True Artificial Intelligence will change everything | Juergen Schmidhuber | TEDxLakeComo
-
0:02 - 0:06[Music]
-
0:06 - 0:15When I was a boy, I wanted to
maximise my impact on the world, -
0:15 - 0:20and I was smart enough to realise
that I am not very smart. -
0:21 - 0:29And that I have to build a machine that
learns to become much smarter than myself, -
0:30 - 0:34such that it can solve all the problems
that I cannot solve myself, -
0:35 - 0:36and I can retire.
-
0:38 - 0:42And my first publication on
that dates back 30 years: 1987. -
0:43 - 0:48My diploma thesis, where I already try
to solve the grand problem of AI, -
0:49 - 0:53not only build a machine that learns a
little bit here, learns a little bit there, -
0:53 - 0:58but also learns to improve
the learning algorithm itself. -
1:00 - 1:04And the way it learns, the way
it learns and so on recursively, -
1:05 - 1:11without any limits except
the limits of logics and physics. -
1:13 - 1:16And, I'm still working
on the same old thing, -
1:16 - 1:20and I'm still pretty much
saying the same thing, -
1:20 - 1:23except, that now, more people
are listening. -
1:25 - 1:30Because, the learning algorithms that we
have developed on the way to this goal, -
1:30 - 1:34they are now on three thousand
million smartphones. -
1:35 - 1:37And all of you have them
in your pockets. -
1:40 - 1:46What you see here are the five most
valuable companies of the Western world: -
1:46 - 1:51Apple, Google, Facebook,
Microsoft, and Amazon. -
1:52 - 1:58And all of them are emphasising that AI,
artificial intelligence, -
1:58 - 2:00is central to what they are doing.
-
2:02 - 2:06And all of them,
are using heavily, -
2:07 - 2:11the deep learning methods that my team
has developed since the early nineties, -
2:11 - 2:14in Munich and in Switzerland.
-
2:14 - 2:19Especially something which is called:
the long short-term memory. -
2:19 - 2:24Has anybody in this room ever heard
of the long short-term memory? -
2:24 - 2:28Or the LSTM? Hands up,
anybody ever heard of that? -
2:28 - 2:33Okay. Has anybody never heard of
the LSTM? -
2:37 - 2:46I see we have a third group in this room:
who didn't understand the question. -
2:49 - 2:52The LSTM is a little bit like your brain:
-
2:53 - 2:57it's an artificial neural network
which also has neurons, -
2:58 - 3:03and in your brain, you've got about
100 billion neurons. -
3:03 - 3:10And each of them is connected to
roughly 10,000 other neurons on average, -
3:11 - 3:15Which means that you have got
a million billion connections. -
3:16 - 3:20And each of these connections has
a strength which says, -
3:20 - 3:25how much this neuron over here, influences
that one over there at the next time step, -
3:25 - 3:30And in the beginning all these connections
are random and the system knows nothing, -
3:30 - 3:36but then, through a smart learning algorithm,
it learns from lots of examples, -
3:37 - 3:43to translate the incoming data,
such as video through the cameras, -
3:43 - 3:46or audio through the microphones,
-
3:46 - 3:49or pain signals through the
pain sensors. -
3:49 - 3:52It learns to translate that
into output actions, -
3:52 - 3:55because some of these neurons are
output neurons, -
3:55 - 3:58that control speech muscles
and finger muscles. -
3:59 - 4:03And only through experience,
it can learn to solve -
4:03 - 4:08all kinds of interesting problems,
such as driving a car -
4:10 - 4:14or, do the speech recognition
on your smartphone. -
4:14 - 4:17Because, whenever you
take out your smartphone, -
4:17 - 4:19an Android phone, for example,
and you speak to it, -
4:19 - 4:24and you say: "Ok Google, show me
the shortest way to Milano." -
4:24 - 4:26Then it understands your speech,
-
4:27 - 4:29because there is a LSTM in there,
-
4:29 - 4:32which has learned to understand speech.
-
4:32 - 4:35Every 10 milliseconds,
100 times a second, -
4:35 - 4:38new inputs are coming
from the microphone, -
4:38 - 4:40and then translates it
-
4:40 - 4:49after thinking into letters which is then
question to the search engine and it has -
4:49 - 4:54long to do that by listening to lots of
speech from women from me all kinds of -
4:55 - 4:57people and that's how since
-
4:57 - 5:012015 Google speech recognition is
now much better than it used to be -
5:02 - 5:08the basic lsdm cell looks like that I
don't have the time to explain that but at -
5:08 - 5:14least I can list the names of
the brilliant students in my lab who made -
5:14 - 5:18that possible and what are
the big companies doing with -
5:18 - 5:26that well speech recognition is only
one example if you are on Facebook is -
5:26 - 5:29anybody on Facebook okay I
use sometimes clicking at -
5:29 - 5:33the translate button because somebody sent
you something in a foreign language and -
5:33 - 5:38then you can translate it is anybody
doing that yeah whatever you do that you -
5:38 - 5:42are waking up again
a long short term memory and lsdm which -
5:42 - 5:49has learned to translate text in
one language into translated text and -
5:49 - 5:57Facebook is doing that four billion times
a day so every 50 every second -
5:57 - 6:0350,000 sentences are being translated by
an LST am working for -
6:04 - 6:06Facebook and another 50,000 in the
second and another -
6:06 - 6:1350,000 and to see how much this thing
is now permitting the modern world -
6:13 - 6:21just note that almost 30 percent of
the awesome computational power for -
6:21 - 6:24interference and all these Google Data
-
6:24 - 6:29Centers all these data centers of Google
are all over the world is used for LST on -
6:29 - 6:32almost 30 percent if you have an
-
6:32 - 6:39Amazon echo you can ask a questions and it
answers you and the voice that you hear -
6:39 - 6:43it's not a recording it's
an LS TM network which has learned from -
6:43 - 6:54training examples to sound like
a female voice if you have an iPhone and -
6:54 - 6:57you're using
the quick type it's trying to predict what -
6:57 - 7:00you want to do next given
all the previous context of what you did -
7:00 - 7:05so far again that's an LS DM which
has to do that so it's on -
7:05 - 7:15a billion iPhones you are a large audience
by my standards but when we started -
7:15 - 7:22this work decades ago in the early
90s only few people who were interested -
7:22 - 7:26in that because computers were so slow and
you couldn't do so much with it and I -
7:26 - 7:33remember I gave a talk at
a conference and there was just -
7:33 - 7:37one single person in the audience
a young lady I said young lady it's -
7:38 - 7:42very embarrassing but apparently
today I'm going to give this talk just to -
7:42 - 7:54you and she said okay but please hurry I
am the next speaker since then we -
7:54 - 8:01have greatly profited from the fact
that every five years computers again in -
8:01 - 8:06ten times cheaper which is
an old trend that has held since 1941 at -
8:06 - 8:12least since this man Conrad Susan built
the first working program control computer -
8:12 - 8:20in Berlin and he could could do roughly
one operation per second one and then -
8:20 - 8:26ten years later for the same prize one
could do 100 operations 30 years later -
8:26 - 8:301 million operations were
the same price and today after 75 years we -
8:30 - 8:35can do a million billion times as much for
the same price and the trend is not about -
8:35 - 8:44to stop because the physical limits are
much further out there rather soon and not -
8:45 - 8:49so many years or decades we will for
the first time have -
8:49 - 8:55little computational devices that
can compute as much as a human brain and -
8:55 - 8:59this a trend doesn't break 50 years
later there will be -
8:59 - 9:04a little computational device for
the same price that can compute as much as -
9:04 - 9:10all 10 billion human brains taken
together and there will not only be one of -
9:10 - 9:13those devices but
many many many everything -
9:13 - 9:18is going to change already in
2011 computers were fast enough such that -
9:18 - 9:22our deep learning methods for
the first time could achieve -
9:22 - 9:28a superhuman pattern-recognition result and
was the first superhuman result and -
9:28 - 9:32the history of
computer vision and back then computers -
9:32 - 9:36were 20 times more expensive than today so
today for the same price we can do -
9:36 - 9:4520 times as much and just a few five years
ago five years ago when computers were -
9:45 - 9:5010 times more expensive than today we
already could win for the first time -
9:50 - 9:54medical imaging competitions what you see
behind me is a slice through -
9:54 - 10:00the female breast and the tissue that you
see there has all kinds of -
10:00 - 10:05cells and normally you need
a trained doctor a trained the solid who -
10:05 - 10:10is able to detect
the dangerous cancer cells or -
10:10 - 10:15pre-cancer cells now our stupid network
knows nothing about cancer knows nothing -
10:15 - 10:19about vision it knows nothing in
the beginning but we can train it -
10:19 - 10:25to imitate the human teacher
the doctor and it became as good or better -
10:25 - 10:31than the best competitors and
very soon all of medical diagnosis -
10:31 - 10:36is going to be superhuman and
it's going to be mandatory because -
10:36 - 10:42it's going to be so much better than
the doctors after this all kinds of -
10:42 - 10:48medical imaging startups
were founded focusing just on this because -
10:48 - 10:54it's so important we can also use lsdm
to train robots one important thing I -
10:54 - 11:01want to say is that we not only have
systems that slavishly imitate what humans -
11:01 - 11:09show them no we also have a eyes that set
themselves their own goals and -
11:09 - 11:14like little babies invent
their own experiment to explore -
11:14 - 11:19the world and to figure out what you
can do in the world without a teacher and -
11:19 - 11:23becoming more and
more general problem solvers in -
11:23 - 11:28the process by learning new skills on top
of old skills and this is going to -
11:28 - 11:34scale we call that artificial curiosity or
a recent password is power plain -
11:34 - 11:39learning to become a more and
more general problems over by -
11:40 - 11:44learning to invent like a scientist
one new interesting goal after -
11:45 - 11:50Nathan and and it's going to scale and I
think in not so many years from now for -
11:50 - 11:56the first time we are going to have
an animal like -
11:56 - 12:01AI you don't have that yet on the level of
a little crowd which -
12:01 - 12:07already can learn to use two worlds for
example little monkey and once we have -
12:07 - 12:11that it may take just a few decades to do
the final step towards -
12:11 - 12:17human level intelligence because
technological evolution is about -
12:17 - 12:22a million times a million times faster
than biological evolution and -
12:22 - 12:30biological evolution needed
3.5 billion years to evolve a monkey -
12:30 - 12:34a monkey from scratch but then just
a few tens of millions of years -
12:34 - 12:39afterwards to evolve
human level intelligence we have -
12:39 - 12:43a company which is called Mason's
like birth in English -
12:43 - 12:47Mason's but spelled in
a different way which is trying to make -
12:47 - 12:51this a reality and build
the first true general and purpose AI at -
12:53 - 13:00the moment almost all research in AI is
very human centric and it's all about -
13:00 - 13:06making human lives longer and
healthier and easier and making humans -
13:06 - 13:12more addicted to their smartphones but
in the long run a eyes are going to -
13:12 - 13:17especially the smart ones are going to set
themselves their own goals and I have -
13:17 - 13:22no doubt in my mind that they
are going to become much smarter than we -
13:22 - 13:26are and what are they going to do of
course they are going to realize what we -
13:26 - 13:31have realized a long time ago namely that
most of the resources in -
13:31 - 13:38the solar system or in general are not in
our little biosphere they are out there in -
13:38 - 13:44space and so of course they
are going to emigrate and of course they -
13:44 - 13:53are going to use trillions of
self-replicating robot factories to expand -
13:54 - 13:57in form of growing
-
13:57 - 14:01AI bubble which within a few
hundred thousand years is going to cover -
14:01 - 14:07the entire galaxy by senders and receivers
such that a eyes can travel the way they -
14:07 - 14:16are already traveling in my lab by radio
from sender to receiver Wireless so what -
14:16 - 14:23we are witnessing now is much more than
just another Industrial Revolution this is -
14:24 - 14:29something that transcends
humankind and even life itself -
14:29 - 14:35the last time something so important
has happened was maybe 3.5 billion years -
14:35 - 14:42ago when life was invented a new type of
life is going to emerge from -
14:42 - 14:46our little planet and
it's going to colonize and transform -
14:47 - 14:52the entire universe the universe is
still young it's only 13.8 billion years -
14:52 - 14:58old it's going to become much older than
that many times more many times older -
14:58 - 15:03than that so there's plenty of time
to reach all of it or all of -
15:03 - 15:09the visible parts totally within
the limits of light speed and physics -
15:10 - 15:15a new type of life is going to make
the universe intelligent now of course we -
15:15 - 15:22are not going to remain the crown of
creation of course not but there is still -
15:22 - 15:29beauty in seeing yourself as part of
a grander process that leads the cosmos -
15:29 - 15:36from low complexity towards
higher complexity it's a privilege to live -
15:36 - 15:40at a time where we can
witness the beginnings of that and where -
15:40 - 15:50we can contribute something to
that thank you for your patience
- Title:
- True Artificial Intelligence will change everything | Juergen Schmidhuber | TEDxLakeComo
- Description:
-
- Video Language:
- English
- Team:
- closed TED
- Duration:
- 15:56
Show all