YT YouTube2Text

YouTube2Text → Transcripts → Will AI Take My Job? with Karen Hao

Will AI Take My Job? with Karen Hao — Transcript

by Hasan Minhaj · 8,714 words · 1,434 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:00I'll be starring in the new Walt Disney
0:01picture Tron Ares. Tron Ares in theaters
0:04October 10th.
0:06I let Disney scan my body for that
0:09movie.
0:10Did I [ __ ] up?
0:12>> Under what rights are they allowed to
0:14use it?
0:14>> I didn't read the contract. [laughter]
0:17>> Then yeah, you did.
0:21Sorry.
0:22>> As the future licensed trademark of the
0:24Walt Disney Corporation, I was curious
0:26to learn how artificial intelligence
0:28will transform my humanity into
0:30shareholder value. So, I decided to talk
0:33with Karen Hao, author of the new book
0:35Empire of AI. It explores how companies
0:38like OpenAI, Microsoft, and Google are
0:40behaving like [music] modern empires,
0:42seizing and extracting natural
0:44resources, exploiting labor, and
0:47justifying [music]
0:47it all as a civilizing mission to
0:49modernize the world. They basically do
0:52everything empires [music] do except
0:53bomb brown countries.
0:55At least not yet. Now, this sounds bad,
0:58but it also feels
1:02I don't know, just inevitable.
1:04You know? Anytime new technology [music]
1:06comes along, most of us just shrug our
1:08shoulders. We're just like, "All right.
1:11I guess I'm letting the Burger King app
1:12scan my iris." But according to Karen,
1:15>> [music]
1:16>> it doesn't have to go down like this.
1:18With more democratic control, we could
1:20focus AI development on things we
1:22actually need, like health care, [music]
1:25education, clear air, and most
1:28importantly,
1:29in my opinion,
1:31can we please get some robot buttlers?
1:38>> [music]
1:42>> You wrote this book called Empire of AI,
1:44but specifically, to get more specific,
1:46I want to talk about the three major
1:48camps that exist with AI.
1:50>> Yeah.
1:50>> I feel like you got the AI optimists.
1:52>> Mhm.
1:53>> It's here. It's coming. This is a
1:55revolution. You got the AI skeptics.
1:58>> Yeah.
1:58>> This is the end.
2:00Then you have the AI haters. This is
2:03overblown. This is stupid. This is dumb.
2:07Stop.
2:08Which camp are you in?
2:10>> Kind of not any of the three.
2:12>> Okay.
2:13>> I would say I'm in the AI accountability
2:15camp, which recognizes that the AI
2:18industry has consolidated an
2:20extraordinary amount of power.
2:21>> Mhm.
2:22>> And that's an odd to the title of my
2:24book, where I call these AI companies
2:27new forms of empire and how much
2:30economic and political power they have
2:32really concentrated.
2:33Um and I don't I'm not a skeptic in that
2:37I do think there are many types of AI
2:40technologies that can be profoundly
2:42beneficial.
2:43But the type that Silicon Valley has
2:46decided to invest in, which is these
2:47colossal models that they brand as
2:50general everything machines,
2:52>> Yes.
2:52>> is not the path that we should be
2:55focused on.
2:56>> I want to get into this idea of AI as an
2:58empire.
2:59>> Yeah.
3:00>> But
3:00before we get there, let's talk about
3:02what you're mentioning here a little
3:04bit, which is
3:05there's a lot of people in the media and
3:08on the internet, and specifically a lot
3:09of dudes talking very confidently about
3:13what AI even is and what these machines
3:17are doing.
3:17>> Yeah.
3:18>> What can we confidently say? What can we
3:21be skeptical about? And and how big is
3:25the I don't know delta? Cuz that's where
3:27I'm at. I think there is just a huge
3:28>> [laughter]
3:29>> I don't know, and that delta is like
3:31massive. Literally why the show is
3:33called Awesome and Hash doesn't know.
3:35Like there's a lot I don't know. So you
3:37tell me, where where we at? I cuz you
3:39seem to be very measured in your
3:41approach with this.
3:42>> Yeah, so I think we kind of have to do a
3:44little bit of a history lesson. Yeah, so
3:46apologies in advance, but AI has been
3:49around for a really really long time. It
3:50was founded as a research field in 1956
3:53at Dartmouth University by a group of
3:55professors and researchers. And the
3:57original intent was to recreate human
4:00intelligence in computers.
4:03Also, when the term artificial
4:05intelligence was coined, it was
4:07specifically coined by this assistant
4:09professor at Dartmouth called John
4:10McCarthy, who years later said, "I
4:14invented this term because I needed
4:16money." So, they were trying to get
4:18funding for work that they were actually
4:20already doing under a different name.
4:22And I think this is kind of a really key
4:24point for understanding the the
4:26craziness of AI discourse today.
4:29>> Yeah.
4:30>> First of all, there is just so much
4:32anthropomorphization of this technology
4:34because of that original pegging of this
4:38field to the idea of intelligence. And
4:41the problem with doing that is there's
4:42no scientific consensus around what
4:45human intelligence is. So, if you're
4:47trying to recreate that in computers,
4:49you're going to run into a lot of
4:51problems where how do you measure if
4:53you've done that? What is the basis of
4:55our intelligence? And therefore, how do
4:58you recreate it? What should it look
4:59like? Who should it serve?
5:01>> Yeah. I
5:02I love that and I and we were chatting
5:04before this interview started. I was
5:06with, you know, the team.
5:08And we were talking about this idea of
5:11when technology comes out, it is always
5:14given this human interface.
5:16>> Yes.
5:16>> And the joke that I had was I don't know
5:17if you remember in the '80s and '90s,
5:18there were all these movies about like
5:21if you turn the TV on, that the the TV's
5:23going to suck you in. Or Tron,
5:25[laughter] hey, if you you turn you're
5:26going to enter the mainframe. And in
5:30technology and humans, there's always
5:33this thing of oh, they have to be like
5:35us. When in reality, you're like, hey,
5:36this is just a bunch of wires and copper
5:39and microchips. It is not a human being,
5:41it doesn't have a soul. So, I I take
5:43your point of
5:44there is no consensus around what does
5:46it mean to be an intelligent human or
5:48what does human intelligence even mean?
5:50I'll tell you this, there's not a lot of
5:51intelligent humans I know. So, I I
5:52totally understand why you're like there
5:54is no consensus.
5:55>> There's [laughter] no consensus.
5:56>> Yeah, yeah, that's a wild card.
5:59Um, but but if I'm correct me if I'm
6:02hearing this right, you're saying
6:05a lot of times this idea of human
6:07intelligence is given the framing of oh,
6:09it beat a human being at chess.
6:12>> Right, exactly.
6:13>> Excel model faster than a human being.
6:15>> Exactly.
6:16>> mean is that the totality of human
6:18intelligence? Am I Am I hearing that
6:20right?
6:20>> Yeah, exactly. It's like okay, it can do
6:23certain tasks better than humans.
6:25But what's the significance of that? And
6:29ultimately what OpenAI as a company
6:32tries to do is create this rhetoric
6:34where
6:35when you recreate human intelligence,
6:37which apparently they are going to do
6:40soon, so they say, uh, that it's somehow
6:44going to have you know, profound utopian
6:48consequences. We are going to have so
6:50much abundance in the world, so much
6:52prosperity in the world that this
6:54intelligence is going to help us solve
6:56cancer, it's going to help us solve
6:57climate change. Um, and that is
7:02based in you know, a fiction of how this
7:05technology actually works.
7:06>> The book is obviously called Empire of
7:08AI.
7:11Why is AI like an empire? How is it like
7:14an empire? And like most empires, when
7:17will it start bombing brown countries?
7:20>> So, in the history of European
7:23colonialism, empires of old had several
7:26features to them. They laid claim to
7:28resources that were not their own, but
7:29they redesigned the rules to suggest
7:32that they were their own.
7:35They exploited a lot of labor, they
7:36didn't pay that labor or they paid that
7:39labor very little, and empires were
7:42always in competition with each other.
7:44So, the British Empire was always
7:45saying, "We are better than the Dutch
7:47Empire." And the French Empire was like,
7:49"We are better than the British Empire."
7:50So, there was this concept that there
7:51were evil empires and there were good
7:54empires. And the reason why the good
7:56empires had to be empires
7:58is because they needed to take down the
8:00evil empire. They needed to be strong.
8:02So, that's why they're extracting all
8:04these resources, they're exploiting all
8:05this labor to fortify themselves.
8:08And they're doing it ultimately under
8:09the civilizing mission of we are doing
8:12this for the benefit of all of humanity.
8:15We are actually bringing religion to all
8:18of these heathens and giving them an
8:21opportunity to access heaven instead of
8:23hell.
8:24Empires of AI not only literally use
8:27this rhetoric now, but they check off
8:30all of the characteristics of empire
8:31building. They lay claim to resources
8:33like the intellectual property of
8:35artists, creators, writers, and then
8:36they redesign the rules to say, "Well,
8:38this is actually fair use to train our
8:40AI models on this." And then they
8:43exploit a lot of labor both in that they
8:45contract workers in primarily global
8:48south countries, going to the ground
8:49countries,
8:50and paying them very, very little
8:52amounts of money to take out toxic,
8:55abusive,
8:56sexist, racist speech out of their
8:58models, and also in the sense that they
9:01are ultimately creating labor automating
9:03machines. So, the
9:06the definition that OpenAI officially
9:08uses for AGI is highly autonomous
9:10systems that outperform most humans in
9:13economically valuable work.
9:16>> Got it. Okay, so things that drive
9:19shareholder value.
9:20>> Exactly. And so, you can imagine if a
9:24worker is going to the bargaining table
9:27and sitting across from a CEO, and both
9:29of them think, "Wait a minute. If I
9:32resist, or wait a minute, if that guy
9:34resists, I can just hire an AI instead."
9:37That worker can't bargain for rights
9:39anymore. So, that's the labor
9:40exploitation that's happening under
9:42empires of AI. And they have this
9:44aggressive competition where OpenAI
9:47frames itself as we need to be the good
9:49guys so bad guys can't create AGI.
9:52>> Yeah.
9:53>> All under a civilizing mission of
9:54they're doing it for the benefit of
9:55humanity.
9:56>> sounds so common to
9:59you know, I'm 39 years old so I heard
10:01this with the web 1.0 and the Googles
10:03and all that sort of stuff. This idea of
10:06we're going to do good in the world.
10:08>> Yeah.
10:08>> Um one There's this quote by Sam Altman
10:11in 2013 that I want to take a look at,
10:14which is "The most successful founders
10:16do not set out to create companies. They
10:19are on a mission to create something
10:20closer to a religion. And at some point
10:22it turns out that forming a company is
10:24the easiest way to do so."
10:27Is Sam Altman forming a cult?
10:30>> I've started increasingly thinking that
10:32the best way to actually understand the
10:34AI world is Dune.
10:36>> Okay.
10:37>> Where you create this mythology. So,
10:40like
10:41Paul Atreides, his mom, right? She
10:43creates this mythology around him being
10:47the coming of the Messiah.
10:49And most people who hear this myth for
10:52the first time, they don't realize that
10:54it was hand-crafted
10:55to control the people and make sure that
10:59Paul would ultimately have power.
11:02And eventually as he steps into this, he
11:06starts to forget that the myth was
11:07originally a creation and he starts to
11:09believe in it himself, right? And I
11:12think this is exactly what's happening
11:14in the AI world with the kind of
11:16rhetoric that they use where they talk
11:18about building digital gods and digital
11:20demons literally.
11:21>> Yeah.
11:22>> Is that they created this mythology at
11:24some point someone created a mythology
11:28around the the extraordinary power of
11:31these technologies and the need to usher
11:34it in carefully
11:36very conveniently by the people that
11:38created that mythology.
11:40And now we are out of place where
11:42essentially
11:44everyone that exists in this ecosystem
11:46in Silicon Valley has forgotten or has
11:49come to believe or maybe always believed
11:52that this is their sole purpose and this
11:54is what they need to do for the world.
11:56>> One of the things you talk about in the
11:57book is that he he talks about Napoleon
11:59constantly. And and Sam
12:03sometimes often times according to his
12:05critics lies. By the way, Sam, if you
12:07want to come on the show, we'd love to
12:08have you on the show.
12:09But what is that? What what's going on
12:12with that? What is with this obsession
12:13with
12:14you're you're you're on Slack but you're
12:16somehow
12:18quoting Napoleon and Marcus Aurelius and
12:20Socrates and the ancient kings of old.
12:23What what is that?
12:24>> I mean, you know, Sam Altman is a
12:26product of Silicon Valley and we've seen
12:28this character before, right? Mark
12:30Zuckerberg is also obsessed with the
12:32emperors of old and they literally
12:35colloquially say to one another in these
12:38spaces
12:39how do you build this empire? We are
12:41building empires. Sam Altman says
12:44has said when he was president of Y
12:46Combinator which was his previous job.
12:48YC was is one of the most prestigious
12:51startup accelerators in Silicon Valley.
12:53He said the thing that I was proudest of
12:55is that I built an empire. So I think
12:57this is like they look up to these
12:59historical figures who really not just
13:04built
13:05the empire or in Napoleon's case failed
13:07to build his empire but how they went
13:11about doing it and the thing that Altman
13:14has said he really admired about
13:16Napoleon is his ability to fully
13:19understand
13:20what people want and
13:22that is exactly how people describe
13:25Altman's superpower is he is really good
13:29at knowing what you want and then saying
13:32a story based on what you want that
13:35makes you really really want a piece of
13:37the future he's selling.
13:38>> got to spend some time
13:40in OpenAI's offices.
13:43What's the tea? What's the C-suite
13:45gossip? What did you take away from that
13:47time?
13:48>> So, I was the first journalist to
13:50profile OpenAI. So, I embedded within
13:52the company for 3 days in August of 2019
13:55back when pretty much no one had heard
13:57>> Sure.
13:57>> of OpenAI.
13:58>> Pre-COVID? This is This is another time.
14:00>> Yeah, exactly.
14:01Yeah, it was a it was a different Yeah,
14:02and my
14:03>> [laughter]
14:04>> my profile published in February 2020.
14:06So, truly um right on the cusp of us
14:09entering a different era.
14:11But, at the time OpenAI was founded as a
14:14nonprofit.
14:16It was founded on the principles of
14:17being totally transparent, doing the
14:20work of advancing AI without any
14:23commercial incentive,
14:24>> Right.
14:24>> and open-sourcing it to everyone, being
14:27collaborative with everyone.
14:29By the time I got to the offices in
14:31August 2019,
14:33those were starting to change. OpenAI
14:36had restructured, so it nested a
14:38for-profit within the nonprofit. It got
14:40a billion dollars from Microsoft. And I
14:43noticed when I was at the company, wait
14:47a minute, they say that they're
14:49collaborative publicly, but they're
14:51telling me internally, we need to be
14:55number one, otherwise our mission does
14:57not work.
14:58>> Right.
14:59>> And I thought,
15:01that's competitive. There's a tension
15:03here. And then they said, we are
15:06transparent, we're going to open-source
15:07everything.
15:08And then internally, they were like,
15:10there are certain things that we cannot
15:12talk about. You cannot see.
15:14And I was like, wait a minute, that's
15:16really secretive.
15:16>> Oh, got it. So, there's a dissonance
15:18here.
15:18>> Yeah.
15:19>> Yeah.
15:19>> So, what they're saying publicly to
15:22accumulate a lot of goodwill and to
15:24grease the wheels for a lot of
15:25accumulation of capital is actually not
15:28how they're operating behind closed
15:29doors.
15:29>> Aren't
15:31all tech companies kind of like this
15:32where they they all kind of talk like
15:34they're UNICEF? Like we're here for the
15:36global good and then until there is a
15:39bag involved or until they're strapped
15:40for cash and they need money or
15:42investors or
15:44more capital or or market share.
15:46>> Exactly. And that was what I realized
15:48was they had positioned themselves as
15:50anti-Silicon Valley as a new form of
15:54tech organization that was going to do
15:57things better than the previous era of
15:59Silicon Valley. And then I realized,
16:00wait a minute, no. This is actually a
16:02continuation. And now that we fast
16:05forward all the way to present day, I
16:07mean OpenAI is one of the most
16:09capitalistic companies in Silicon
16:12Valley. They just raised $40 billion
16:14at a $300 billion valuation, which is
16:17the largest private tech investment
16:20fundraise in the history of Silicon
16:22Valley and places them as one of the
16:24most valuable companies in the history
16:27of
16:28private startups in the history of
16:30Silicon Valley.
16:31>> Obviously OpenAI is huge. ChatGPT is
16:34huge. Microsoft acquiring them, huge. So
16:37they're they're now a player. They are
16:40here and they're probably
16:42at least for the near future, they're
16:44here to stay. They're a huge company and
16:45clearly with a with a sizeable uh market
16:49cap and huge investment. There is now
16:51this talk about artificial intelligence
16:54and these AI companies creating
16:55commercial products.
16:57>> Yeah.
16:58>> What does that mean?
16:59>> [gasps]
17:00>> Honestly,
17:02what people should know is that means
17:05they're trying to get more of your data
17:07because they are trying to figure out
17:10how to make their
17:13products, their technology so attractive
17:15that they can continue building them.
17:16>> Don't they already have all my data?
17:18Facebook, Apple, Netflix, Google, you
17:19got everything.
17:21>> That's what's wild and that's why I
17:22think we've reached the moment where we
17:24can no longer talk about these as
17:25companies and we have to talk about them
17:27as empires. Is the amount of data that
17:30they need
17:32has completely eclipsed the amount of
17:34data that social media companies took
17:36from us.
17:36>> Really? How?
17:37>> Yeah, so so if you just look at Meta,
17:39which has also entered the AI race, I
17:42mean Meta literally has 4 billion users'
17:46data from their previous era as a social
17:49media company and they were using that
17:52to create their really lucrative ad
17:54targeting algorithms, right? But even
17:57then, The New York Times reported last
17:59year that Meta was having conversations
18:01about we don't have enough data, we need
18:03to potentially buy Simon & Schuster, we
18:06need to potentially ignore all the data
18:08privacy rules that we set up after
18:10Cambridge Analytica, we need to
18:11potentially ignore all the copyright
18:14rules and just acquire more and more and
18:16more because with the current repository
18:19of data that we have, the 4 billion
18:21users, we cannot outcompete OpenAI. Like
18:25that is an order of magnitude, maybe
18:27multiple orders of magnitude more data
18:29that we're talking about.
18:30>> Data's always,
18:31you know, esoteric and when I try to
18:32talk about this with even people my age,
18:34my generation, even a generation
18:36younger, there is this acceptance of
18:38like, "Hey, I clicked accept on the
18:39iTunes user agreement, you got my data."
18:41Like I never
18:43had privacy. Everything is compromised
18:46anyways. From the Nest camera that's
18:47letting me know [laughter]
18:49who's dropping off what
18:50>> Yeah.
18:51>> to my apartment to every single one of
18:53my photos when I'm trying to upload a
18:55reel on Instagram. Access to all photos?
18:58Sure. I need to make this carousel dope.
19:00So, for me as a human being, I'm a
19:04husband, I am a father,
19:06I got two kids. I'm like, "Look, you got
19:08my data and this is to make my world
19:10better?" Yeah. "When will you give me
19:14the robot butlers?
19:15>> Yeah.
19:16>> You got all my data. When are the robot
19:18butlers getting here? I'm talking about
19:19the people that are going to make my
19:21life better. Please cook my food. Please
19:24Please clean my baby's booty.
19:27We argue about the dishes. People come
19:29over. We got more dishes. Da da da. Do
19:31the dishes. Fold my laundry. Cook the
19:33food. Like
19:35Why
19:36I don't want another app on my rectangle
19:38of sadness. I want the robot butlers. Is
19:40that going to happen with these AGI and
19:42AI machines or not?
19:44>> Totally not. And here's why.
19:46>> Oh, no.
19:46>> [laughter]
19:47>> Here's why you should stop giving all of
19:49your data to these companies.
19:52You're seeding a lot of control and
19:55agency of your life without actually
19:57getting much in return. Now, there used
19:58to be a time, I think, when there was
20:01kind of a fair trade-off of, "Okay, I
20:03get a little bit more convenience. I get
20:05some kind of technology that I've never
20:06gotten before. I get to connect with my
20:08long-lost elementary school friend."
20:11But we have reached a point where these
20:15companies they have gone gotten so much
20:18economic and political leverage. They're
20:21developing such a controlling influence
20:22over all spheres of society, including
20:25scientific production, including
20:27geopolitics, that they are reaching, or
20:30I believe have reached, an inflection
20:33point where they can start acting in
20:35their self-interest with basically no
20:37consequence. And originally, the bargain
20:41of giving [clears throat] data to
20:42companies is they will give you
20:44something in return. But these companies
20:46have reached empire status where they
20:48don't actually have to give you anything
20:50in return anymore.
20:51>> But what about like the way Grok can
20:53give me a demented photo of Bill Gates
20:56and Elon Musk having lunch? Like isn't
20:58that a
21:00exchange? Or summarizing a very, very
21:04dense 89-page
21:07P&L report into something I can quickly
21:10make a decision on. Is that a fair
21:12exchange in your
21:13>> Grok is a great example of how these
21:16companies operate because in order to
21:19train Grok, Elon Musk set up a
21:21supercomputer called Colossus in the
21:23Memphis, Tennessee area
21:26and completely hijacked local democratic
21:28processes to put it up as quickly as
21:31possible and start powering it with
21:34unlicensed methane gas turbines that are
21:37now pummeling [snorts] that area with
21:40huge amounts of air pollution.
21:43Talking about black and brown
21:44communities.
21:45And so
21:47yes, there are thing there are certainly
21:50interesting utilities that come out of
21:52these tools and there's certainly people
21:53that actually benefit a lot from using
21:56these tools.
21:57But the supply chain of producing these
21:59tools has already illustrated to us the
22:02logic of what's happening here, which is
22:04that these companies don't actually care
22:07about preserving people's right to even
22:11clean air
22:12in order to ultimately produce something
22:15that they are trying to use to then
22:17accumulate more data and get more money.
22:20>> Got it. Do you mind if we back up a sec
22:22and just
22:23can you define what AGI is?
22:28Because there's a lot of people that are
22:29watching the show, that listen to the
22:31show, that
22:32it sounds like we're all talking about
22:34something different. What is AGI and is
22:37it ex machina or not?
22:39>> AGI is whatever the companies need it to
22:41be.
22:42If they want to sell you a convenient
22:44product, they are going to talk about
22:45AGI as the movie Her
22:48and say, "This is going to make your
22:50life so amazing. It's an operating
22:52system for your life."
22:54If they want to talk to Congress to ward
22:56off regulation, AGI is suddenly this
22:58mythical object that will solve climate
23:01change and cure cancer.
23:03And so AGI morphs, and that's why no one
23:06really can say what AGI means because it
23:10shapeshifts based on what the companies
23:12need it to be.
23:13>> Yeah, it's that's the thing I keep
23:14seeing in different settings. It can be
23:16congressional testimony, it can be a
23:18Super Bowl commercial.
23:19>> Yeah.
23:20>> So AGI is going to cure cancer. AGI It's
23:23going to solve climate change. Or I
23:24mean, the one that struck me, I have
23:26older parents, they go AGI is going to
23:28be able to look at your parents'
23:30bloodwork and identify exactly what's
23:32wrong with them.
23:33>> Yeah.
23:33>> And so for me, I'm like, that's awesome.
23:35>> Exactly.
23:36>> But at its core is artificial general
23:38intelligence, a machine that just takes
23:40complex data sets.
23:42>> Yeah.
23:43>> Essentially unknown variables and then
23:45just crunches out the answer to that
23:47data set. At its core, is it that?
23:48>> That's what AI at its core is at the
23:50moment at the moment.
23:52>> Okay.
23:52>> Which The reason why I say that is
23:54because there are many different
23:56techniques that could be used to
24:00automate certain types of tasks um that
24:04traditionally we think only humans could
24:06do. And it just so happens that right
24:08now we are in a realm where the
24:10technique is very, very much data-driven
24:12data processing.
24:13>> Right. Are you more pro task-specific
24:17AI? Like, are you in alignment on hey,
24:19I'm for a product if it's specifically
24:22about bloodwork.ai.
24:25>> [laughter]
24:25>> I just literally take everybody's
24:28bloodwork, you know, grandma's bloodwork
24:30at Kaiser, and I'm going to tell you,
24:32hey, she may have a likelihood for X or
24:35Y disease.
24:38And then are you arguing
24:41against more kind of this general data
24:43scraping AGI of like, just give me
24:45everything and I'll tell you about it
24:47later.
24:48>> Absolutely. I That's exactly right. Like
24:51These companies are trying to build
24:53everything machines. The problem with
24:55everything machines is that they can't
24:57actually do everything. They do some
24:59things for some people because also time
25:02and time again we've seen through the
25:03history of AI development that models
25:06have embedded biases based on the data
25:08that they're trained on, based on who
25:10gets to leave data on the internet and
25:13who gets to shape these technologies.
25:16Um and so ultimately when you position
25:19your product as an everything machine,
25:21not only are people going to be really
25:23confused and start using it for things
25:24that it's actually not that good at and
25:26it could lead to a lot of harm like
25:28people asking ChatGPT to read their
25:30medical records. Like ChatGPT's not
25:32actually designed to be able to do that
25:36because it's not 100% accurate 100% of
25:38the time. It's a probabilistic machine.
25:42Um and so
25:44in the task-specific approach, not only
25:47is that better for consumers in terms of
25:50it being super clear like how are you
25:52supposed to use this AI model to make
25:54sure you get the maximum
25:55>> benefits from it.
25:55>> Right.
25:56>> It also is a way better for developers
25:59to develop tools that work because then
26:02there's a very well-scoped space in
26:05which they can test all of the different
26:07failure modes of this technology and
26:09continue shoring them up. You cannot
26:11test all the failure failure modes for
26:13an everything machine.
26:14>> Right. So
26:17let's play devil's advocate here. If if
26:20I'm arguing for the everything machine,
26:23what if I go
26:25Karen, I hear you.
26:27I'm figuring it out. I'm iterating as
26:30they say in Silicon Valley. I'm moving
26:31fast and I'm breaking stuff, but my my
26:34North Star cardinal direction is
26:36something good and I do want to do this
26:38with good intent.
26:40What's your response to that? Is it no,
26:43like good intentions is the path to hell
26:45or what what do you say to that?
26:48I would say that we need to look at how
26:53AI is being developed right now and the
26:54harms that it's creating right now all
26:56around the world.
26:57>> world consequences.
26:58>> The real world consequences because that
27:01is those are the data points that that
27:04is our evidence to understand
27:08what this technology is going to do for
27:09us in the future.
27:10>> You know, what's interesting is every
27:11empire has these thing called sacrifice
27:14zones. You know, the British Empire
27:16obviously had India and Africa. Those
27:17were sacrifice zones. Here in America,
27:19we we have our sacrifice zones. iPhones
27:22made in China. We have Bengali kids
27:24making our Nikes. We're aware of this
27:26and this has existed for a long time.
27:28Who are the invisible people of the AI
27:30empire that we're not seeing right now?
27:33>> So, in my book I go to Kenya, I go to
27:35Chile, I go to Uruguay, to Colombia. And
27:39in Kenya for example, Kenya and
27:41Colombia, I was talking with workers
27:44that are contracted by these companies,
27:47these AI companies to do some of the
27:50worst work in the AI supply chain. So,
27:53Kenyan workers, Open AI went there. They
27:55were at a moment in their history as a
27:58company where they realized, wait a
28:00minute, we need to start commercializing
28:02and if we start putting models that can
28:04spew anything in the hands of users,
28:07it's not going to be a huge commercial
28:08success if it starts spewing a lot of
28:10hate speech. So, we need a put a content
28:12moderation filter around it.
28:14>> Right. And content moderators are like
28:15human beings that literally have to look
28:17at things as awful as child pornography
28:19to snuff, like really bad stuff. So that
28:22it doesn't end up in your feed while
28:23you're texting in traffic. And Kenya is
28:24that sacrifice zone where it is long
28:24served as a backstop
28:25>> texting in traffic. And Kenya is that
28:28sacrifice zone where it is long served
28:30as a backstop for the internet of the
28:33global north. And so, Open AI shows up,
28:35they contract these workers and they
28:37ask, hey,
28:39label all of these worst like text from
28:43the worst parts of the internet and AI
28:46generated text where we prompted an AI
28:49model to imagine the worst text on the
28:51internet, read that day in and day out,
28:54label it into a detailed taxonomy where
28:57you have to say, is this sexual content?
29:00Is it sexual abuse content? Is it sexual
29:02abuse content that involves children?
29:05And those workers, like all content
29:07moderators, ended up psychologically
29:10devastated. And not only them, because
29:12these individuals are part of
29:14communities, they're people that depend
29:15on them. And I write about a man named
29:18Mofa Okinyi, who I met who was one of
29:21the Kenyan workers contracted by Open
29:23AI, where he completely changed his
29:26personality. He was on the sexual
29:28content team. And his wife had no idea
29:32what was going on, because he had no way
29:36to tell her, "Oh, I'm reading sex
29:38content all day." ChatGPT hadn't come
29:40out yet. There was no conception of what
29:42this work was for.
29:45And one day, she texts him and says, "I
29:50like I would want fish for dinner." He
29:52goes out, buys three fish, one for him,
29:54one for her, one for her daughter, his
29:56stepdaughter, who he loved and adored
29:58and called his baby girl. And when he
30:00shows up back home, they've left
30:03completely. All their stuff is gone, and
30:05his wife texts him,
30:07"You've changed. I don't know the man
30:09you are anymore."
30:10And she never comes back.
30:18>> This is a
30:20really heavy stuff, and
30:24what I took away
30:26from the book and what you're talking
30:28about
30:30really is this very
30:33modern, updated, but classic I call it
30:36critique of capitalism. And how
30:41the benefits
30:45whether those be social or business
30:47profits,
30:48are not equally distributed.
30:52But then I got to thinking, I go,
30:55"Have the benefits of technology and
30:56these tech companies and these empires
30:58ever been equally distributed?" Is this
31:01the story of man, sadly?
31:04I I haven't been able to reconcile that.
31:06How have you processed all of this?
31:09>> To me, it's I cite a book in the book
31:13called Power and Progress, which was
31:15written by two MIT economists, Daron
31:17Acemoglu and Simon Johnson. They just
31:18won the Nobel Prize to economics last
31:20year.
31:21And they say exactly this, that that
31:24over the 1,000 years of technology, they
31:27analyzed 1,000 years technology's
31:29been around for longer, but analyzing a
31:31thousand years of technology history,
31:34there is a consistent pattern that we
31:36see in every technology revolution,
31:39that the elites are the ones that have
31:42the money, the influence, the power to
31:44actually rally enough resources around
31:48creating certain new technologies, but
31:50it's also created in their image. And
31:53consistently, there's a lot of fallout
31:55that comes from that, where people who
31:57do not live like them, who do not look
32:00like them,
32:01end up being harmed. Either their jobs
32:04are lost or worse, you know?
32:06Um, but the thing that Silicon Valley
32:10will always tell you is that is
32:12justification for why this technology
32:15revolution is happening in the same way.
32:17And to me,
32:19it's like, wait a minute. Most of these
32:21technology revolutions that have
32:22happened in the last 1,000 years were
32:24when we didn't have human rights in
32:26existence. We didn't have democracy.
32:29People didn't believe in their own
32:30agency and their right to
32:31self-determination.
32:33And this technology revolution is now in
32:36an era where we have all those things.
32:38So, we should want better. We should
32:41want more. And we should actually
32:43reinvent the way the technology
32:45revolutions happen so that they don't
32:47just repeat all of the terrible things
32:50that happen in previous revolutions
32:52without any rights.
32:53>> You're going to be doing the media
32:54rounds talking about this book, and
32:56you're probably going to hear, and
32:57you've heard this probably in online
32:58discourse, but you're going to hear it
33:00as you do the rounds, "Hey Karen, I'm
33:01sorry, the genie is out of the bottle."
33:03I call it the genie is out of the
33:04bottle.pdf
33:06paradox. Hey, guess what?
33:09Whether this company does it, another
33:12company will do it. Whether America does
33:14it, China will do it. Somebody is going
33:16to do it. So, you better get on board
33:17and just give in.
33:20Um what do you say to that? What do you
33:22say to this like
33:24"Hey, it's already happening, boomer, so
33:26get on board." Like what do you
33:28>> [laughter]
33:28>> I'm not saying you're a boomer. I get
33:30told that. But you know what I mean? I
33:31get told this all the time by
33:32techno-optimists. Yeah. Like it's
33:34happening.
33:35>> Yeah.
33:35>> So, do you want to be
33:36>> Of course they're going to tell you that
33:38because that's like the the the a
33:40feature of empire is they're made to
33:43feel inevitable.
33:44>> Yeah.
33:45>> And that is that is part of their power.
33:47Their persuasive power is you can't stop
33:49it. It's it's an unstoppable force. But
33:51the thing is every empire has fallen in
33:54history because they're actually really
33:56weak at their foundations. And the way
33:59that I think about how we can actually
34:01contain the empire is thinking about the
34:03full supply chain of AI development.
34:06These companies, in order to do what
34:07they do, they actually need resources
34:10from us. They need our data. They need
34:13the land, energy, and water to power
34:15their data centers. They need that
34:17labor. They need talent, the AI
34:19researchers that are working within
34:21their labs. And they also need consumers
34:24to buy their technologies, to deploy
34:26them into classrooms, into healthcare,
34:28into all these different spaces. And all
34:31of these things are what I like to think
34:34of as sites of democratic contestation.
34:37There are already movements happening
34:39where artists are glazing their work
34:42when they put it up on the internet in
34:44online portfolios such that there's no
34:47difference to the naked eye, but when an
34:49AI model trains on it, it breaks the AI
34:51model apart. And that's one form of
34:53resistance of
34:55if you're not going to give me if you're
34:57not going to ask for my consent, if
34:58you're not going to compensate me, you
35:00don't get this data for free. And there
35:02are already
35:04workers strikes in the Hollywood writers
35:07who are saying, "We're not going to
35:08allow AI to be deployed in certain
35:11contexts. There need to be guidelines
35:13and conditions around when AI is and
35:15isn't deployed in our work." There are
35:18activists all around the world that are
35:20fighting back data centers that are just
35:23landing in their communities. And by the
35:25way, like these data centers often come
35:27in without any transparency.
35:29Like Meta built a data center in New
35:31Mexico under a shell company name called
35:34Greater Kudu LLC. And it wasn't until
35:37the deal was done that they went,
35:39"Surprise, it's Meta."
35:41And so, all of these residents are
35:43rising up being like, "We need more
35:45transparency. We need you to guarantee
35:47that either if you bring in a data
35:49center, you give us jobs, or you tell us
35:53that you're not going to use above a
35:55certain amount of water, a certain
35:56amount of energy, or you don't come at
35:58all." And so, I think we have to sort of
36:00remember that
36:02Silicon Valley has done a really good
36:03job of creating this culture where they
36:07make you feel like everything that you
36:09own is actually what they own.
36:11But we have to remember that we actually
36:13own this data, we own these spaces, we
36:17have a right to
36:19elect officials that protect our
36:22life-sustaining water.
36:24And if everyone actually remembers that
36:28and asserts, "Hey, we want AI to be be
36:31developed this way. We want it to be
36:33deployed this way." Companies have to
36:35follow. They're ultimately businesses.
36:38>> Is there a central place where
36:40collective action can gather around
36:43a common almost set of human rights
36:46>> Yeah.
36:47>> in the face of this AI revolution?
36:50>> And I think not necessarily central
36:51place, more of a distributed many, many
36:54places. You know, if you're a parent
36:57you are a parent.
36:58>> a parent.
36:59>> The the fact that your school is
37:01implementing certain technologies that
37:03are going to affect your kids, like
37:05build a parent group or parent-teacher
37:07coalition and be like, "Hey, let's
37:08actually talk about this before you
37:11start, you know, using facial
37:12recognition on my kid. Before you start
37:14turning my kid into a QR code. Let's
37:16actually set some guidelines around what
37:19kinds of technologies we do want to use
37:21and don't want to use." If you are going
37:24to your doctor's office, like ask, "What
37:27AI do you use and can I opt out?" And
37:31maybe get together with other patients,
37:33other the nurses in the office and ask
37:36like, "Can we create guardrails around
37:38that, too?" When you go to work, your
37:41job is almost definitely now talking
37:43about like, "How do we adopt AI? What is
37:45our AI policy?" Get together a group of
37:48co-workers, talk to your boss, like,
37:49"Let's have a meeting about this."
37:52>> So, this is really rubber meets the
37:55road. And what's funny is, you know,
37:57sometimes people go, "Hey, listen,
37:58collective action, that's a privilege. I
38:00got to pay the bills." So, let's
38:02actually talk about the bills, your job.
38:04Will AI take my job?
38:07What's your stance on that? What's going
38:09to happen?
38:10>> I think AI can absolutely take people's
38:13jobs because of the way that Silicon
38:16Valley has started pitching the
38:17technology to try and earn back all the
38:20money that they're spending, which
38:21they're going to executives and saying,
38:24"We can make your workforce a lot
38:27cheaper by giving you these AI tools."
38:29But, there was a really funny headline
38:31recently where a company [clears throat]
38:33declared, "We have entered the AI era."
38:35And then fired a bunch of people to
38:37replace them with AI tools. And then a
38:38few weeks later, they were like, "Oops,
38:41this these AI tools are not good enough.
38:43Can you please come back?"
38:44>> Right.
38:44>> So, the reason why AI is going to
38:47automate jobs is not always going to be
38:49because the AI tools are actually up to
38:52snuff. It's because people are putting
38:54the cart before the horse and just
38:56getting rid of workers,
38:57being pulled into this allure that AI is
39:01the solution.
39:02>> Sh- How should I think about it because
39:04I've heard two different versions? And I
39:05And it It sounds like the story you just
39:07told me is simultaneously both. AI will
39:10take your job, and there will be
39:11corporate downsizing because because of
39:13it. But then simultaneously, it may
39:15create new jobs because these systems,
39:17these AI models, are slightly or majorly
39:21flawed. So, what is it? Is it Is it a
39:24little bit of both?
39:25>> It is going to be a little bit of both
39:26because at the end of the day, these
39:28aren't actually everything machines. The
39:30companies have lists of economically
39:33valuable tasks that they are trying to
39:36design these systems to perform
39:38particularly well at. And um I had a
39:40trove of documents that I had access to
39:42of the tasks that they were trying to
39:44specialize these models in in the book.
39:48And they tried to target the most
39:50lucrative industries, entertainment,
39:52media, finance, healthcare, because
39:55those are the industries where they can
39:56show up to the executives who pay the
39:58big bucks.
39:59>> Mhm.
39:59>> And so, that is where
40:02these models might get really good. And
40:04you know, these companies are investing
40:06a lot in automating coding, which is
40:08particularly um something that AI is
40:10good at. It's super computational.
40:13>> Right.
40:13>> And [snorts] so, there will be certain
40:15things that will certainly like AI
40:18models will be technically competent at
40:21replacing a human. There will be many
40:23other things that it will not be.
40:26But that won't necessarily have a
40:28bearing on whether that person keeps
40:31their job anyway because ultimately it's
40:33not actually AI taking your job, it's
40:35humans. It's an executive deciding that
40:38your job is now redundant.
40:41>> Interesting. I
40:43hear this with
40:45the education system as well.
40:47There's this idea of like, well
40:50AI can it can code, it can text, it can
40:54write, it can summarize
40:56and it can analyze complex data sets.
40:59You might as well be illiterate.
41:02What do you say to that?
41:03For some reason I firmly disagree, but
41:06what do you what do you say to that? IS
41:07IT COMING
41:07>> YEAH, NO, I DO FIRMLY DISAGREE. I mean
41:11this is like
41:13in order for a democracy to function I
41:14mean this is this I'm I'm getting really
41:17high level here, but
41:18>> get philosophical here.
41:20>> In order for democracy to function we
41:22need critical thinking skills. We need
41:24agency. We need to be able to be
41:26independent from the crutches that
41:28Silicon Valley is trying to sell us, you
41:30know?
41:30>> Mhm.
41:31>> And so ultimately I mean the best thing
41:33is for technology to be assistive to
41:36people, not to totally gouge out their
41:38brains. [laughter]
41:40>> Right.
41:41Do you feel unfortunately as someone who
41:44you know, you write and you cover
41:46stories like this for a living
41:49do you see it frying our brains?
41:52>> I do see it frying a lot of people's
41:54brains, but
41:56the thing that has been really amazing
41:58is at the same time there is now more
42:01[clears throat] conversation than ever
42:03before about AI and whether it's good or
42:05whether it's bad, what do we want out of
42:07it? Like I've been covering this since
42:092018 and this is the first time that I
42:12mean we are having actual global
42:14conversations about the ethics of this
42:17yeah and so that is I think a sign that
42:20it's going to take a lot of hard work to
42:22readjust the the vehicle that's
42:25bulldozing its way in one direction
42:28but we're going to get there.
42:30>> In 2023
42:32this conversation around AI really took
42:34center stage in my industry around the
42:36WGA and the SAG strikes.
42:39Um
42:40there was
42:42strikes and then there were
42:43negotiations. How do you think
42:46the strike went and how do you think
42:48it's played out since? What what have
42:51you noticed that's good or bad about
42:53what went down?
42:56>> It definitely showed that collective
42:58action is an extremely important
43:01mechanism to hold on to for demanding
43:05certain protections against AI
43:08but the the specific details of like how
43:10specific like what they negotiated I
43:13couldn't say
43:14how it actually has
43:17resisted or or been resilient under the
43:20test of time but I think to me it was
43:23really amazing that
43:26they actually got the executives to the
43:29negotiating table to actually put AI up
43:32for discussion and that is something
43:34that every industry
43:38any worker anywhere can learn from.
43:40>> Um I'll be starring in the new Walt
43:42Disney picture Tron Ares Tron Ares in
43:45theaters October 10th.
43:47I let Disney scan my body for that
43:50movie.
43:52Did I [ __ ] up?
43:54>> Under what rights are they allowed to
43:55use it?
43:56>> I didn't read the contract. [laughter]
43:59>> Then yeah you did.
44:02Sorry.
44:03>> I did ask for free tickets to
44:04Disneyland. That's all I did and then I
44:06walked in.
44:07>> I mean, you know, if it works for you.
44:09If that's a fair trade.
44:12>> [music]
44:14>> As I scrolled through the thing, I I
44:16almost felt I was like, well, is there
44:18just an AI thing that can summarize what
44:19I'm about to sign?
44:21>> [laughter]
44:23>> [ __ ]
44:25>> But, you know, I think a lot of people
44:27are starting to feel the way that you're
44:29feeling in this moment of, wait a
44:30minute. There are some things that I did
44:32in the past that maybe I should
44:33reconsider how I do in the future. And I
44:36think that is exactly
44:39what's going to help.
44:40>> So, what is the alternative? How should
44:43we look at the next 5, 10, 15, 20 years?
44:47Um, while I'm, you know, still
44:48negotiating my lower back pain and
44:51I do have a modicum of sanity.
44:53I For real, I I think about this like
44:55the next 20, 25 years of my life. What
44:57What are the alternatives to what we
44:59currently have and
45:01what what should we do?
45:03>> Collective action. Also, investing in
45:05different types of AI technologies.
45:09This specific paradigm of growth at all
45:12costs, scale at all costs, that's coming
45:15out of Silicon Valley with respect to
45:17how they develop AI models,
45:19we don't need to do that. Um, there's an
45:23amazing organization called Climate
45:25Change AI. It's a nonprofit that is
45:28dedicated to
45:29putting out white papers and doing
45:32research on all the different AI tools
45:34that could be used to help with fighting
45:36the climate crisis.
45:38And most pretty much all of their
45:40recommendations actually have nothing to
45:43do with generative AI. So, for example,
45:46they recommend optimization models to
45:49help better integrate renewable energy
45:51into the grid because you need to be
45:53able to predict how much renewable
45:55energy generation there's going to be
45:57when the sun shines, when the wind
45:58blows, and then And need to be able to
46:01figure out how to actually distribute
46:02that effectively among all the people
46:04that are demanding that energy. That's a
46:07problem that AI is perfect at and is
46:10just one little piece of the general
46:12resiliency climate change equation.
46:14>> Right. And that's task-specific AI.
46:16You're like
46:16>> That's task-specific AI.
46:18>> designed to do this.
46:19>> Yes. And we've also seen, you know, the
46:22Nobel Prize was awarded to a team that
46:25created AlphaFold at DeepMind. AlphaFold
46:30helped, it was also task-specific AI
46:32that helped with predicting protein
46:35structures from their sequences, which
46:37is really critical for drug discovery,
46:40for understanding disease. And so that
46:42was a really great advance in AI and
46:44health care that has nothing to do with
46:46generative AI. And I think we need to
46:48invest in more of these approaches by
46:50ultimately
46:51asking, what do we need as a society to
46:55live in a sustainable, equitable future?
46:59We need
47:00We need our rights. We need clean air.
47:01We need clean water. We need better
47:03health care, better education. We need
47:06to not have an environmental crisis.
47:08And then think about, well, how do we
47:11integrate any technology, not just AI,
47:13in service of that? Rather than suddenly
47:17ask how we serve technology.
47:20>> I have been thinking about the way
47:24even collective action. That's
47:27important, but I've been thinking about
47:28the way, how does the government get
47:29involved? And And the the toughest part
47:32with government, you know this, is if
47:33you look at Congress and Senate, it's a
47:35[ __ ] retirement home. I mean, Chuck
47:37Grassley's in his 90s. I think he just
47:40maybe, fingers crossed, knows what
47:42iMessage is.
47:44How can our government officials hold
47:47any technology company accountable when
47:50you have an analog government trying to
47:52compete with an AI revolution?
47:54>> We obviously have a huge
47:57vacuum of leadership at the top in the
47:59US right now. But the beautiful thing
48:00about democracy is you can also have
48:02leadership at the bottom.
48:04And we cannot actually wait around right
48:06now for policy makers and regulators to
48:08move.
48:10So we got to move.
48:11>> Mhm.
48:12The The scary thing that I think about
48:14all the time is
48:16when I read history, when you look at
48:18any company that has been able to
48:21acquire exorbitant amounts of wealth and
48:24deliver returns for shareholders,
48:26there's always what's written legally,
48:28here [snorts] the 12 rules, right?
48:31And they just find rule number 13 that
48:34euro steps
48:35>> Yeah.
48:36>> past what's legal. And once you add a
48:39new rule that's technically not illegal,
48:41you then conflate the legal
48:43with the ethical. Hey, I'm not breaking
48:44the law.
48:45So it's totally fine.
48:46>> Yeah.
48:47>> You clearly see that in finance. That
48:49happens all the time. They are masters
48:51of understanding what is legally allowed
48:53and just let's add an addendum or two
48:55that just works around that.
48:57>> How do we get ahead of that? How do you
48:59play defense against that?
49:01>> I mean, there's some really interesting
49:04case studies in history of this
49:07collective action helping to do that
49:10thing. Like when you talk about the
49:12fashion industry, I mean, there were
49:14some serious environmental and labor
49:16harms coming out of the fashion
49:17industry.
49:18>> Right.
49:18>> And it was None of it was illegal.
49:21But there was such a huge movement among
49:24consumers of Wait a minute, we don't
49:26want to buy clothes that are created in
49:29buildings that are collapsing on people
49:31and leaving leaving them dead.
49:32>> Right.
49:33>> We want to buy sustainable
49:36ethically sourced clothes where workers
49:39are paid what they're actually the value
49:42that they create. And it [clears throat]
49:43actually created entirely new markets
49:46for sustainable fashion, for ethically
49:48sourced fashion. Yeah. So the solution
49:50at the time wasn't like no one wear any
49:52clothes. The solution was to shore up
49:54the supply chain and create enough
49:55pressure that there are new markets born
49:58from the consumer demand.
50:00And I think there are you know, there
50:02are many many other examples of
50:04different industries that have led to
50:06that kind of transformation because
50:08people wanted better. They wanted better
50:10than the law.
50:11>> Are you um are you familiar with
50:13Cassandra from Greek mythology?
50:15>> Yeah.
50:15>> [laughter]
50:15>> Okay.
50:16So, if you if for those of you that
50:18aren't aware uh [laughter] in Greek
50:20mythology, a Cassandra Cassandra was a
50:22prophet uh whose prophecies always came
50:25true but were never believed by the
50:26people. [laughter] So, so she would tell
50:28you what's going to happen and it would
50:30fall on deaf ears.
50:32Do you feel like you are a Cassandra
50:34when it comes to AI?
50:34>> No. Actually, I've been amazed by how
50:38many people I talk to around the world
50:41who are like, "Oh, yeah. This is exactly
50:45what I'm feeling." And that has been
50:47amazing.
50:48>> This has been an amazing conversation,
50:51Karen. I loved chatting with you.
50:53Uh do you have any final thoughts that
50:55you want to leave our audience with? Uh
50:57but this has been so rad. Thank you for
50:59doing your work.
51:00>> Just that I'm your number one fan and
51:02everyone should continue watching us on
51:04Monage.
51:05>> Thank you. Thank you, Karen. Appreciate
51:06you being on the show.
51:08>> Thank you so much for having me.
51:09>> All right.
51:10Mom, you got a competitor right here.
51:13>> [laughter]
51:13>> Sorry, Simone.
51:15Okay. This was so lovely. Thank you.
51:17>> Thank you.
51:21>> [music]

About this transcript

This page contains the full transcript of Will AI Take My Job? with Karen Hao by Hasan Minhaj, generated from the public captions YouTube serves with the video. The transcript has 8,714 words across 1,434 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.