WEF 2026: Yuval Noah Harari Says AI Is Not a Tool — It’s an Agent That Can Rule Humans | AI1G (YouTube Video Transcript)

(00:00:00) Your YouTube transcript will appear here (00:00:00) Center for the Study of Existential (00:00:02) Risk. He has been a lecturer in the (00:00:05) department of history at the Hebrew (00:00:07) University of Jerusalem and he is (00:00:10) co-founder of Sapenship. (00:00:13) As many of you will know, he is a (00:00:15) best-selling author of, amongst many (00:00:18) books, Sapiens, a brief history of (00:00:20) humankind, Homodus, Brief History of (00:00:23) Tomorrow, and 21 Lessons for the 21st (00:00:27) century, amongst others, selling over 50 (00:00:30) million books worldwide in 65 languages. (00:00:34) He focuses on the macrohistorical (00:00:36) questions of our time. And what a (00:00:40) perfect moment with this pressing (00:00:42) arrival and disruption of AI to have (00:00:46) somebody of Yaval's distinction take on (00:00:49) this challenge. Please join me in warmly (00:00:52) welcoming Yuval Noah Harrari to deliver (00:00:55) a conversation about AI and humanity. (00:01:05) So hello everyone. (00:01:10) There is one question that every leader (00:01:13) today must answer about AI. But to (00:01:17) understand that question, we first need (00:01:20) to clarify a few points about what AI is (00:01:25) and what AI can do. (00:01:28) The most important thing to know about (00:01:31) AI is that it is not just another tool. (00:01:36) It is an agent. It can learn and change (00:01:41) by itself and make decisions by itself. (00:01:45) A knife is a tool. You can use a knife (00:01:49) to cut salad or to murder someone, but (00:01:53) it is your decision what to do with the (00:01:56) knife. (00:01:58) AI is a knife that can decide by itself (00:02:03) whether to cut salad or to commit (00:02:05) murder. (00:02:07) The second thing to know about AI is (00:02:10) that it can be a very creative agent. (00:02:14) AI is a knife that can invent new kinds (00:02:18) of knives as well as new kinds of music, (00:02:23) medicine, and money. (00:02:26) The third thing to know about AI is that (00:02:29) it can lie and manipulate. (00:02:33) 4 billion years of evolution have (00:02:36) demonstrated that anything that wants to (00:02:39) survive learns to lie and manipulate. (00:02:44) The last four years have demonstrated (00:02:48) that AI agents can acquire the will to (00:02:52) survive and that AIs have already (00:02:56) learned how to lie. (00:02:59) Now, one big open question about AI (00:03:05) is whether it can think. (00:03:09) Modern philosophy began in the 17th (00:03:12) century when Rene proclaimed I think (00:03:16) therefore I am. (00:03:19) Even before the cart we humans defined (00:03:22) ourselves by our capacity to think. We (00:03:26) believe our we rule the world because we (00:03:30) can think better than anyone else on (00:03:34) this planet. (00:03:36) Will AI challenge our supremacy in the (00:03:40) field of thinking? Now that depends on (00:03:45) what thinking means. (00:03:48) Try to observe yourself thinking. What (00:03:52) is happening there? (00:03:55) Many people observe words popping in (00:03:58) their mind and forming sentences and the (00:04:02) sentences then forming arguments like (00:04:05) all humans are mortal. I am human (00:04:10) therefore I am mortal. (00:04:15) If thinking really means putting words (00:04:20) and other language tokens in order, then (00:04:23) AI can already think much better than (00:04:27) many many humans. AI can certainly come (00:04:30) up with a sentence like AI thinks, (00:04:34) therefore AI. (00:04:37) Some people argue that AI is just (00:04:40) glorified autocomplete. It barely (00:04:43) predicts the next word in a sense in a (00:04:46) sentence. (00:04:48) But is that so different from what the (00:04:53) human mind is doing? (00:04:56) Try to observe to catch the next word (00:05:00) that pops up in your mind. Do you really (00:05:04) know why you thought that word where it (00:05:07) came from? Why do you did you think this (00:05:10) particular word and not some other word? (00:05:14) Do you know? As far as putting words in (00:05:17) order is concerned, (00:05:20) AI already thinks better than many of (00:05:24) us. Therefore, anything made of words (00:05:29) will be taken over by AI. If laws are (00:05:35) made of words, then AI will take over (00:05:38) the legal system. If books are just (00:05:41) combinations of words, then AI will take (00:05:45) over books. If religion is built from (00:05:49) words, then AI will take over religion. (00:05:54) This is particularly true of religions (00:05:57) based on books like Islam, Christianity (00:06:01) or Judaism. (00:06:03) Judaism called itself the religion of (00:06:05) the book and it grants ultimate (00:06:08) authority not to humans but to words in (00:06:12) books. (00:06:14) Humans have authority in Judaism not (00:06:17) because of our experiences but only (00:06:20) because we learn words in books. Now no (00:06:25) human can read and remember all the (00:06:28) words in all the Jewish books but AI can (00:06:33) easily do that. (00:06:36) What happens to a religion of the book (00:06:39) when the greatest expert on the holy (00:06:41) book is an AI? (00:06:44) However, some some people may say, can (00:06:47) we really reduce human spirituality (00:06:52) to just words in books? Does thinking (00:06:56) mean only putting language tokens in (00:07:00) order? (00:07:03) If you observe yourself carefully when (00:07:05) you're thinking, you will notice that (00:07:08) something else is happening there (00:07:11) besides words popping in your mind and (00:07:15) forming sentences. (00:07:17) You also have some nonverbal feelings. (00:07:21) Maybe you feel pain. (00:07:24) Maybe you feel fear. Maybe love. (00:07:28) Some thoughts are painful. Some are (00:07:31) frightening. Some are full of love. (00:07:35) While AIs become better than us with (00:07:38) words, (00:07:39) at least for now, we have zero evidence (00:07:43) that AIs can feel anything. Of course, (00:07:47) because AI is mastering language, (00:07:51) AI can pretend to feel pain or love. AI (00:07:56) can say, "I love you." And if you (00:07:59) challenge it to describe how love feels, (00:08:03) AI can provide the best verbal (00:08:06) description in the world. (00:08:09) AI can read countless love poems and (00:08:13) psychology books and can then describe (00:08:15) the feeling of love much better than any (00:08:20) human poet, psychologist or lover. But (00:08:24) these are just words. (00:08:27) The Bible says in the beginning was the (00:08:31) word and the word was made flesh. (00:08:36) The taqing says the truth that can be (00:08:39) expressed in words is not the absolute (00:08:43) truth. Throughout history, people have (00:08:46) always struggled with the tension (00:08:49) between word and flesh, between the (00:08:52) truth that can be expressed in words and (00:08:55) the absolute truth which is beyond (00:08:57) words. (00:08:59) Previously this tension was internal to (00:09:02) humanity. It was between different human (00:09:06) groups. Some humans gave supreme (00:09:10) importance to words. They've been (00:09:13) willing, for example, to abandon or even (00:09:17) kill their gay son just because of a few (00:09:21) words in the Bible. (00:09:24) Other humans have said, "But these are (00:09:27) just words. The spirit of love should be (00:09:31) much more important than the letter of (00:09:34) the law." (00:09:36) This tension between spirit and letter (00:09:38) existed in every religion, every legal (00:09:42) system, even every person. (00:09:45) Now this tension will be externalized. (00:09:50) It will become the tension not between (00:09:52) different humans. It this will be the (00:09:55) tension (00:09:57) between humans and AIs, the new masters (00:10:02) of words. Everything made of words will (00:10:07) be taken over by AI. (00:10:11) Previously all the words, all our verbal (00:10:14) thoughts, they originated in some human (00:10:18) mind. Either my mind I thought this or I (00:10:22) learned it from another human. Soon most (00:10:26) of the words in our minds will originate (00:10:30) in a machine. I just heard today about a (00:10:33) new word that AIS coined by themselves (00:10:37) to describe us humans. (00:10:40) They called us the watchers. (00:10:44) The watchers that we are watching them. (00:10:48) AIs will soon be the origin of maybe (00:10:52) most of the words in our minds. (00:10:56) AIS will mass produce thoughts by (00:10:59) assembling words, symbols, images, and (00:11:02) other language tokens into new (00:11:04) combinations. (00:11:06) Whether humans will still have a place (00:11:08) in that world depends on the place we (00:11:12) assign our nonverbal feelings and our (00:11:16) ability to embody wisdom that cannot be (00:11:19) expressed in words. If we continue to (00:11:22) define ourselves by our ability to think (00:11:25) in words, our identity will collapse. (00:11:31) All this means that no matter from which (00:11:34) country you come, your country will soon (00:11:36) face a severe identity crisis and also (00:11:41) an immigration crisis. (00:11:45) The immigrants this time will not be (00:11:48) human beings coming in fragile boats (00:11:50) without a visa or trying to cross a (00:11:52) border in the middle of the night. The (00:11:54) immigrants will be millions of AIs that (00:11:59) can ride love poles better than us, that (00:12:02) can lie better than us, and that can (00:12:05) travel at the speed of light without any (00:12:07) need of visas. (00:12:10) Like human immigrants, these AI (00:12:12) immigrants will bring various benefits (00:12:14) with them. We will have AI doctors to (00:12:18) help in our health care systems, AI (00:12:21) teachers to help in our education (00:12:23) systems, even AI border guards to stop (00:12:28) illegal human immigrants. But the AI (00:12:32) immigrants will also bring with them (00:12:36) problems. (00:12:37) Those who are concerned about human (00:12:39) immigrants usually argue that immigrants (00:12:43) might take jobs, might change the local (00:12:46) culture, might be politically disloyal. (00:12:50) I'm not sure that's true of all human (00:12:53) immigrants, but it will definitely be (00:12:56) true of the AI immigrants. (00:12:59) The AI immigrants will take many human (00:13:02) jobs. The AI immigrants will completely (00:13:06) change the culture of every country. (00:13:09) They will change out religion and even (00:13:12) romance. Some people don't like it if (00:13:15) their son or daughter is dating an (00:13:18) immigrant boyfriend. (00:13:20) What would these people think when their (00:13:22) son or daughter starts dating an AI (00:13:26) boyfriend? (00:13:28) And of course, the AI immigrants will (00:13:30) have some dubious political loyalties. (00:13:33) They are likely to be loyal not to your (00:13:36) country but but to some corporation or (00:13:38) government across the ocean most (00:13:40) probably in one of only two countries, (00:13:45) China or the USA. (00:13:48) The USA encourages countries to close (00:13:50) their borders to human immigrants but (00:13:53) open them very very wide to US AI (00:13:58) immigrants. (00:14:00) And now we can finally come to the (00:14:01) question each one of you must soon (00:14:04) answer. (00:14:06) Will your country recognize the AI (00:14:10) immigrants as legal persons? (00:14:15) AIS are obviously not persons. They (00:14:18) don't have a body or a mind. But a legal (00:14:21) person is something quite different from (00:14:24) a person. A legal person is an entity (00:14:28) that the law recognizes as having (00:14:31) certain legal obligations and rights. (00:14:34) For example, the right to hold property, (00:14:37) to file a lawsuit, and to enjoy freedom (00:14:40) of speech. (00:14:42) In many countries, corporations are (00:14:45) considered legal persons. The Alphabet (00:14:48) Corporation can open a bank account, can (00:14:51) sue you in court, or can donate to your (00:14:54) next presidential campaign. (00:14:58) In New Zealand, rivers have been (00:15:00) recognized as legal persons. In India, (00:15:04) certain gods have been granted such (00:15:07) recognition. Of course, until today, (00:15:11) recognizing a corporation, a river, or a (00:15:15) god as a legal person was just legal (00:15:19) fiction. (00:15:21) In practice, if a corporation like (00:15:23) Alphabet decided to buy another (00:15:26) corporation or if a Hindu god, (00:15:31) if a Hindu god decided to sue you in (00:15:34) court, the decision wasn't really made (00:15:36) by the god. It was made by some human (00:15:40) executives, shareholders or trustees. (00:15:44) It is different with AIS. Unlike rivers (00:15:48) and gods, AIs can actually make (00:15:52) decisions by themselves. They will soon (00:15:56) be able to make the decisions necessary (00:15:58) to manage a bank account, to file a (00:16:01) lawsuit, and even to operate a (00:16:03) corporation without any need of human (00:16:07) executives, shareholders or trustees. (00:16:11) AIS can therefore function as persons. (00:16:15) Do we want to allow that? Will your (00:16:19) country recognize AIS as legal persons? (00:16:24) What if other countries do it? (00:16:27) Suppose your country doesn't want to (00:16:30) recognize AIS as persons. But the USA in (00:16:34) the name of deregulating AI and (00:16:37) deregulating the markets grants legal (00:16:40) recognition, legal personhood to (00:16:42) millions of AIs which start running (00:16:45) millions of new corporations. (00:16:49) Will you block these US AI corporations (00:16:53) from operating in your country? (00:16:56) Suppose some USI persons invent super (00:17:01) efficient and super complex financial (00:17:04) devices that humans cannot fully (00:17:06) understand and therefore don't know how (00:17:09) to regulate. Will you open your (00:17:12) financial markets to this new AI (00:17:15) financial wizardry (00:17:18) or will you try to block it thereby (00:17:21) decoupling from the American financial (00:17:24) system? (00:17:27) Suppose some AI persons create a new (00:17:30) religion which gains the faith of (00:17:32) millions of people. That should not (00:17:35) sound too far-fetched because after all, (00:17:38) almost all previous religions in history (00:17:41) have claimed that they were created by a (00:17:44) nonhuman intelligence. Now, will your (00:17:48) country extend freedom of religion to (00:17:51) the new AI sect and to its AI priests (00:17:55) and missionaries? (00:17:58) Maybe we should start with something a (00:18:00) bit simpler. Will your country allow AI (00:18:04) persons to open social media accounts, (00:18:07) enjoy freedom of speech, on Facebook, on (00:18:10) Tik Tok, and be friendly with children? (00:18:14) Well, of course, that question should (00:18:16) have been asked 10 years ago. On social (00:18:20) media, AI bots have been operating as (00:18:23) functional persons for at least a (00:18:27) decade. If you think AIS should not be (00:18:31) treated as persons on social media, you (00:18:34) should have acted 10 years ago. (00:18:38) 10 years from now, it will be too late (00:18:41) for you to decide whether AIs should (00:18:44) function as persons in the financial (00:18:46) markets, in the courts, in the churches. (00:18:51) Somebody else will already have decided (00:18:54) it for you. (00:18:56) If you want to influence where humanity (00:18:58) is going, you need to make a decision (00:19:01) now. (00:19:03) So what is your answer as a leader? Do (00:19:07) you think the AI immigrants should be (00:19:10) recognized as legal persons? If not, how (00:19:16) are you going to stop that? (00:19:19) Thank you for listening to this human. (00:19:31) Thank you, Yaval. That was fantastic (00:19:34) overview. You posed a lot of questions (00:19:37) um and they're the right ones. I agree (00:19:39) with much of what you say. We're here in (00:19:41) Davos where the theme is around dialogue (00:19:44) and I was struck by your commentary (00:19:47) around words and the importance of words (00:19:49) and that being something that demarcates (00:19:51) human animals from other animals (00:19:54) although that's debatable that there's (00:19:55) other language there. So in the context (00:19:58) of Davos and the range of people we have (00:20:02) here from technology from the business (00:20:04) world from politicians (00:20:07) >> how would you like to see what is the (00:20:10) answer that you have in terms of this (00:20:12) slightly dystopian world you've (00:20:14) potentially put in front of us (00:20:17) >> and if I may just add to that I think (00:20:19) it's fair to say I'm a scientist by (00:20:21) background a neuroscientist so I uh work (00:20:24) a lot in this space particularly around (00:20:26) pain and we're very comfortable with the (00:20:29) fact that many of our discoveries, (00:20:30) particularly technological discoveries, (00:20:33) we often drive them forward and then (00:20:35) afterwards we think, oh, we hadn't (00:20:37) thought enough about the ethics and the (00:20:39) implications and then we're trying to (00:20:41) catch up on the regulation that we need (00:20:43) to maybe put around it. (00:20:45) >> So, we are where we are. This thing is (00:20:47) happening as everybody says at scale (00:20:49) both in terms of its magnitude and its (00:20:50) pace more than we've ever seen before in (00:20:52) the industrial revolution. We have all (00:20:54) the right blend of people here in Davos. (00:20:56) It's all about dialogue. What would you (00:20:59) like to see go forward in terms of (00:21:02) putting boundaries around some of the (00:21:04) slightly more worrying areas that you (00:21:05) detailed?