00:00:00
humans rule Earth without competition
00:00:03
but we're about to create something that
00:00:05
may change that our last invention the
00:00:08
most powerful tool weapon or maybe even
00:00:10
entity artificial super
00:00:13
intelligence this sounds like science
00:00:16
fiction so let's start at the beginning
00:00:19
intelligence is the ability to learn
00:00:22
reason acquire Knowledge and Skills and
00:00:24
use them to solve problems intelligence
00:00:27
is power and we're the species that
00:00:29
exploited it the most so much so that
00:00:32
Humanity broke the game of Nature and
00:00:35
took control but the journey there
00:00:37
wasn't straightforward for most animals
00:00:40
intelligence costs too much energy to be
00:00:42
worth it still if we track intelligence
00:00:45
in the tree of species over time we can
00:00:47
see lots of diverse forms of
00:00:49
intelligence emerge the earliest brains
00:00:52
were in flatworms 500 million years ago
00:00:55
just a tiny cluster of neurons to handle
00:00:57
basic body functions it took hundreds of
00:01:00
millions of years for species to
00:01:01
diversify and become more complex life
00:01:05
conquered New environments gained new
00:01:07
senses and had to contend with Fierce
00:01:09
competition over resources but in nature
00:01:12
all that matters is survival and brains
00:01:15
are expensive so for almost all animals
00:01:17
a narrow intelligence fit for a narrow
00:01:19
range of tasks was enough in some
00:01:22
environments animals like birds
00:01:24
octopuses and mammals evolved more
00:01:26
complex neural structures for them it
00:01:29
paid off to have more energy consuming
00:01:31
skills like Advanced navigation and
00:01:33
communication until 7 billion years ago
00:01:36
the hominins emerged we don't know why
00:01:39
but their brains grew faster than their
00:01:41
relatives something was different about
00:01:43
their intelligence very slowly it turned
00:01:46
from narrow to General from a
00:01:48
screwdriver to a multi-tool able to
00:01:51
think about diverse Problems 2 million
00:01:55
years ago homo erector saw the world
00:01:57
differently from anyone before as
00:01:59
something be understood and transformed
00:02:02
they controlled fire invented tools and
00:02:05
created the first culture we probably
00:02:08
emerged from them around 250,000 years
00:02:11
ago with an even larger and more complex
00:02:14
brain it enabled us to work together in
00:02:16
large groups and to communicate complex
00:02:19
thoughts we used our intelligence to
00:02:21
improve our lives to ask how things work
00:02:24
and why things are the way they are with
00:02:26
each Discovery we asked more questions
00:02:29
and pushed forward Ward preserving what
00:02:31
we learned outpacing what evolution
00:02:33
could do with genes knowledge Builds on
00:02:36
knowledge progress was slow at first and
00:02:39
then sped up exponentially agriculture
00:02:42
writing medicine astronomy or philosophy
00:02:44
exploded into the world 200 years ago
00:02:47
science took off and made us even better
00:02:49
at learning about the world and speeding
00:02:51
up progress 35 years ago the internet
00:02:54
age began today we live in a world made
00:02:57
to suit our needs created by us us for
00:03:00
us this is incredibly new we forget how
00:03:04
hard it was to get here how enormous the
00:03:07
steps on the intelligence ladder were
00:03:09
and how long it took to climb them but
00:03:11
once we did we became the most powerful
00:03:14
animal in the world in a
00:03:16
heartbeat but we may be in the process
00:03:18
of changing this we're building machines
00:03:21
that could be better at the very thing
00:03:23
that gave us the power to conquer the
00:03:25
planet Humanity's final invention
00:03:30
artificial
00:03:31
intelligence artificial intelligence or
00:03:34
AI is software that performs mental
00:03:36
tasks with a computer code that uses
00:03:38
silicon instead of neurons to solve
00:03:41
problems in the beginning AI was very
00:03:43
simple lines of code on paper mere
00:03:46
proofs of concept to demonstrate how
00:03:48
machines could perform mental tasks only
00:03:51
in the 1960s did we start seeing the
00:03:53
first examples of what we would
00:03:55
recognize as AI a chatbot in 1964
00:03:59
approach program to sort through
00:04:00
molecules in
00:04:02
1965 slow specialized systems requiring
00:04:06
experts to use them their intelligence
00:04:08
was extremely narrow built for a single
00:04:11
task inside a controlled environment the
00:04:13
equivalent of flat worms 500 million
00:04:16
years ago doing the minimum amount of
00:04:18
mental work progress in AI research
00:04:21
paused several times when researchers
00:04:23
lost hope in the technology but just
00:04:25
like changing environments create new
00:04:27
niches for Life the world around AI
00:04:30
changed between 1950 and 2000 computers
00:04:33
got a billion times faster while
00:04:35
programming became easier and widespread
00:04:38
in 1972 AI could navigate a room in 1989
00:04:42
it could read handwritten numbers but it
00:04:45
remained a fancy tool no match for
00:04:47
humans until in 1997 an AI shocked the
00:04:51
World by beating the world champion in
00:04:53
chess proving that we could build
00:04:55
machines that could surpass us but we
00:04:57
calmed ourselves because a chest spot is
00:05:00
quite stupid not a flatworm but maybe a
00:05:03
bee only able to perform a specialized
00:05:06
narrow task but within this narrow task
00:05:09
it's so good that no human will ever
00:05:11
again beat AI a chess as computers
00:05:15
continued to improve AI became a
00:05:17
powerful tool for more and more tasks in
00:05:20
2004 it drove a robot on Mars in 2011 it
00:05:23
began recommending YouTube videos to you
00:05:26
but this was only possible because
00:05:28
humans broke down problems into easyto
00:05:30
digest chunks that computers could solve
00:05:32
quickly until we taught AIS to teach
00:05:37
themselves rise of the self-learning
00:05:41
Machines this is not a technical video
00:05:43
so we're massively oversimplifying here
00:05:46
in a nutshell the sheer power of
00:05:47
supercomputers was combined with the
00:05:49
almost endless data collected in the
00:05:51
information age to make a new generation
00:05:54
of ai ai experts began drastically
00:05:57
improving forms of AI software called
00:05:59
neur neural networks enormously huge
00:06:01
networks of artificial neurons that
00:06:03
start out being bad at their tasks they
00:06:06
then used machine learning which is an
00:06:08
umbrella term for many different
00:06:10
training techniques and environments
00:06:11
that allows algorithms to write their
00:06:13
own code and improve themselves the
00:06:16
scary thing is that we don't exactly
00:06:17
know how they do it and what happens
00:06:20
inside them just that it works and that
00:06:23
what comes out the other end is a new
00:06:25
type of AI a capable black box of code
00:06:29
the these new AIS could Master complex
00:06:31
skills extremely quickly with much less
00:06:34
human help they were still narrow
00:06:36
intelligences but a huge step up in 2014
00:06:40
Facebook AI could identify faces with
00:06:42
97% accuracy in 2016 an AI beat the best
00:06:47
humans in the incredibly complex game of
00:06:49
Go in 2018 a self-learning AI learned
00:06:53
chess in 4 hours just by playing against
00:06:56
itself and then defeated the best
00:06:58
specialized chess bot since then machine
00:07:01
learning has been applied to reading
00:07:03
image processing solving tests and much
00:07:06
more many of these AIS are already
00:07:08
better than humans for whatever narrow
00:07:10
task they were trained but they still
00:07:12
remained a simple tool AI still didn't
00:07:15
seem that big of a deal for most people
00:07:17
and then came the chatbot chat GPT the
00:07:21
work that went into it is massive it
00:07:23
trained on nearly everything written on
00:07:24
the internet to learn how to handle
00:07:26
language which it now does better than
00:07:28
most people it can summarize translate
00:07:31
and help with some math problems it's
00:07:34
incredibly more broad than any other
00:07:36
system just a few years ago not crushing
00:07:39
any single Benchmark but order of them
00:07:41
at once many large tech companies are
00:07:44
spending billions to build powerful
00:07:46
competitors AI is already transforming
00:07:49
customer service banking Healthcare
00:07:51
marketing copyrighting creative spaces
00:07:54
and more AI generated content has
00:07:57
already taken hold of social media
00:07:59
YouTube and news websites elections are
00:08:02
expected to be inundated by Propaganda
00:08:04
and
00:08:05
misinformation no one is sure how much
00:08:07
good or harm can come from adopting AI
00:08:09
everywhere change is scary there will be
00:08:12
winners and losers one of the biggest
00:08:15
questions governments and corporations
00:08:17
have now is how to manage the transition
00:08:19
to an AI boosted economy all these
00:08:22
potential gains or risks are just the
00:08:24
result of today's AI chat gpt's
00:08:27
intelligence is a major step up but it
00:08:30
remains narrow one it can write a great
00:08:32
essay in seconds it doesn't understand
00:08:35
what it's writing but what if the AIS
00:08:38
stopped being
00:08:39
narrow General AI what makes humans
00:08:43
different from current AI is our general
00:08:46
intelligence humans can technically
00:08:48
absorb any piece of knowledge and start
00:08:50
working on any problem we're great at
00:08:52
many very different skills and tasks
00:08:54
from playing chess to writing or solving
00:08:56
science puzzles not equally of course
00:08:59
some of us are experts in some fields
00:09:01
and beginners in others but we can
00:09:03
technically do all of them in the past
00:09:06
AI was narrow and able to become good at
00:09:08
one skill but was rather bad in all the
00:09:10
others simply by building faster
00:09:13
computers and pouring more money into AI
00:09:15
training will get us new more powerful
00:09:18
generations of AI but what is the next
00:09:21
step for AI is to become a general
00:09:24
intelligence like us an
00:09:26
AGI if the AI improvement process
00:09:29
continues as it has been it's not
00:09:31
unlikely that AGI could be better in
00:09:33
most or even all skills that humans can
00:09:36
do we don't know how to build AGI how it
00:09:39
will work or what it will be able to do
00:09:42
since narrow AIS today are capable of
00:09:45
mastering one mental task quickly HGI
00:09:47
might be able to do the same with all
00:09:49
mental tasks so even if it starts out
00:09:52
stupid and HGI might be able to become
00:09:55
as smart and capable as a human while
00:09:58
this sounds like science sence fiction
00:10:00
most AI researchers think this will
00:10:02
happen sometime this Century maybe
00:10:04
already in a few years humanity is not
00:10:07
ready for what will happen next not
00:10:09
socially not economically not morally
00:10:13
earlier we defined intelligence as the
00:10:15
ability to learn reason acquire
00:10:17
Knowledge and Skills and use them to
00:10:19
solve problems all things humans excel
00:10:22
at an AGI as intelligent as even an
00:10:25
average human would already disrupt
00:10:27
modern civilization because they're not
00:10:29
Bound by the same limitations as we are
00:10:32
today's AIS like chat gbt already think
00:10:35
and solve the tasks they were made for
00:10:37
at least 10 times faster than even very
00:10:39
skilled humans maybe AGI will be slower
00:10:43
but it may also be faster maybe much
00:10:45
faster and since HGI are software you
00:10:49
could copy them endlessly as long as you
00:10:51
have enough storage and run them in
00:10:54
parallel there are 8 million scientists
00:10:56
in the world now imagine an AI copied a
00:10:59
million times and put to work imagine 1
00:11:02
million scientists working 24/7 thinking
00:11:05
10 times faster than humans without
00:11:07
being distracted only focused on the
00:11:10
task they've been given what if suddenly
00:11:13
HGI could do all intelligence-based jobs
00:11:16
in the world from interpreting law to
00:11:18
coding to creating animated YouTube
00:11:20
videos better faster and much cheaper
00:11:23
than humans would whoever controls this
00:11:26
AGI suddenly own the economy
00:11:29
and thinking bigger human progress is
00:11:32
our intelligence applied to problems so
00:11:35
what could a million agis
00:11:37
achieve solve fundamental questions of
00:11:40
science like dark energy invent new
00:11:43
technology that gives us Limitless
00:11:44
energy fix climate change cure aging and
00:11:48
cancer but then again sadly humans apply
00:11:52
their intelligence not just for the
00:11:53
benefit of all what if the agis are
00:11:56
tasked to guide drones or pull the
00:11:58
triggers in war or to engineer a virus
00:12:01
that only kills people with green eyes
00:12:04
or to create the most profitable social
00:12:06
media so addictive that people starve in
00:12:08
front of their screens the creation of
00:12:11
AGI could reasonably be as big of an
00:12:13
event as taming fire or electricity and
00:12:16
give whoever invents it equally as much
00:12:19
power but now let's go one step further
00:12:23
what if the potential of AGI doesn't
00:12:25
stop here intelligence explosion
00:12:29
intelligence and knowledge build and
00:12:31
accelerate each other but humans are
00:12:33
limited by biology and evolution once we
00:12:37
evolved the right Hardware our software
00:12:39
outpaced evolution by orders of
00:12:41
magnitudes and within a heartbeat we
00:12:43
ruled this planet but our software
00:12:46
basically hasn't changed much since then
00:12:48
which is why we have obesity and destroy
00:12:50
the climate for short-term gains since
00:12:53
AGI is software on a computer once it's
00:12:56
smart enough to do AI research the rate
00:12:58
of AI progress should speed up a lot and
00:13:01
that results in better AI That's better
00:13:04
at AI research without much human
00:13:06
involvement it may even be possible that
00:13:08
AI could learn how to directly improve
00:13:10
itself in which case some experts fear
00:13:12
this feedback loop could be incredibly
00:13:15
fast maybe just months or years after
00:13:18
the first self-improving hii is switched
00:13:20
on maybe it would actually take decades
00:13:23
we simply don't know this is all
00:13:25
speculative but such an intelligence
00:13:27
explosion might lead to a true
00:13:29
superintelligent entity we don't know
00:13:32
what such a being would look like what
00:13:33
its motives or goals would be what would
00:13:36
go on in its inner world we could be as
00:13:39
laughably stupid to superintelligence as
00:13:42
squirrels are to us unable to even
00:13:45
comprehend its way of
00:13:46
thinking this hypothetical scenario
00:13:49
keeps many people up at night humanity
00:13:52
is the only example we have of an animal
00:13:55
becoming smarter than all others and we
00:13:57
have not been kind to what we perceive
00:14:00
as less intelligent beings AGI might be
00:14:03
the last invention of humanity it's
00:14:06
possible that it could become the most
00:14:08
intelligent and therefore most powerful
00:14:10
being on earth a God in a box that could
00:14:13
exercise its power to bring unimaginable
00:14:16
wealth and happiness to humans while
00:14:18
securing our future or it could subvert
00:14:21
civilization and bring about our end
00:14:24
with Humanity unable to come up with a
00:14:26
way to stop it we'll look at some of
00:14:28
these potential Futures in more videos
00:14:30
but for now let's wrap up the only thing
00:14:33
we know for sure is that today right now
00:14:36
many of the largest and richest
00:14:38
companies in the world are racing to
00:14:40
create ever more powerful AIS whatever
00:14:43
our future is we are running towards
00:14:48
it who knows how long we have until we
00:14:51
must confront our AI future luckily you
00:14:54
still have plenty of time to prepare for
00:14:56
it if you're learning on Brilliance that
00:14:58
is
00:14:59
brilliant will make you a better thinker
00:15:01
and Problem Solver in just minutes a day
00:15:04
with thousands of bite-sized Hands-On
00:15:06
lessons on just about anything you may
00:15:08
be curious about including AI their
00:15:11
latest course how llms work takes you
00:15:14
under the hood of real language models
00:15:17
it demystifies Technologies like chat
00:15:19
gbt with interactive lessons on
00:15:21
everything from how models build
00:15:23
vocabulary to how they choose their next
00:15:25
word you'll learn how to tune llms to
00:15:27
produce output with exactly your desired
00:15:29
tonality whether it's poetry or a cover
00:15:32
letter and you'll understand why
00:15:34
training is really everything by
00:15:36
comparing models trained on Taylor Swift
00:15:38
lyrics and the legal speech of big Tech
00:15:40
terms and conditions it's an immersive
00:15:42
AI Workshop allowing you to experience
00:15:45
and harness the mechanics of today's
00:15:46
most advanced tool we've also partnered
00:15:49
with brilliant to create a series of
00:15:51
lessons to take your scientific
00:15:52
knowledge to the next level these
00:15:54
lessons let you further explore the
00:15:56
topics in our most popular videos from
00:15:58
rapes and mum metabolism to climate
00:16:00
science and supern noi each lesson on
00:16:03
brilliant is interactive like a
00:16:04
one-on-one version of a cazar video and
00:16:07
you can get started Whenever Wherever
00:16:10
right from whatever device you'd like to
00:16:12
get handson with ktz kazak lessons and
00:16:14
explore everything brilliant has to
00:16:15
offer from Ai and programming to maths
00:16:18
science and Beyond start your free
00:16:20
30-day trial by signing up at
00:16:22
brilliant.org
00:16:23
nutshell there's even an extra perk for
00:16:25
courts gazar viewers anyone signing up
00:16:27
through our link will get get 20% off an
00:16:30
annual membership once their trial ends
00:16:33
[Music]
