:09:36.803
and images we take[br]every day here on Earth.

0:09:36.827,0:09:39.931
We need a way to tell our algorithms[br]what images look like

0:09:39.955,0:09:43.204
without imposing one type[br]of image's features too much.

0:09:43.755,0:09:45.648
One way we can try to get around this

0:09:45.672,0:09:48.734
is by imposing the features[br]of different kinds of images

0:09:48.758,0:09:52.888
and seeing how the type of image we assume[br]affects our reconstructions.

0:09:54.542,0:09:58.033
If all images' types produce[br]a very similar-looking image,

0:09:58.057,0:10:00.114
then we can start to become more confident

0:10:00.138,0:10:04.341
that the image assumptions we're making[br]are not biasing this picture that much.

0:10:04.365,0:10:07.355
This is a little bit like[br]giving the same description

0:10:07.379,0:10:10.375
to three different sketch artists[br]from all around the world.

0:10:10.399,0:10:13.259
If they all produce[br]a very similar-looking face,

0:10:13.283,0:10:15.076
then we can start to become confident

0:10:15.100,0:10:18.716
that they're not imposing their own[br]cultural biases on the drawings.

0:10:20.040,0:10:23.355
One way we can try to impose[br]different image features

0:10:23.379,0:10:25.820
is by using pieces of existing images.

0:10:26.374,0:10:28.534
So we take a large collection of images,

0:10:28.558,0:10:31.276
and we break them down[br]into their little image patches.

0:10:31.300,0:10:35.585
We then can treat each image patch[br]a little bit like pieces of a puzzle.

0:10:35.609,0:10:39.887
And we use commonly seen puzzle pieces[br]to piece together an image

0:10:39.911,0:10:42.363
that also fits our telescope measurements.

0:10:46.600,0:10:50.343
Different types of images have[br]very distinctive sets of puzzle pieces.

0:10:51.367,0:10:54.173
So what happens when we take the same data

0:10:54.197,0:10:58.327
but we use different sets of puzzle pieces[br]to reconstruct the image?

0:10:58.351,0:11:02.081
Let's first start with black hole[br]image simulation puzzle pieces.

0:11:03.941,0:11:05.532
OK, this looks reasonable.

0:11:05.556,0:11:08.250
This looks like what we expect[br]a black hole to look like.

0:11:08.274,0:11:09.467
But did we just get it

0:11:09.491,0:11:12.805
because we just fed it little pieces[br]of black hole simulation images?

0:11:12.829,0:11:14.709
Let's try another set of puzzle pieces

0:11:14.733,0:11:17.242
from astronomical, non-black hole objects.

0:11:18.274,0:11:20.400
OK, we get a similar-looking image.

0:11:20.424,0:11:22.660
And then how about pieces[br]from everyday images,

0:11:22.684,0:11:25.469
like the images you take[br]with your own personal camera?

0:11:26.672,0:11:28.787
Great, we see the same image.

0:11:28.811,0:11:32.177
When we get the same image[br]from all different sets of puzzle pieces,

0:11:32.201,0:11:34.247
then we can start to become more confident

0:11:34.271,0:11:36.237
that the image assumptions we're making

0:11:36.261,0:11:39.182
aren't biasing the final[br]image we get too much.

0:11:40.046,0:11:43.299
Another thing we can do is take[br]the same set of puzzle pieces,

0:11:43.323,0:11:45.812
such as the ones derived[br]from everyday images,

0:11:45.836,0:11:49.436
and use them to reconstruct[br]many different kinds of source images.

0:11:49.460,0:11:50.731
So in our simulations,

0:11:50.755,0:11:54.530
we pretend a black hole looks like[br]astronomical non-black hole objects,

0:11:54.554,0:11:58.403
as well as everyday images like[br]the elephant in the center of our galaxy.

0:11:58.427,0:12:01.595
When the results of our algorithms[br]on the bottom look very similar

0:12:01.619,0:12:03.715
to the simulation's truth image on top,

0:12:03.739,0:12:07.085
then we can start to become[br]more confident in our algorithms.

0:12:07.109,0:12:08.976
And I really want to emphasize here

0:12:09.000,0:12:10.934
that all of these pictures were created

0:12:10.958,0:12:13.894
by piecing together little pieces[br]of everyday photographs,

0:12:13.918,0:12:16.353
like you'd take with your own[br]personal camera.

0:12:16.377,0:12:19.823
So an image of a black hole[br]we've never seen before

0:12:19.847,0:12:24.331
may eventually be created by piecing[br]together pictures we see all the time

0:12:24.683,0:12:27.328
Imaging ideas like this[br]will make it possible for us

0:12:27.352,0:12:29.971
to take our very first pictures[br]of a black hole,

0:12:29.995,0:12:32.442
and hopefully, verify[br]those famous theories

0:12:32.466,0:12:34.887
on which scientists rely on a daily basis.

0:12:35.731,0:12:38.339
But of course, getting[br]imaging ideas like this working

0:12:38.363,0:12:41.685
would never have been possible[br]without the amazing team of researchers

0:12:41.709,0:12:43.596
that I have the privilege to work with.

0:12:43.920,0:12:45.083
It still amazes me

0:12:45.107,0:12:48.458
that although I began this project[br]with no background in astrophysics,

0:12:48.482,0:12:51.101
what we have achieved[br]through this unique collaboration

0:12:51.125,0:12:53.884
could result in the very first[br]images of a black hole.

0:12:54.408,0:12:57.106
But big projects like[br]the Event Horizon Telescope

0:12:57.130,0:12:59.944
are successful due to all[br]the interdisciplinary expertise

0:12:59.968,0:13:01.758
different people bring to the table.

0:13:02.182,0:13:03.888
We're a melting pot of astronomers,

0:13:03.912,0:13:06.144
physicists, mathematicians and engineers.

0:13:06.168,0:13:08.042
This is what will make it soon possible

0:13:08.066,0:13:10.579
to achieve something[br]once thought impossible.

0:13:10.603,0:13:12.859
I'd like to encourage all of you to go out

0:13:12.883,0:13:14.979
and help push the boundaries of science,

0:13:15.003,0:13:18.904
even if it may at first seem[br]as mysterious to you as a black hole.

0:13:18.928,0:13:20.102
Thank you.

0:13:20.126,0:13:25.689
(Applause)