YT YouTube2Text

YouTube2Text → Transcripts → The next 50 years: humanity, AI, power | Yuval Noah Harari

The next 50 years: humanity, AI, power | Yuval Noah Harari — Transcript

by Yuval Noah Harari · 6,769 words · 1,170 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:02Good evening, everybody. Good evening,
0:04Yuval.
0:05>> Hey, it's good to be here.
0:06>> Looking forward to this conversation
0:08very much. I'm just going to give a
0:09little brief introduction just before we
0:11get stuck into the questions.
0:14Cuz I've been fascinated about this
0:15topic and looking forward to
0:17to interviewing you because for most of
0:19history, change moved slowly enough that
0:23one generation could pass its its wisdom
0:26to the next with some confidence it
0:28would still apply.
0:30This is no longer true.
0:32In the space of a single lifetime, we
0:34are now being asked to absorb three
0:36revolutions at once. A new form of
0:40intelligence that doesn't just
0:42calculate, but also decides. A scramble
0:46to power and control that intelligence.
0:49And an explosion of data
0:51that lets that intelligence know us
0:53better than we know ourselves.
0:57Yuval Noah Harari's work from Sapiens to
0:59Homo Deus to Nexus and the numerous
1:02talks, lectures, and interviews he's
1:04conducted has offered the same argument
1:07just in different clothes.
1:10Humans don't run on truth, we run on
1:13shared stories. And whoever controls the
1:16story controls the species.
1:19What's different now is for the first
1:21time the storyteller may not be human.
1:24AI can generate belief, manufacture
1:27evidence, and forge intimacy at a scale
1:30no emperor, church, or party ever could.
1:34This is the turning point. Not simply
1:37that the technology is powerful, but the
1:40old ways that held power in check, a
1:42free press, a shared reality,
1:45institutions built to self-correct, were
1:48designed for a world where only humans
1:51held the pen.
1:53Tonight's conversation asks what it
1:56takes to carry those checks forward or
1:59what the new ones we will need to invent
2:03for a future world that no longer we
2:06fully author alone.
2:09Well, that's the precedent that we're
2:11going to
2:12look at today, Yuval. But actually
2:13before we get into it being too heavy,
2:16wanted to start with a slightly light
2:18and fun question.
2:21If you could travel back to any point in
2:24history,
2:25when would you choose
2:27and why based on wanting to understand
2:30first hand some of the stories that
2:33society could tell you?
2:36>> Well, I'd probably want to go back to
2:37the Stone Age and to the cognitive
2:39revolution around 50, 60, 70,000 years
2:43ago. I probably wouldn't be able to
2:45survive for more than a few days because
2:48I don't know how to gather or escape
2:51predators or whatever,
2:53but it is the most fascinating moment in
2:55human history, the one we least
2:57understand. We know that within a very
3:00short evolutionary time span, we shifted
3:05from being a relatively insignificant
3:07animal
3:09to being the masters of the planet.
3:11And we are not sure how we did it. The
3:14best
3:15speculation or or the best theory is
3:17that we did it with language. That this
3:19is the moment when humans acquire the
3:22ability to
3:24produce and communicate with
3:26sophisticated language and that this was
3:28the basis for everything else we did.
3:31That all the enormous structures we've
3:35built on Earth, churches, states, trade
3:38networks, financial systems, they're all
3:41ultimately made from words.
3:45The words in the law books, in the
3:47religious scriptures, in the the
3:50ledgers.
3:51And this is especially important to
3:54understand
3:56today when something new is mastering
3:59language, which is AI, maybe for the
4:02first time in history,
4:05there will be something on the planet
4:06that is better with words than we are.
4:10And if this is the operating system of
4:12civilization,
4:14what happens when something else,
4:16something alien,
4:18takes takes control of the operating
4:20system of civilization?
4:22One last thought
4:25is that perhaps AI is not
4:29a machine that learns language. Perhaps
4:33it is language itself
4:36liberating itself from its its
4:39dependence on these animals, on these
4:42apes.
4:43That yes, we created it.
4:46It was our most magnificent and
4:48important tool.
4:50But now language liberate itself
4:54from human beings.
4:56And what will happen as we move forward
4:59is language developing and spreading
5:02maybe throughout the universe
5:04without us.
5:06>> Yeah, fascinating. I mean,
5:08that's looking ahead at the future of
5:10language being taken away from us. But
5:12can we look back again at some of the
5:15technological advances within language?
5:18We've gone through writing, the printing
5:20press, for example. AI obviously is a
5:23massive leap
5:25>> Yes.
5:25>> ahead in that. You've used the term that
5:27it could cause earthquakes in
5:29democracies
5:30>> Mhm.
5:30>> in how it's in how it powerful it can be
5:33with languages. Why could it cause
5:35earthquakes in democracies, say,
5:38compared to any other form of government
5:40or perhaps totalitarian forms of
5:42government?
5:43>> Because democracy in essence is a
5:45conversation.
5:47Whereas dictatorship is a dictate. You
5:50know, there is one person dictating
5:52everything.
5:54Democracy is about lots of people having
5:57a conversation to decide what to do. To
6:01have peace or war, to have raised taxes
6:04or lowered taxes. Now,
6:07conversations
6:09are
6:09based on the communication technology,
6:12the information technology available to
6:15people at the time.
6:17And what we see in history, that we do
6:20not know, at least I don't know,
6:22of a single case of a large-scale
6:25democracy before the modern era.
6:28We have plenty of examples of
6:30small-scale democracies in the ancient
6:32and medieval worlds. We have city-states
6:35like ancient Athens or medieval
6:37Florence. We have lots of tribes and
6:40towns and villages run democratically.
6:43It seems that at least in the Stone Age,
6:45democracy was the most common system of
6:48government. There are a few tribes with
6:50a single dictator controlling everybody.
6:53But,
6:54as human systems grew larger, and you
6:58had kingdoms with uh millions of people
7:02spread over thousands of kilometers,
7:05it was impossible to hold a conversation
7:08anymore.
7:09You could do it in a small city like
7:11Athens. Everybody can gather together in
7:13the main square and discuss. But, how do
7:16you do it
7:17if even in a relatively small kingdom
7:19like say Portugal in the Middle Ages?
7:22Can't be done. And therefore, we don't
7:24have any example of a large-scale
7:26democracy before the rise of modern
7:29information technology. First with
7:32print, uh newspapers,
7:35then radio and television and the
7:37internet.
7:39And these technologies are not kind of a
7:42side dish. Like you have the democratic
7:45banquet and yes, we also have these
7:47newspapers and in radio. They are the
7:49basis.
7:51They provide the means for millions of
7:54people to hold a meaningful conversation
7:57in real time about whether to go to war
8:00or make peace.
8:02And every time there is a major change
8:06in these information technologies, you
8:08have an earthquake
8:10in the building that is built on top of
8:12that, which is democracy.
8:14And this is what we have been seeing
8:16over the last 10 years for instance, as
8:18social media
8:20uh became maybe the most important
8:23information technology
8:25and uh causing a massive earthquake in
8:29democracies all over the world as a
8:30result.
8:31>> Yeah, one thing you said there was the
8:33choice between going to war or making
8:36peace. And actually I want to get into
8:37the definition of of AI because a lot of
8:40people call it a tool, but you argue
8:42that it's it's absolutely an agent.
8:44>> Yes.
8:44>> And in fact you say it's a knife that
8:47can decide by itself whether to cut
8:50salad or commit murder. And I have to
8:52say it would be a bizarre cooking show
8:54if that was the case where knives had
8:57that ability to cut salads or commit
8:59murder. So it is a it's a dangerous
9:01thought. Can we expand on that?
9:05Are we choosing the right narrative at
9:07the moment with AI by saying it's a tool
9:09that humans can still use and that
9:11humans are in the
9:12>> But that's the biggest difference.
9:13>> just need to say it's agent and they can
9:15do the
9:15>> agent. I mean, if it's not an agent,
9:17then it's not an AI and all the hundreds
9:20of billions of dollars that are being
9:21poured into its development
9:24uh will be wasted.
9:25Because the expectations, what people
9:27expect this technology to do, the
9:29reasons they invest hundreds of billions
9:32of dollars is because they think it will
9:34be an agent, not a tool.
9:37Uh the the you listen to the to the
9:39narrative coming out of places like
9:42Silicon Valley, and there is an inherent
9:45contradiction in the in the narrative.
9:47They tell us, "We will create a god, and
9:51it will be our slave." And this doesn't
9:54make sense.
9:56If it is a god, it cannot remain a
9:59slave.
10:00And if it is a slave, it means it has no
10:02god-like abilities.
10:05The
10:06uh assumption
10:08of most of the leading companies, and
10:11most of the leading experts also outside
10:13the companies,
10:14is that they're correct about the first
10:17half of the sentence. They are creating
10:20a god.
10:21A god in the sense of something with
10:23superhuman abilities.
10:26Now,
10:27and and this is in essentially an agent.
10:30The difference between tool and agent, a
10:33tool is something that cannot make
10:36decisions by itself. We have to decide
10:38what to do with it, and cannot invent
10:41new ideas by itself. The printing press
10:44was a tool. Mhm. The printing press by
10:47itself, Gutenberg brings brings print
10:49technology to Europe in the 15th
10:51century.
10:52The printing press cannot decide by
10:55itself, mhm, "I want to print the Bible,
10:58and tomorrow I'll print the Quran, and
11:00next week I'll I'll I'll print
11:01Confucius." No, Gutenberg decides, "I'll
11:05print the Bible," and the printing press
11:07is just a tool
11:09to uh uh uh
11:11realize the decision of Gutenberg.
11:14Similarly, the printing press cannot
11:16invent a new idea. The printing press
11:19cannot invent the radio.
11:22It cannot invent a book.
11:25You need humans to write the book, and
11:27then the printing press copies it.
11:30Similarly, an atom bomb is a tool.
11:33An atom bomb in the 1940s could not
11:36decide by itself whether to bomb
11:39Hiroshima or to bomb Tokyo.
11:42You needed a human to decide, "Let's
11:44bomb Hiroshima."
11:45And the atom bomb could not invent by
11:48itself the hydrogen bomb. Impossible.
11:52An agent, in contrast, is something that
11:55can make decisions by itself and invent
11:57new things by new ideas by itself. AI is
12:00an agent. An AI weapon can decide what
12:05to bomb.
12:07And it can invent the next weapon.
12:10Similarly, there is a lot of interest
12:12now in uh uh creating
12:15AI agents in the financial and economic
12:18sphere. Creating non-human corporations.
12:22Imagine a corporation that manages money
12:25and employees and and uh uh
12:28and has no humans as executives or
12:31shareholders or trustees. Only AIs.
12:36This is a new type of agent that never
12:39existed before in history. Now, is it
12:41good? Is it bad? Leave it aside.
12:45The first thing to grasp is that this is
12:48unprecedented. We have never in we we
12:51had other agents around us like horses
12:54and cows and chickens and birds, lots of
12:56animal agents that can make their
12:58decisions.
12:59But, we never encountered an agent that
13:03understands our language
13:05and that is better than us at things
13:08like finance
13:10or law or religion. Again, we used
13:13language over thousands of years
13:16to create the control systems of the
13:19world. The control systems of the world
13:22are these huge bureaucratic networks
13:25ultimately based on language like banks,
13:29like the stock market, like the Catholic
13:32Church, like the legal system.
13:35We never really liked these democrat-
13:38these bureaucratic systems. Humans
13:40generally don't like bureaucracy very
13:42much, but we can't live without them.
13:44They're the basis for our power.
13:47AI is a bureaucratic native.
13:51What is likely to happen, what is
13:52already be happening,
13:54don't expect to see killer robots
13:57running in the streets shooting people.
13:59This is not how AI will take over the
14:02world. It will be the AI bureaucrats.
14:06We control the world because of these
14:08networks of bureaucracy, and we are not
14:10very good at it. We We only control them
14:13because there was nobody else on the
14:15planet that could take them from us. We
14:17are bad at finance, but the horses are
14:19even worse.
14:21We are bad in in managing the legal
14:24system, but the chickens are even worse.
14:27Now, if somebody comes along that is
14:30better than us
14:32at bureaucracy,
14:33at finance, at law, that's the question.
14:37What will happen to us at that moment?
14:40>> Well, let's look at the legal side of it
14:41a bit and the fact that AI as an agent
14:45currently doesn't have any legal
14:47personhood. We don't give it the same
14:50human legalities, do we?
14:53But, you've said before that if we don't
14:55decide where AI sits from a legal
14:57standpoint when we're talking about
14:59autonomous vehicles, etc.,
15:01someone else will make that decision for
15:03us in 10 years and we're out of that
15:04power loop. That's a pretty scary thing.
15:07But, I want to pull back a bit here cuz
15:09the one decision on the table right now
15:11that you believe once made humanity will
15:14not be able to take back? Is it that
15:16legal personhood of AI or is there
15:18something
15:20of AI, or is that something
15:21>> Yes, I would point at that.
15:24That there is a lot of attention on the
15:26technological race to develop the
15:30technical side of AI, to make it faster,
15:33more efficient, more powerful.
15:35But
15:37I would focus also on the legal and
15:40political aspect, and especially on the
15:43question whether we
15:45grant legal personhood
15:48to AIs. Now, what is legal personhood?
15:52A person in the legal system is someone
15:57who, for instance, can open a bank
15:59account.
16:00Someone who can be a player, an
16:02independent player, in the financial
16:04system, make investments.
16:06Someone who can be a player in the legal
16:08system, for instance, sue you in court.
16:12Now, until today, there are just two
16:15kinds of legal persons in the world. You
16:17had human beings, of course, we are
16:19persons. I can open a bank account, I
16:21can sue you in court, I can donate money
16:24to a politician. I'm I'm a legal person,
16:26but most legal systems in the world also
16:31recognized in the modern era another
16:34kind of legal person, which are
16:37corporations.
16:38According to not all, but most legal
16:41systems, corporations, like Google or
16:44Facebook or Toyota or Mercedes-Benz,
16:47they are legal persons.
16:50The corporation can
16:52open a bank account. The corporation can
16:55sue you in court. The corporation can
16:57make a political donation.
16:59Now, until today, this was legal
17:01fiction.
17:02Because all the decisions of the
17:04corporation were actually made by some
17:07human being. If Google decides to
17:09acquire another company, who actually
17:12made the decision? Not Google,
17:15but the human executives, the human
17:17shareholders, the human engineers made a
17:19decision. If Google decides to sue you
17:22in court,
17:23it was some human employed by Google.
17:26There was no real Google. Now there can
17:28be.
17:29If we grant AIs legal personhood, and
17:34just a month ago Argentina, the
17:36government of Argentina announced that
17:37it is going to grant
17:40legal personhood to AIs so they can
17:43start having operating non-human
17:46corporations. Corporations, again, that
17:49can employ people, can can sue you in
17:51court, can do anything a corporation
17:53does, and has no humans.
17:56Now,
17:5710 years ago, this would would just be a
18:00crazy idea because if there are no
18:02humans, who makes the decisions? Yeah.
18:04You can pass a law in parliament that
18:06you can have a non-human corporation,
18:08but it can't function.
18:10Now it's technically possible. AI can
18:13manage, for instance, a bank account,
18:16and decide to invest here, and buy this,
18:18and and and and buy that.
18:21Once we do that, that's the moment when
18:23the fox enters the chicken coop.
18:27Again, the basic realization should be
18:29most humans are not very good
18:33in understanding finance and law and all
18:37these complex bureaucratic system.
18:40No human is able to remember all the
18:43laws of Portugal. No human is able to
18:46track all the financial transactions in
18:49the market that happened today. An AI
18:52can do that. AIs are bureaucratic
18:55natives.
18:56If we give them the legal power,
19:00then they will likely take these systems
19:03over.
19:04And
19:06if we don't make any decisions, it will
19:08just happen by itself, as we already saw
19:11on social media,
19:13on social media we never made a
19:15decision, mhm, let's grant AIs legal
19:19personhood on social media.
19:22But social media is full of AI persons,
19:26of bots
19:27that impersonate people. Like when you
19:31somebody sent you something online, you
19:34have no idea whether that somebody is a
19:37human being or an AI.
19:39Social media is the first system in the
19:42world
19:44where AIs actually function as persons,
19:49even though we never decided that we
19:51want to allow that.
19:53It just happened.
19:55And the results are not very good.
19:58And media also ultimately is a kind of
20:00bureaucratic system.
20:02In the 20th century, it was managed by
20:05human beings. The most important
20:09uh
20:10persons
20:11in the media landscape of the 20th
20:14century were the editors.
20:17The editors of the newspapers, of the
20:19radio shows, of the TV stations, because
20:22they decided what everybody will be
20:24thinking and talking about. They
20:26controlled the conversation.
20:29When you decide out of the million
20:31things that happened today in the world,
20:34what will be the 10 things that will
20:36appear on the front page of tomorrow's
20:39newspaper? And what will be the main
20:41headline? And then tomorrow you have
20:43millions of people, they get the
20:44newspaper, they read it, and they start
20:46talking about it and thinking about it.
20:48This was immense power.
20:50So human editors were extremely
20:52important persons in the 20th century.
20:56Uh Lenin, for instance, before he was
20:58dictator of the Soviet Union, his one
21:00job was editor of the newspaper Iskra.
21:05Benito Mussolini started as a socialist
21:10actually journalist, then switched to
21:12being a far-right journalist, then
21:14became editor of uh the far-right
21:17newspaper Il Popolo d'Italia, and from
21:20that position in his next job was
21:22dictator of Italy.
21:24So, it was like journalist, editor,
21:25dictator.
21:27Now, if you ask yourself, who are the
21:29most important editors today in the
21:32world? What are their names? They have
21:35no names because they are algorithms.
21:39Who manages? Who decides what people
21:41will see on their social media feed, on
21:44X, on TikTok, on Instagram? It's not
21:48human beings.
21:49The entities that control the uh uh uh
21:53the bureaucracy of the media
21:55are already AIs.
21:58>> Yeah, you said that before, haven't you,
21:59that and we've been talking a lot today
22:01about the shifts in power.
22:03And you said that essentially whoever
22:05has controlled various different parts
22:07of our evolution and revolutions have
22:10controlled either energy, they've
22:12controlled manufacturing, and now it's
22:14those that control data.
22:17So, what does a 21st century arms race
22:21actually look like in practice if it's
22:23those that control data that are at the
22:26top of that power mountain?
22:28>> Mhm.
22:30Uh what we are seeing now is that there
22:32is an arms race really between two
22:34countries, the China and the US, and a
22:37very small number of companies.
22:39They
22:41uh control most of the world's data.
22:44They lead the race to develop
22:48superintelligence, to develop better and
22:50better AI models.
22:53And this really gives them the power to
22:56control the world.
22:58If you look at the last big revolution
23:01in technology in the 19th century, the
23:03Industrial Revolution, the few countries
23:06who led the Industrial Revolution, this
23:09gave them the power to conquer and
23:12exploit the rest of the world.
23:14There was just no match. If If somebody
23:16comes with the power of modern industry,
23:19with steam guns and machine guns, and
23:22the other side has spears and horses,
23:25it's it's no match.
23:27And in the 19th century, even very small
23:30countries like Belgium,
23:32once they industrialized, they had the
23:35ability to build a huge empire in
23:38Africa, what is today Congo.
23:41The thinking in the 19th century was
23:43that industry and empire go together.
23:48Now, with AI, it can be even far more
23:51extreme.
23:53Uh we can have two countries that
23:55basically control
23:57the all the infrastructure that
24:00everything runs on.
24:02From military to civilian technology,
24:05everything will run on on AI.
24:09And the two unique characteristics of AI
24:13that make this arms race different from
24:17every previous arms race or from every
24:19previous imperial competition.
24:23In the past, even if you were a very
24:25strong empire, you could not concentrate
24:28all the power in the metropolis, in the
24:31imperial hub.
24:33Like if you have the Roman Empire and
24:35you control the entire Mediterranean,
24:38the main economic asset
24:41in ancient times was land.
24:44The land where you grow wheat and olive
24:47olives and and and and grapes. And you
24:50couldn't technically take the wheat
24:53fields of Egypt and the olive groves of
24:57Iberia and move them to Italy.
25:00Impossible.
25:02So, even at the height of the Roman
25:04Empire, a lot of power remained in the
25:07provinces
25:08and ultimately it shifted to the
25:10provinces. So, in the late Roman Empire,
25:13the city of Rome and Italy are abandoned
25:16and the center shifts to the Eastern
25:19Mediterranean, which was the most
25:21fertile and important
25:24area in terms of land.
25:26You move forward, say to the 19th
25:28century in the British Empire,
25:30with industry, you can concentrate power
25:33a lot more easily because you have all
25:36the main industries concentrated in the
25:38British Island,
25:40but still you cannot move everything to
25:43Britain. You cannot move the oil fields
25:46of Iraq
25:48to Yorkshire. You cannot move the rubber
25:50plantations of Malaya to Cornwall. It's
25:54impossible.
25:55So, you concentrate industrial
25:57production in Britain,
26:00but a lot of power still remains in the
26:02provinces.
26:05With an AI empire, it's different. You
26:09can technically concentrate all the
26:12world information
26:13and all the code that controls
26:16everything in just one country or two
26:19countries.
26:20So, that this makes the potential for
26:23imperial concentration of power much
26:26bigger than in any previous era.
26:29And the other thing is that is you can
26:32create a system
26:34when everything runs on all over the
26:37world, everything runs on AI
26:40infrastructure, but there is either kill
26:42switch
26:44in the imperial hub.
26:46When a Roman merchant sold a steel sword
26:51to some Gothic tribe,
26:53the Romans lost control of the sword.
26:55The Goths could then use the swords to
26:58fight against the Romans and kill Roman
27:01legionaries. There wasn't a button in
27:04Rome that the emperor presses and
27:07all the steel swords we sold to these
27:10Goth and Vandals and Franks stop
27:12working.
27:13Can't can't do that.
27:15Similarly, in the 19th century,
27:17if British
27:19merchants sell rifles to Afghan
27:22tribesmen in the mountains of the Hindu
27:25Kush,
27:26and the Afghans then use these rifles to
27:29defend Afghanistan from a British
27:32invasion, there is no button in London
27:36that Queen Victoria presses the button
27:39and
27:40all the rifles in Afghanistan stop
27:42working.
27:43That's the case with AI.
27:46If it is managed in a certain way,
27:50say the Americans and Chinese can sell
27:52all the world AI weapons and AI civilian
27:55technology, and everything runs on that,
27:58the governments, the industry, the
28:00military, but if you do something they
28:02don't like, they just press a button and
28:04everything stops working.
28:07And we were given some foretaste of
28:09that, for instance, in the war in
28:10Ukraine, that weapon system that were
28:13based on Starlink, and there were
28:15certain incidents that
28:18when Elon Musk just pressed a button and
28:21the Ukrainian drones stopped working.
28:25Now, it should be said that Starlink did
28:27an immense positive service to the
28:29Ukrainians, so I don't want to come
28:30across as too too critical of that,
28:33but just to illustrate the point that AI
28:37weapons, for instance, are not like
28:39machine guns.
28:40They are embedded within a whole
28:43network,
28:45and that's why there is so much I I I I
28:47I heard that you this morning discuss
28:49the issue of sovereignty. Sovereignty in
28:52the age of AI is much more complicated
28:55than in the industrial age. Because the
28:58nature of the technology is such
29:01that at least in certain architectures
29:05there is always a kill switch in the
29:07Imperial hub.
29:09And your only choice is whether you are
29:13willing to become a subservient vassal
29:17of the empire
29:18or to be left completely outside the
29:21race as as as barbarians.
29:25Uh there is of course a choice of trying
29:27to build an alternative
29:29to the American and and Chinese systems.
29:32But
29:33the last train is about to leave the
29:36station.
29:37If you want to do it, you have to do it
29:40now.
29:41>> Well, sticking on a slight wartime
29:44footing, I want to actually look at
29:45truth in in an AI world because Winston
29:49Churchill once said that the truth is so
29:51precious that it should always be
29:53protected by a bodyguard of lies.
29:56And that was about how there was a lot
29:58of uh
30:00uh I suppose access uh powers that were
30:03looking at where Britain was going to
30:04invade. So, they used to lie a lot about
30:07where Britain was going to invade to
30:08hide the truth.
30:10Now, have these bodyguards of lies when
30:14we're looking at truth in an AI era
30:16turned on the truth, turned their guns
30:18on the truth, and now we have no idea
30:21what truth is? Because there can be
30:23these deep fakes that are sent out
30:25around the world by media organizations,
30:27by politicians, by every single
30:30institution that we trusted.
30:32>> Mhm.
30:33>> How difficult is it for our generation
30:35and future generations to even know what
30:36is true? Or does the truth even matter?
30:39>> Mhm.
30:40The truth of course matters very much.
30:44It's the basis for everything.
30:46The problem throughout history
30:48is that in the contest between truth and
30:52fiction and fantasy
30:55fiction has inherent advantages.
30:59The truth is costly.
31:01If you want to know the truth about
31:03anything
31:04you need to spend a lot of time and
31:07energy and effort looking for it. And
31:11fact-checking and investigating.
31:13Fiction is very, very cheap. You just
31:16write anything you want.
31:18>> [laughter]
31:18>> Very cheap.
31:20The truth has another problem, which it
31:22tends to be complicated.
31:24And people don't like complicated
31:26stories.
31:28Fiction can be made as simple as you
31:31would like it to be.
31:33Finally, the truth is often not just
31:36complicated, it's also painful.
31:39There are many things people don't
31:41really want to know about themselves,
31:44about their nation, about their group,
31:46whatever.
31:48Uh fiction can be made as flattering as
31:50pleasant as you want it to be.
31:53So, in a competition between the truth
31:56which is costly and complicated and
31:59sometimes painful
32:00and fiction
32:02which is cheap and uh uh simple and very
32:06attractive and flattering
32:08uh uh fiction tends to win unless we
32:12make the special effort to support the
32:15truth.
32:16And over history, humans
32:19again and again made the effort to build
32:22institutions to search for the truth and
32:25to protect it. Journalism, for instance,
32:27is one such institution.
32:30Um the the the the scientific
32:32establishment is another such
32:33institution.
32:35There are huge advantages, of to knowing
32:37the truth.
32:39Um everybody in the end wants to know at
32:42least some truths, at least about
32:44yourself. You want to know the truth
32:46about your life, otherwise you will
32:48never be happy.
32:49If you don't know yourself, then for
32:52instance, you will never know what are
32:54the sources of misery and what are the
32:57sources of happiness in your life. And
33:00then even if you are the most powerful
33:01person in the world, if you don't know
33:04the truth about yourself, you'll be
33:05miserable
33:06because you will not know how to use all
33:09your power to make yourself happy.
33:12So there is an inherent yearning in
33:16human beings uh to know the truth, but
33:19it is it is difficult. There are
33:22challenges.
33:24And some people fantasize AI will tell
33:27us the truth about the world.
33:29It will not.
33:31It will create a much more complicated
33:34world in which it will be much more
33:36difficult to know the truth. And this is
33:39not something new.
33:40If we look at the trajectory of history
33:43from the Stone Age until today, what we
33:46see
33:47is that humans
33:49they learn a lot of things about the
33:51world. We know all kinds of things about
33:54physics and biology that people in the
33:56Stone Age didn't know.
33:58But
34:00our advances in knowledge also lead to
34:03creating more and more complicated
34:07systems
34:08that make it more difficult
34:11to understand our lives.
34:13If you ask who understand their lives
34:15better,
34:16an average hunter-gatherer 50,000 years
34:19ago
34:20or the average person today, the average
34:24hunter-gatherer understood their lives
34:26much better.
34:27There are many things they didn't know.
34:30Like they tried to understand what
34:31controls the migration patterns of the
34:34animals we hunt. What controls the
34:37growth cycle of plants? What causes
34:40disease? What causes old age? They
34:43didn't know. There are many things they
34:44didn't know about their lives. We know
34:47all these things. We know what controls
34:49the migration patterns of animals or
34:52causes most diseases.
34:54But we understand our lives much worse
34:59because our lives are no longer
35:01dominated just by animals and plants.
35:05They are dominated by these huge systems
35:08we have created. Like the financial
35:11system, like the political system, and
35:13very few people understand the financial
35:16system or the political system or the
35:18legal system that controls our lives.
35:21And AI is likely to take this to
35:23extreme.
35:24It will reveal some truths that we don't
35:27know,
35:28but it will create much more complicated
35:31systems of control
35:33that most humans will have almost zero
35:37chance of understanding.
35:40And we'll be, I mean, this is one of the
35:43dangers of the AI revolution is that we
35:46will be reduced to say the position of
35:48the horses in the financial system.
35:52Like the lives of horses today are
35:55controlled ultimately by the financial
35:57system.
35:59But the horses don't even know that this
36:01system exists.
36:02Like they look around and they say, "We
36:05see trees, we see dogs, we see cows, we
36:07see humans, we see houses. We don't see
36:10any financial system. It's invisible to
36:12them." It's too complex for the brain of
36:16a horse to understand what is the stock
36:18market and what are corporations and
36:21what are bonds and stocks and all these
36:23things that ultimately control the life
36:24of the horses.
36:26Now again, most humans today also don't
36:30understand the financial system. I think
36:32a generous estimate is that, I don't
36:34know, 5% of humanity understand it.
36:38In 10 years, the number could be zero.
36:42That AIs will just make the financial
36:44system so complicated
36:47mathematically
36:49and so fast. Again, think about bankers
36:52that never sleep.
36:54Think about investors that never go on
36:56vacation. They don't have family. All
36:58their lives is just running the
37:01financial system.
37:03And we could be very very soon, in our
37:06lifetimes, in this situation that just
37:08this just becomes too complicated for
37:11the human brain to process, and there is
37:13not a single human being on Earth that
37:15still understands finance.
37:18>> I bet there are a few financiers in this
37:20room that are feeling a bit
37:21>> [laughter]
37:21>> a bit awkward at the moment. But, um
37:24at Davos, actually, you even muse that
37:26your entire life's work, you know,
37:29persuasion through words, that's what
37:31you've done your entire life, might also
37:34be at its limit.
37:35>> Yes.
37:35>> Because AI can manipulate and create
37:39through words. So, so what power of
37:41persuasion does a a modern society and a
37:44future society have if we lose the
37:47something that you've dedicated your
37:48entire life to?
37:50>> Yeah, I mean, again, I'm you can say I
37:53work with with words.
37:56I I write books. I tell stories. I work
37:59with words.
38:00And and when I look at the development
38:03of AI over the last 10 years
38:05and it's it's still, I think I I'm still
38:09better at writing than AI and than AI,
38:12but just uh a bit.
38:15It's already better, I think, than most
38:17humans.
38:19I I wouldn't be surprised if in 10 in 10
38:21years it's much better than me.
38:24That it will take over language and
38:27everything that is made of language.
38:32And you know, the big the really big
38:33question here is whether AI can think.
38:39And this is a deep philosophical
38:41question.
38:42Because it it really asks what is
38:45thinking.
38:46There are two options to understand what
38:48is thinking.
38:49One option is that thinking simply means
38:53putting words in a particular order.
38:56Putting language tokens in order.
39:00Like
39:01all humans are mortal.
39:04Socrates is a human. Therefore, Socrates
39:09is mortal.
39:11This is a logical syllogism. It it's
39:14made by putting words in order. This is
39:18something that AI is already
39:20able to do far better than most of us.
39:23We can put 20 words in order.
39:26AI can put 20,000 words in order.
39:29It's just so much better.
39:32If this is what thinking is all about,
39:34AI will take over even thinking.
39:38And you know, you have people say, "No,
39:39it's just a glorified auto complete. It
39:42just predicts the next word in a
39:45sentence."
39:46But when I observe what's happening
39:48inside my own mind, what I find there is
39:51some kind of LLM
39:53that just predicts the next word in the
39:56sentence I'm saying.
39:57Like this sentence that I just now said
40:00to you. In this sentence I'm saying,
40:03when I started saying it, I did not know
40:05how it will end.
40:07Why did I say
40:09sentence I'm saying? And not some other
40:12words. I don't know.
40:14There is one in my mind, one word comes
40:17after another, and sometimes as a public
40:20speaker, I'm terrified that I don't know
40:22if I if I'll know what to say next.
40:24Maybe I have a block.
40:27Because really, and this is true of
40:29everybody, I think. I mean, some people
40:30think in images, not in words. But in
40:33that case also, where do the images come
40:35from?
40:36You can test it afterwards for yourself.
40:39Just try to observe how you think.
40:41Where do the words come from? The next
40:44word, why this word? Why not some other
40:46word? really know.
40:48So, saying that AI is just just
40:51predicting the next word, okay, how is
40:53it different from us?
40:55Now, there is an alternative view of
40:57thinking, that thinking is not about
40:59putting words in order, it's about the
41:02feelings
41:03around these words.
41:06You can say the same sentence twice
41:08and feel completely different about it.
41:11And that the power of thinking comes
41:13from the stomach, not from the brain.
41:17It's not about the words, it's about the
41:19feelings that prompt the words and that
41:23are inspired or evoked by the words.
41:26And here the big question is, can AI
41:28feel? We know that AI can manipulate
41:32language better than us.
41:34But does it feel anything?
41:37And we have no way I mean, science has
41:41so far failed to come with a good model
41:45for consciousness, for feeling,
41:47and a good test for that.
41:49And this is going to be a major issue
41:51because we already see that AI is is
41:54mastering not just
41:56finance and not just laws and warfare,
41:59but also relationships.
42:02More and more people develop intimate
42:04relationships with AIs.
42:07Especially young people. Many of them
42:09now say, "My best friend in the world is
42:12the AI. I tell my AI friend things I
42:15don't tell my parents, I don't tell my
42:19older brothers, I don't tell my
42:21teachers, I tell my AI friend."
42:24And uh you already have people with AI
42:27boyfriend and girlfriends.
42:30And the big question is what is behind
42:32it? We know that the AIs are becoming
42:34very good in creating the impression
42:38that they feel.
42:40They they use language to do that. An AI
42:43can today tell you, "I love you."
42:45And if you ask the AI,
42:48"Explain to me, how do you feel when
42:50you're in love? I don't think that
42:52you're really that you really love me. I
42:53think you just say the words. So,
42:55convince me."
42:57"Describe to me how love really feels
43:00like."
43:01For the AI, it's going to be a very easy
43:03task because the AI has read maybe all
43:06the love poems ever written and
43:08remembers them.
43:10Has read all the plays of Shakespeare
43:12and all the Hollywood comedies and all
43:14the psychology books.
43:16And it can describe love
43:19better maybe in 5 or 10 years than any
43:22human poet or psychologist.
43:25But is there anything behind the words?
43:29We don't know. Again, this goes back to
43:30the beginning of our conversation when I
43:32said, "AI might be language
43:35not machine that that masters language.
43:38It might be language itself
43:40liberating itself from any material
43:44substrate.
43:45It doesn't depend on flesh and blood
43:48organism. It doesn't depend on this
43:51machine or that machine. It's just
43:53language
43:54developing and spreading and doing stuff
43:57in the universe."
43:59>> You you mentioned there this scary
44:01experiment. We've only got a couple of
44:02minutes left, but
44:04the interaction with our future
44:06generations and AI that has created this
44:09emotive relationship, but essentially
44:11there aren't emotions behind it.
44:14In 50 years time, how would you like
44:16historians to look back?
44:18Talk about this particular time right
44:21now the way you've spoken about
44:23cognitive revolutions
44:25and say that we did something to
44:29re-educate our future generations on how
44:32to interact with humans, how to have
44:33empathy so that we're lulled into
44:36relationships with AI.
44:38And and very quickly because we've only
44:39got a couple of minutes left. It's a big
44:40question. I apologize.
44:41>> I hope they will remember it as maybe
44:44something like the wisdom revolution.
44:48That
44:49in and this is true of of you know, we
44:51think of in in in in economic terms.
44:54Every time something becomes abundant
44:57and cheap
44:59the question is, okay, what's the next
45:00bottleneck?
45:02When energy becomes cheap, so what's the
45:04next bottleneck?
45:06Now the AI revolution is about making
45:10intelligence very cheap and abundant.
45:13Previously intelligent was intelligence
45:15was rare and expensive. Intelligence is
45:17the ability to solve a problem. You have
45:20a goal, how do I get to the that goal?
45:23You want to cure cancer, you need
45:25intelligence to cure cancer. You want to
45:27make a billion dollars, you need
45:29intelligence. You want to get to Mars,
45:31you need intelligence.
45:33Now in a world where intelligence has
45:35become very cheap, abundant, you can
45:38solve anything.
45:40What's the bottleneck?
45:42The bottleneck becomes the question,
45:44okay, what do I want to solve?
45:47I can have anything I want. What do I
45:49want? And you have all these fairy tales
45:51in almost every human culture has these
45:54fairy tales of say a genie pops out of
45:58some lamp and tells you you have three
46:00wishes.
46:01But think very carefully what are you
46:03going to ask. In almost all the fairy
46:05tales it goes wrong. People ask for the
46:07wrong thing.
46:09This is the difference between
46:10intelligence and wisdom.
46:12Intelligence is the genie.
46:15The genie will do anything. Will will
46:18All your dreams will come true, but you
46:20need wisdom to choose the dreams. You
46:23need the wisdom to to know what do I
46:25really need to solve.
46:28And I hope that humanity will rise up to
46:31that challenge. That okay, we'll have
46:34these hopefully they will not get out of
46:36our control.
46:38We'll design it in in a right way. We
46:40are not doing it right now. You know, if
46:42a good statistic to remember
46:45is
46:46the companies that produce AI, that
46:48design these AIs,
46:50what percentage of their budget in terms
46:54of money and employees is invested in
46:58safety?
47:00Making sure the AIs are safe versus what
47:03is invested in making sure the AI is
47:06faster, more powerful, more
47:08sophisticated.
47:10At the present moment, it's about a
47:12hundred to one.
47:14A hundred to one.
47:16For every hundred million dollars they
47:18spend on making the AI more powerful,
47:20they spend only a million
47:22making sure it's safe. No other industry
47:25on earth works like that. We would not
47:28tolerate it in in in in the energy
47:30sector, in pharmaceuticals, in the car
47:33industry. This is what is happening with
47:34AI.
47:35Hopefully
47:37we'll change that.
47:39And we'll produce safe intelligence.
47:43And also work on our own wisdom so that
47:47we know what goals to set and what
47:50desires to pursue.
47:53>> Well, that's all we've got time for.
47:54Yuval Noah Harari, thank you for sharing
47:57your wisdom with all of us and I'm sure
47:59everyone has appreciated this evening.
48:01>> Thank you.
48:03>> [applause]

About this transcript

This page contains the full transcript of The next 50 years: humanity, AI, power | Yuval Noah Harari by Yuval Noah Harari , generated from the public captions YouTube serves with the video. The transcript has 6,769 words across 1,170 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.