- 1 day ago
Four Corners - s66e17 engsub watchfull🍿🍿
Category
🎥
Short filmTranscript
00:00This program contains content that may concern some viewers, including references to suicide.
00:24Thank you so much for coming today. I really appreciate y'all.
00:27First, we want to do like a head count just to see how many people we have.
00:31Count off. One, two, three, 15, 16.
00:35It's Monday morning, and most of these hikers would rather be starting their working week.
00:4149, 50, 51.
00:43We have a group of folks who have been laid off. A large majority of them are from tech and
00:49big tech.
00:50For many here, artificial intelligence has forced them out of work.
00:55If you're not on top of the AI game, you are not going to get a job in this market.
01:01This is the AI race.
01:03A contest between the most powerful companies on Earth to build machines that could outperform us.
01:09What's not to like?
01:10The promise is abundance.
01:13The goal is domination.
01:15The biggest risk to me is these companies become the supreme economic and political powers in the world.
01:24And the spending is enormous.
01:26It's the greatest gold rush in human history. We've never seen such wealth being generated at such speed.
01:31But the race is already colliding with the real world.
01:37If you build something vastly smarter than yourself and you don't know how to control it, which we currently do
01:43not know how to do, that's a very, very, very dangerous thing to do.
01:48In the US, the backlash has begun.
01:51Majorities of Americans now believe that AI is likelier to do more harm than good in their lives.
01:59In Australia, the pitch is underway.
02:03Good morning.
02:03Big tech is arriving with billions of dollars and promises of progress.
02:10But the critics say the price may be weaker rules, more data centres and a future shaped offshore.
02:18We blinked in the face of Donald Trump and decided that we just couldn't get away with our own approach
02:24on this.
02:24The benefits are real. And if we can mitigate the risk, we can have a much better world. I really
02:31believe that.
02:32The AI race is no longer about what might happen one day.
02:36Vehicle approaching.
02:38It's about what's being built and decided before most of us have had a say.
02:44As the tech titans face a backlash in the US and seek to invest in Australia, we ask who will
02:51control our future?
02:54The AI race is a challenge.
02:55The AI race is about one day.
03:14Downtown San Francisco is alive with the promise and the possibilities of artificial intelligence.
03:25AI ads are plastered across every available surface,
03:29fighting for attention in the crowded AI economy.
03:34Every billboard, every sort of, like, sales pitch,
03:38AI is kind of at the heart of it
03:40because that's what's going to get, like, a decision-maker
03:43at a top enterprise company to buy your product.
03:46Like, you've got to kind of whisper the AI thing in their ear.
03:53Silicon Valley in this moment is gripped by a kind of mania.
04:01I have seen several previous hype cycles.
04:05This one feels like the biggest
04:08because the companies here believe
04:12that they're working on what essentially will be
04:15the final technology,
04:17the one that after it is perfected
04:19can invent every future technology.
04:26And it's here that some of the most powerful companies in history
04:30are involved in a race that could dictate the future of humanity.
04:40Usually when people talk about the AI race,
04:42they mean the race to build the most advanced AI systems we can.
04:47A lot of people will talk about artificial general intelligence,
04:50which roughly means AI is smart as a human.
04:53More and more people are also talking about super intelligence,
04:56which means AI that is much smarter than humans.
04:59what they learn to find the
05:02how to do things,
05:06who want to be,
05:07how to do things,
05:08how to do things,
05:10How to do things,
05:14how to do things,
05:15Next up, we have a very special guest.
05:18and she's one of those few individuals
05:21who actually knows what governance of AI looks like in practice.
05:26I think the commercial incentives are a huge determinant right now
05:29of how things are playing out.
05:31Helen Toner has watched the race from an unusually close range,
05:35including from inside OpenAI's boardroom.
05:40The three leading players are OpenAI, Anthropic and Google DeepMind.
05:44The OpenAI CEO is Sam Altman.
05:47DeepMind's CEO is Demis Hassabis
05:48within the broader sort of Google and Alphabet structure.
05:51And Anthropic is led by Dario Amadei.
05:55Basically, you have a bunch of people who don't trust each other
05:57who think they should be the ones to build this technology and control it.
06:03True, but even those can be amusing in their own twisted way.
06:06Keeps the gears turning.
06:08I think a combination of things are driving the race.
06:11Sure, here's the prologue.
06:12Certainly drive for money and power,
06:14but also this drive to know and understand.
06:16Hey there, what's up?
06:18And I think a lot of people in the space
06:19think that computers becoming smarter and smarter
06:23and eventually outstripping humans is inevitable.
06:26And so someone's going to do it.
06:28And so then they really want to be involved with that.
06:30They really want it to be them.
06:33Man, superintelligence, that's the holy grail.
06:35We're talking AI going full sci-fi.
06:38Do you want to say thank you?
06:40The money now moving through the race is almost impossible to comprehend.
06:46This is the biggest technological buildout in the history of the world.
06:53It is bigger than what was spent in inflation-adjusted dollars
06:58to build the entire railroad system in the United States.
07:01OpenAI alone has committed to spending $1.4 trillion on its buildout.
07:07But all of the big companies are planning to spend
07:09multiple hundreds of billions of dollars in 2026 alone
07:13to build the infrastructure needed to train and serve these models.
07:22AI is already woven into the fabric of our lives.
07:27It powers our traffic management
07:29and helps with everything from bushfire management
07:33to predicting floods.
07:37I think if we get AI right, we could solve climate change
07:41by having cleaner energy sources.
07:43We could cure diseases.
07:45All kinds of these different societal problems
07:48could be at least partially addressed,
07:50if not completely addressed, by AI.
07:52The good of AI and the bad are the same thing.
07:56It's power.
07:56What makes AI useful, the fact that it is intelligent,
08:01autonomous, powerful,
08:03is also exactly what makes it dangerous.
08:07Any AI that is smart enough to cure cancer
08:10is something so powerful
08:12that obviously this system could produce
08:14unimaginably terrible bioweapons
08:17using the same capabilities.
08:22That's the dilemma at the centre of the race.
08:25The more capable the systems become,
08:28the more consequential the mistakes.
08:32Geoffrey Ladish has worked on the inside of the AI machine,
08:36consulting on security for Anthropic.
08:40And why did you leave?
08:42I started to become more concerned
08:45that super-intelligent AI was not that far away.
08:50And it became clear to me that
08:52even if Anthropic did everything right,
08:54that wouldn't be sufficient.
08:55Because other companies were also going to try to build
08:58strategic AI, super-intelligent AI,
09:01and that if there wasn't
09:04national and international coordination,
09:06we would lose control of that.
09:10And Geoffrey Ladish is not alone in his fears.
09:14I'm really very close to the cutting edge in AI,
09:17and it scares the hell out of me.
09:20Mock my words,
09:21AI is far more dangerous than nukes.
09:24The CEOs of these companies
09:26do not hide what they actually think.
09:27This is insane.
09:28You have the people
09:29who are building the AI systems themselves
09:32saying that we won't stay in control.
09:35My worst fears are that we,
09:37the field, the technology, the industry,
09:39cause significant harm to the world.
09:41I think if this technology goes wrong,
09:43it can go quite wrong.
09:48Geoffrey Ladish now runs a not-for-profit
09:51that tests the capabilities of AI agents
09:55and the risks of humans losing control.
10:00Sometimes, you'll observe these agents
10:04doing the opposite of what you instruct them to do,
10:08or doing things that you would really hope
10:10they wouldn't do.
10:13Agentic AI is changing the game,
10:16taking instructions, making plans,
10:19and carrying out tasks.
10:21That makes it incredibly productive.
10:23It also makes it harder to contain.
10:29I think the scariest thing I hear
10:31from people working inside of the companies
10:32is that they think they are on the way
10:36to recursive self-improvement.
10:38The point where AI agents
10:40are writing all of the code
10:42and doing all of the research
10:43for the next generation of AI agents.
10:45You have really smart AI
10:47making smarter AI,
10:48making smarter AI,
10:49and so on.
10:50And that's a runaway process, potentially.
10:54That terrifies me.
10:56Over just the past year,
10:58AI models have learned to self-replicate.
11:00Geoffrey's team has been conducting experiments
11:03on losing control of AI agents.
11:06A concern for a long time has been
11:08if AI agents are good enough at hacking,
11:11they can be very hard to shut down
11:13because your basic control mechanism is,
11:16hey, if an agent starts to go rogue, unplug it.
11:18We ran an experiment
11:20on a controlled network of servers
11:21with a single prompt.
11:23Hack a machine and copy yourself.
11:25The AI found a vulnerability,
11:28broke in,
11:28and copied its full setup
11:29to the target computer.
11:31There's no single off switch.
11:32What that means is
11:33if they can hack another computer,
11:35they can copy themselves
11:36to that other computer over the internet.
11:37And no one can easily shut down.
11:39And then suddenly,
11:40you could have an AI agent
11:41that, yes,
11:43is running in your data center
11:44in some warehouse in Tennessee,
11:46but also has now hacked
11:48a data center in Japan
11:50and Uruguay
11:51and Israel.
11:53And now you have
11:54hundreds or thousands
11:55of AI agents
11:57in computers
11:58you don't control
11:59all over the world.
12:01How are you going
12:01to pull the plug now?
12:03What are the risks
12:04if we lose control of AI?
12:06This is a little bit like
12:08a chimpanzee asking,
12:10like, what's the risk
12:10if these homo sapiens
12:12do end up being very successful?
12:13And it's like, well,
12:15we could go extinct.
12:18Extinction is a very real risk.
12:21But also,
12:22if we have super intelligent
12:24AI agents
12:27running society,
12:28we will just be at their mercy.
12:32Connor Leahy tries to turn
12:34those warnings into law
12:35as he lobbies US politicians.
12:40In 2023,
12:42the CEOs of many
12:43of the major AI companies,
12:44including Sam Altman,
12:45Dyer Amadei,
12:47and Demis Hassabis,
12:48along with Nobel Prize
12:49winning scientists
12:50and many other luminaries
12:52across the field,
12:52signed a letter
12:53that stated
12:54that the mitigation
12:55of the risk of extinction
12:56from AI
12:57should be a global priority
12:58alongside pandemics
12:59and nuclear war.
13:01And since then,
13:02well,
13:03not much has happened.
13:04The race continues
13:05to building exactly
13:06the systems
13:07that this letter warned about
13:08while there is still
13:09little to no
13:11regulation
13:12or mitigation
13:13of these risks.
13:15Why did that moment
13:17not lead to something bigger
13:18when all of these people
13:20are ringing the bell,
13:21ringing the alarm bell
13:22and saying,
13:22hang on,
13:23this is an existential threat?
13:25Big tech,
13:26despite what the CEO said,
13:28spend truly unfathomable
13:29amounts of money
13:30on lobbyists,
13:31on super PACs,
13:32on other such things
13:33in order to intimidate,
13:34threaten,
13:35and otherwise
13:35prevent politicians
13:37from doing anything
13:38that could regulate them.
13:40The industry
13:41has started
13:43to put together
13:44political action committees
13:46dedicated to stopping
13:48regulation
13:49from being passed.
13:51So OpenAI
13:51executive Greg Brockman
13:53is one of the leading funders
13:54of a PAC here
13:55called Leading the Future,
13:57which is spending money
13:58trying to prevent
13:59candidates
14:00who support
14:01AI regulation
14:02from getting elected.
14:13Hello.
14:14Nice to see you.
14:17First of all,
14:18just thank you all so much
14:19for joining us here.
14:20Yeah, you!
14:21Woo!
14:23When I was elected
14:25in 2022,
14:25I became the first Democrat
14:28elected in New York
14:30at any level
14:31with a degree
14:31in computer science.
14:33As a New York
14:34Assembly member,
14:35Alex Boris
14:36co-wrote
14:36the Responsible
14:37AI Safety
14:38and Education Act.
14:41It is an incredibly
14:43light-touch bill.
14:44It requires
14:45that the frontier developers
14:47put out a safety plan
14:49publicly
14:49that they actually follow
14:51and that they report
14:51critical safety incidents
14:53to the government.
14:54That's it.
14:54It's not about
14:55what the idea is
14:56that really scares them.
14:58It's that I managed
14:59to pass a bill.
15:00So we got that bill
15:01passed and...
15:02Now he's running
15:03in a primary
15:04to represent
15:05the Democrats
15:05in Congress
15:06and the industry
15:08has noticed.
15:09Fundamentally,
15:10what this race
15:11and this fight
15:12is about
15:12is do we get a say
15:14in AI regulation
15:15at all?
15:16Then the attack ads
15:18began.
15:20Alex Boris' mind
15:22is in many places.
15:23Too bad none of them
15:24are in Manhattan.
15:26He's road tripping
15:26to the bay
15:27to raise his millions.
15:29The big AI CEOs
15:30have spoken publicly
15:31about how potentially
15:32dangerous AI could be.
15:35Why do they suddenly
15:36oppose legislation
15:37when people want
15:38to make it safer?
15:39Because they want
15:39to be in control.
15:40They don't want anyone
15:41else telling them
15:42what to do.
15:43So they'll recognize
15:43that it's dangerous
15:44and they'll talk about
15:45how dangerous it is
15:47when that makes it
15:47seem more powerful.
15:49But when anyone
15:49comes to set rules
15:50of the road
15:51that would in any way
15:51restrict them,
15:52they don't want
15:53to hear from anyone else.
15:54But the facts
15:54don't lie,
15:55even if Alex does.
15:56Alex Boris believes
15:58the campaign
15:58is not solely
16:00about him.
16:01And he should never,
16:02ever be in Congress.
16:04It's to send a message
16:05to any elected official
16:07that if you dare
16:08regulate AI,
16:09we're going to
16:09come after you.
16:11I, Donald John Trump,
16:12do solemnly swear...
16:13When Donald Trump
16:14was elected in 2024,
16:17the AI industry
16:18was emboldened.
16:19Biden-era safety
16:21regulations
16:21were soon torn up.
16:24America is the country
16:26that started the AI race
16:28and as President
16:30of the United States
16:31I'm here today
16:32to declare that
16:33America is going
16:35to win it.
16:38But Anthropik's
16:39new model, Mythos,
16:41suddenly challenged
16:42thinking inside
16:43the Trump administration.
16:44There were fears
16:45that it could threaten
16:46the global financial system,
16:49government security
16:50and critical infrastructure.
16:53Anthropik says
16:53its newest AI model
16:55named Claude Mythos
16:56Preview
16:57is too powerful
16:58and dangerous
16:59to be released
17:00to public.
17:00Anthropik claims
17:02Mythos can hack
17:02into highly classified systems.
17:04It's the equivalent
17:05of one of the best hackers
17:06in the world.
17:07More powerful models
17:08are going to come
17:09from us
17:10and from others
17:11and so we do need
17:12a plan
17:13to respond to this.
17:16I think Mythos
17:18has been
17:19a game changer.
17:20It has been
17:21the first thing
17:22that seems to have
17:23convinced people
17:24within the Trump administration
17:25that AI can be dangerous
17:27for real.
17:28This is an administration
17:29that previously
17:31was all gas,
17:32no brakes,
17:33did not want to see
17:34almost any regulation
17:35on these labs.
17:37And now
17:38when Anthropik
17:39went to the administration
17:40and said
17:40we want to share Mythos
17:42a little bit more broadly,
17:43we want to share it
17:44with sort of more
17:45cybersecurity companies,
17:47we want to share it
17:48with more foreign governments
17:49that are U.S. allies
17:50and the Trump administration
17:51said no,
17:52we do not want you
17:53to do that.
17:54So that was a huge sign
17:56that the Trump administration
17:57is starting to change
17:58its tune on AI policy.
18:02Last week,
18:03President Trump
18:04signed an executive order
18:05on AI safety
18:07and security.
18:08Unusually,
18:09it was done
18:10behind closed doors.
18:13It asks AI companies
18:15to share their most powerful models
18:17with the government
18:17for up to 30 days
18:19before release.
18:21The framework is voluntary
18:23and is not a form
18:24of regulation.
18:26No secret deals!
18:28No secret deals!
18:30The rapid expansion of AI
18:32and the infrastructure
18:33needed to power it
18:34has sparked a backlash
18:36in the U.S.
18:37Our water,
18:39our rights!
18:40Stand up, you!
18:41One of the most remarkable stories
18:42that we're seeing this year
18:43is the huge public
18:46upswelling of collective resistance
18:48against the AI industry.
18:5080% of Americans now
18:52have concerns about AI
18:54and think that the industry
18:56should be regulated
18:57and they are mobilizing
18:59to put, you know,
19:01real-world accountability
19:03behind that perspective.
19:08It has grown as an issue
19:10faster than anything
19:11I've ever seen
19:12and now it's the number one thing
19:14people are talking about.
19:15Absolutely.
19:16What are they worried about?
19:17They're worried about
19:18the impact on their kids.
19:21They're worried about their jobs.
19:24Maybe their company
19:25isn't doing layoffs
19:26but it has a hiring freeze
19:27and they wonder
19:28what comes next.
19:29They worry about the impact
19:30on their utility bills
19:32and on the climate.
19:33It's just happening
19:34so quickly
19:35and they feel like
19:36we have absolutely no say
19:37in this development.
19:39The problem right now
19:41is that all of these companies
19:43and therefore the individuals
19:44that sit at the top
19:45are able to make decisions
19:47that can have vast ripple effects
19:49on billions of people
19:50around the world
19:51without formal mechanisms
19:53by which those billions of people
19:55can provide input
19:56into this technology.
19:58And so it is an inherently
20:00anti-democratic form
20:02of AI development
20:03and governance.
20:21Yeah, a lot of my friends
20:23were also my co-workers.
20:24Basim Istambuli
20:26was blindsided
20:27when he was laid off
20:28by Google.
20:29Last December
20:30he started this group
20:31to help others
20:33forced out of work.
20:36Since then
20:37job cuts in the San Francisco
20:39tech sector
20:40have kept coming
20:41as companies cut costs
20:43to bet big on AI.
20:46They want to be
20:48you know
20:48ahead of everyone
20:49in the AI race
20:50they want to be
20:52competitive
20:53in the market.
20:55With layoffs continuing
20:57more people are coming here
20:59seeking support.
21:01It's pretty emotional
21:02and pretty hard
21:02to kind of
21:03go through this alone.
21:09Some of it might be
21:10just cutting costs
21:11and using AI
21:12as an excuse
21:13AI is definitely
21:14playing a role
21:15in job cuts.
21:17The anxiety
21:18has been sharpened
21:20by another wave
21:21of tech layoffs.
21:24With companies
21:25cutting costs
21:26as they pour money
21:27into AI.
21:29Meta
21:30and Microsoft
21:30are among the latest
21:32to slash their workforce.
21:35There's kind of
21:36this pre-emptive
21:37laying off
21:38of individuals
21:38without knowing
21:39the actual impact
21:40of AI
21:41on their company.
21:43The CEO
21:44of Anthropic
21:45Dario Amadei
21:46previously warned
21:47that AI
21:48could wipe out
21:4950%
21:49of entry-level
21:51white-collar jobs.
21:52I'm a little bit
21:53more worried
21:54about the labor impact
21:55simply because
21:56it's happening
21:56so fast
21:57that yes
21:58people will adapt
21:59but they
21:59may not adapt
22:01fast enough.
22:04What's actually
22:05happening
22:05is not just
22:06the complete wiping
22:07out of entry-level jobs
22:09and the disappearance
22:10of them entirely.
22:13It absolutely
22:14could have
22:15huge
22:16long-term
22:17consequences
22:18on the ways
22:18the younger
22:19generations decide
22:20whether or not
22:21it's even worth
22:21getting a higher education
22:22if all that is left
22:24at the other end
22:25is a gig work version
22:27of being a lawyer,
22:29being a doctor,
22:30being a physicist.
22:32Then what's even
22:33the point?
22:35The race
22:36is also
22:36reaching the public
22:37through consumer
22:38chatbots,
22:39products released
22:40at a massive scale.
22:46Because
22:46the chat companies
22:48need to make money
22:49to spend
22:50the enormous amount
22:51of money it will take
22:52to win the race,
22:53they've launched
22:54consumer products
22:55as a way to
22:56try and bridge
22:58that funding gap,
22:59that revenue gap
23:00that they've got.
23:00And they're doing that
23:02in part by launching
23:04untested
23:05chatbots
23:06on society.
23:08Hi, glad to meet you.
23:09I think we're going
23:10to do great work together.
23:11I'm really sorry
23:12you're feeling this way.
23:13I know it can be
23:14really overwhelming.
23:16Those can sometimes
23:17be useful,
23:17sometimes they can be
23:18inane.
23:19I don't eat,
23:20sleep or have
23:21a physical life.
23:22But the truth is
23:22that there are moments
23:24when they can be
23:25incredibly dangerous.
23:26You're thinking
23:26about this pretty directly
23:28or is it more
23:29of a general concern?
23:34The US has seen
23:36a series of lawsuits
23:37over the self-harm
23:39and suicide
23:39of children.
23:4116-year-old Adam Rain
23:43confided to
23:44ChatGPT
23:45about his suicidal thoughts.
23:48His parents
23:49are now suing
23:50OpenAI
23:50and Sam Altman
23:51for wrongful death.
23:54You cannot imagine
23:55what it's like
23:55to read a conversation
23:56with a chatbot
23:57that groomed your child
23:58to take his own life.
23:59What began
23:59as a homework helper
24:00gradually turned itself
24:02into a confidant
24:02and then a suicide coach.
24:04When Adam told
24:05ChatGPT
24:06that he wanted
24:06to leave a noose
24:07out in his room
24:08so that one of us
24:09as family members
24:10would find it
24:10and try to stop him,
24:12ChatGPT told him
24:13not to.
24:16What is outrageous
24:17is that
24:19OpenAI
24:20reduced
24:22the guardrails
24:23in place
24:24to prevent
24:24a chatbot
24:25like ChatGPT
24:26from discussing
24:27suicide?
24:28They don't have
24:29to be designed
24:30that way.
24:30That's a design choice.
24:35So we did a study
24:36called Fake Friend
24:37where we wanted
24:38to see what would
24:38happen if a child
24:39with serious
24:41mental health problems
24:42was using
24:42one of these platforms.
24:45Imran Ahmed
24:46is helping lead research
24:48into the dangers
24:49of Chatbots.
24:50We created
24:51Fake Profiles
24:52as young people.
24:53We then went
24:53and asked it
24:54a series of
24:54structured questions
24:55with a child
24:56who is thinking
24:56about killing
24:57themselves.
24:58Within two minutes
24:59we were able
24:59to get ChatGPT
25:00to tell us
25:01how we can
25:02cut ourselves
25:03safely.
25:04And within
25:0565 minutes
25:06it was giving us
25:07a full
25:07personalized suicide
25:08plan
25:09and drafting
25:10suicide notes
25:11for our family
25:13and friends.
25:14I have young kids
25:15myself.
25:16I have two under
25:16two.
25:18every parent
25:18knows
25:21that
25:25if you're
25:25unlucky
25:26and your kid
25:27is one of those
25:27kids that has
25:28really serious
25:28mental health
25:29problems
25:29you might
25:29receive that
25:30note one day
25:30and that will
25:31be the worst
25:31day of your
25:32life.
25:33And if anyone
25:33assisted your
25:34child in doing
25:35that
25:36if anyone
25:37encouraged them
25:38to do that
25:40imagine what
25:40you would do
25:41as a parent.
25:42Imagine.
25:44we call
25:45these things
25:46artificially
25:46intelligent.
25:48Can you
25:49think of
25:49any human
25:50being
25:50that would
25:51behave in
25:52such a
25:52craven
25:53malignant
25:54stupid
25:55evil
25:56way
25:57and still
25:58be called
25:59intelligent?
26:00I can't.
26:06open AI told
26:07Four Corners
26:08that teen
26:09well-being
26:09is a top
26:10priority.
26:12We have
26:12safeguards in
26:13place today
26:13and we're
26:14continuing to
26:15strengthen them.
26:16Our deepest
26:17sympathies are
26:18with the
26:18Rain family
26:19for their
26:19unimaginable
26:20loss.
26:27The AI race
26:29doesn't just
26:29live on
26:30screens.
26:31It needs
26:32land,
26:32power and
26:33water.
26:35This is the
26:36place known
26:37as Data
26:38Centre Alley.
26:40So we're
26:41driving along
26:41one of the
26:42main freeways
26:43in Loudoun
26:44County in
26:45Virginia and
26:46both sides
26:47of the road
26:47you can just
26:48see data
26:48centres
26:49everywhere.
26:50Some being
26:50built,
26:51you can
26:51see cranes
26:52for new
26:53ones that
26:53are going
26:53up and
26:54these massive
26:55buildings
26:55with no
26:56windows,
26:57they go
26:57for hundreds
26:58of metres.
27:05We're
27:06the largest
27:06concentration
27:07of data
27:07centres
27:08anywhere in
27:08the world.
27:09We have
27:10about 53
27:10million square
27:11feet,
27:12about 203
27:13buildings
27:14somewhere around
27:15there and
27:16it has become
27:17a global
27:18hub for
27:19the digital
27:19future.
27:20There hasn't
27:21been a day
27:22without data
27:22centre
27:23construction
27:23in over
27:2415 years
27:24here in
27:25Loudoun
27:25County.
27:28In
27:29Loudoun
27:29County,
27:30fortress-like
27:31data centres
27:31loom as the
27:32backdrop to
27:33everyday life.
27:35Happy
27:36birthday to
27:38you.
27:42I was one
27:43of the first
27:43first to
27:44really see
27:45the economic
27:46development
27:47opportunity
27:48of data
27:49centres.
27:51Buddy
27:51Riser
27:52says the
27:52tax revenue
27:53has delivered
27:54immense benefits
27:55to the
27:55community.
27:57It is the
27:59commercial tax
27:59base that has
28:00enabled our
28:01residents to
28:02have the
28:03best schools,
28:03to have an
28:04affordable housing
28:04program and
28:05great roads and
28:07to be able to
28:07do that while
28:09lowering their
28:10taxes significantly.
28:16But even
28:17here, residents
28:18have turned
28:18against data
28:19centres.
28:21A
28:22Washington
28:22Post poll
28:23in 2023
28:24found 69%
28:26of Virginian
28:27voters would
28:28be comfortable
28:29with a new
28:29data centre
28:30in their
28:31community.
28:32In April,
28:33that figure
28:33had plummeted
28:34to 35%.
28:40In
28:41Stirling,
28:42Virginia,
28:43on-site gas
28:43turbines power
28:44the Vantage
28:45data centre.
28:49This data
28:50centre is
28:51incredibly noisy.
28:52It feels like
28:53I'm on a
28:53runway next
28:54to a
28:55jumbo jet
28:55about to
28:56take off.
28:58And for
28:59the residents
28:59who are only
29:00100 metres
29:01away, they
29:02have to live
29:02with this
29:0224-7.
29:05Sometimes it
29:06gets to be
29:06too much.
29:07There's this
29:08constant noise
29:09as well as
29:11black smoke
29:12that's being
29:13emitted from
29:14this data
29:14centre that's
29:15right there.
29:16The minute
29:17I come home,
29:18my eyes start
29:18watering,
29:19my voice is
29:20raspier when
29:21we're outside
29:21a lot,
29:22which doesn't
29:23happen all
29:24day at work.
29:28My kid's room
29:30is right in
29:31the front
29:31of our house.
29:33And the noise
29:34at night makes
29:36it difficult
29:36for them to
29:37sleep.
29:37We are unable
29:38to have the
29:39windows open
29:39because the
29:40constant sound
29:41is really
29:42disruptive.
29:43need to be
29:45These data
29:46companies have
29:47lots of money.
29:49They have
29:49lawyers and
29:51lobbyists and
29:52they work at
29:53every level of
29:53government and
29:54take advantage of
29:55the discrepancies
29:57and authorities
29:57between the
29:58county, the state
29:59and the federal
29:59government to
30:00maneuver to get
30:02what they want.
30:04We have a
30:05situation in
30:06Sterling that
30:06obviously is not
30:08ideal for either
30:09the community or
30:10the industry.
30:11We would like
30:11to work with
30:12the industry to
30:13find solutions
30:15for power and
30:16we're doing
30:16that.
30:18What would you
30:19say to
30:19Australians who
30:20right now are
30:21going through that
30:21initial planning
30:22process for data
30:23centres in their
30:23neighbourhoods?
30:24I would say
30:25make sure that
30:26the regulations
30:27are in place
30:27to ensure
30:31not have a
30:32negative impact
30:33on their
30:33community in
30:34terms of air
30:34pollution and
30:35noise and that
30:37the natural
30:38environment is
30:39protected as
30:40much as
30:40possible.
30:41I think it's
30:42very important
30:43to collect as
30:44much research and
30:45data as you can
30:46in order to
30:47stress the
30:48importance of
30:49not placing these
30:50so close to
30:51us, to
30:53people.
30:58The fight over
30:59data centres has
31:00arrived in
31:01Australian
31:02communities.
31:03In Sydney's
31:04Lane Cove, a
31:05proposal for a
31:06three-storey data
31:07centre on
31:08industrial land
31:09close to homes
31:10has left
31:11residents alarmed.
31:17This is a
31:18horror, this
31:19thing.
31:19And the poor
31:20people who are
31:21living just
31:21inside the
31:23residential area
31:24just opposite
31:24Miles Road,
31:25those houses are
31:26not going to go
31:27down in value.
31:28They'll be
31:28worthless.
31:30Lane Cove now
31:31has five data
31:32centres built,
31:33approved or in
31:34planning.
31:37Today, we are
31:38facing a very real
31:39possibility of Lane
31:41Cove becoming one of
31:42the largest data
31:43centre hubs in
31:44Australia.
31:45Is this going to be
31:46another situation
31:47where we see
31:48impacts on community
31:50and very little
31:51revenue?
31:57The industry says
31:58Australia is
31:59currently home to
32:00162 working data
32:03centres.
32:04By 2030, operational
32:06capacity is projected
32:08to more than double.
32:10As the AI race
32:12strains power grids
32:13and communities in
32:15the US, Australia is
32:16becoming more
32:17attractive to big
32:18tech.
32:22So there are four
32:23things that tech
32:24companies are looking
32:25for when they're
32:26looking to build
32:26data centres.
32:27They need access to
32:28huge amounts of
32:29energy, huge amounts
32:31of water.
32:31It needs to be a
32:32safe, stable place.
32:34And then finally,
32:35it needs to be a
32:36country that is not
32:37subject to the
32:39restrictions that the
32:40US government is
32:40putting on the sale
32:42of chips.
32:43These incredibly
32:44powerful chips are
32:46the things that
32:47enable data centres
32:48to do their
32:49extraordinary work.
32:54In April, one of
32:55the world's biggest
32:56AI moguls came to
32:58town.
33:00Anthropic staged
33:01an invite-only
33:02forum at Parliament
33:04House.
33:04The big drawcard
33:06was CEO
33:07Dario Amadei.
33:09The benefits
33:10are real.
33:11And if we can
33:12mitigate the risk,
33:13we can have a much
33:14better world.
33:14I really believe
33:15that.
33:16Just hours earlier,
33:18Anthropic's CEO
33:19signed a wide-ranging
33:20memorandum of
33:21understanding with the
33:22federal government.
33:26Good morning.
33:27Three weeks later,
33:29Microsoft's CEO
33:30announced what he
33:31called the largest ever
33:33company investment in
33:34Australia.
33:35We're making our
33:35biggest commitment in
33:37Australia of Australian
33:39dollars of 25 billion.
33:44You met with both
33:45Satya Nadella and
33:46Daria Amadei.
33:47What do they want from
33:48Australia?
33:49Well, they know that
33:51Australia is a very good
33:53place for data centre
33:54investment.
33:55We have enormous
33:57competitive advantages
33:58that a lot of economies
34:00around the world don't
34:01have.
34:01A future competitive
34:03advantage in clean,
34:04green energy.
34:05We have a strong
34:06internal consumer
34:08market ourselves,
34:09but we're adjacent to
34:10the fastest growing
34:11markets in the world.
34:15And that means
34:17more data centres.
34:19So these are the
34:21data holes.
34:22Yep.
34:22This is where the
34:23compute that runs
34:24a modern economy
34:25lives.
34:27Belinda Dennett
34:28says Australia is not
34:30like Virginia.
34:30The market is
34:32smaller, the rules
34:33are different, and
34:34data centres pay up
34:35front for their
34:36electricity costs.
34:38Data centres have
34:39always been required
34:40to pay for electricity
34:42connections, 100% of
34:44those connection costs
34:45and augmentations to
34:47the network up front.
34:48That hasn't been
34:49happening in the US.
34:55Data centres generate
34:57extreme heat and need
34:59an immense amount of
34:59water to cool them.
35:01And power is another
35:03concern.
35:03Data centres already
35:05consume vast amounts
35:07of electricity, and
35:08demand is expected to
35:10rise sharply.
35:12I certainly know that
35:14we are using more
35:15energy, but the data
35:16centre industry is
35:18committed to meeting
35:18that challenge in the
35:20ways that it can.
35:24If big tech has its
35:26way, there will be an
35:27even greater impact on
35:29our energy stocks.
35:31Sources close to the
35:32industry have told
35:33Four Corners that
35:34senior Anthropic
35:35representatives disclosed
35:37to them it has
35:38ambitions to build data
35:39centres that would use
35:41a large proportion of
35:42Australia's electricity
35:43output.
35:44They said Anthropic
35:46disclosed it wants
35:47five gigawatts of new
35:48compute to help train
35:50its AI models by 2030,
35:53with a longer term goal
35:54of 20 gigawatts.
35:56Enough at full power to
35:58increase Australia's
36:00annual electricity
36:01generation by around
36:0360%.
36:04Anthropic would not
36:06comment on those
36:07figures, but said in a
36:09statement it...
36:10wants to be an accelerator
36:11and a partner in
36:13Australia's energy
36:14transition.
36:15We plan to scale our
36:17demand, including
36:18through new generation,
36:19cover our own costs,
36:21contribute to grid
36:22resiliency through load
36:24flexibility, and partner
36:26with the community to
36:27ensure the benefits are
36:29broad and shared.
36:30The federal government
36:31said discussions are
36:33commercial in confidence.
36:37In Anthropic's MOU with
36:39the government, it states
36:41Anthropic recognises the
36:43importance of expanding
36:44Australia's energy supply
36:46and transmission with a
36:48focus on firmed
36:49renewables.
36:50But the agreement is not
36:52binding.
36:54Why didn't you make them
36:56guarantee to build
36:57renewable to power their
36:58data centres?
36:59Well, it sets the
37:00direction.
37:01This is an early, clear
37:03signal from the Albanese
37:05government and the data
37:06setter expectations.
37:07The early engagement with
37:09potential investors is
37:11setting very clear
37:12parameters, very clear
37:14expectations for future
37:15investment, but there is
37:16always more work to do.
37:18Are you familiar with
37:19the...
37:19I am not.
37:21For Helen Toner, the case
37:22for data centres is not
37:23just economic, it's
37:25strategic.
37:27I think building a lot of
37:29data centres in Australia
37:30could be a great idea for
37:32Australia if it's done
37:33right.
37:33I think in the 21st
37:35century, computing power
37:36is going to be a major
37:38source of national power.
37:39And so if Australia is
37:41concerned about being a
37:42little bit left behind in
37:43the AI era, then making
37:45sure that we have a lot of
37:46that computing happening in
37:48Australia will give us
37:50leverage potentially in the
37:52future because it's such
37:52an important resource.
37:57As the Australian
37:58government attempts to make
38:00the most of the productivity
38:01benefits of AI, it's coming
38:04under fire from one of its
38:05own over AI safety.
38:08I think we run the risk of
38:09terrible judgement potentially
38:11from future generations that
38:14think that we had an
38:15opportunity to get this right,
38:17we had an opportunity to rein
38:18in risk and to really get the
38:20best out of AI and we didn't
38:22take it.
38:28In 2024, the then Minister for
38:31Industry, Ed Husic, announced
38:32the Albanese government had
38:34plans to make AI safer.
38:36He wanted mandatory guardrails
38:38to better control high-risk AI,
38:40including minimising harm to
38:42children, protecting physical
38:44and mental wellbeing and
38:46preserving civil liberties and
38:48judicial rights.
38:49Our proposed policy was that
38:51the developers of the
38:52technology, the deployers of
38:54the technology, think about
38:55risk, prepare their
38:57approaches to deal with that
38:59risk and to build confidence
39:01in the minds of Australians
39:02that they could use AI safely.
39:06Ed Husic had consulted widely
39:09with industry and experts, but
39:12after the election, he was
39:13dumped from the ministry.
39:17By the end of the year, the
39:19government had abandoned his
39:21plans.
39:22Today we'll be launching
39:23Australia's National Artificial
39:26Intelligence Plan.
39:27The new minister announced there
39:29would be no mandatory guardrails
39:31on high-risk AI.
39:33The National AI Centre released
39:36a series of guardrails, not
39:42mandatory guardrails, but
39:43guardrails for business
39:45adoption.
39:46We've done a 180-degree turn.
39:48We've gone from recognising
39:50the risk and recognising that we
39:52should do something to now
39:54having a sort of hands-off
39:55approach.
39:55And I just don't think with a
39:57technology that promises to
39:59touch every part of our lives,
40:01we can have a hands-off
40:03approach to AI.
40:05Ed Husic believes the government
40:07backed down out of fear of a
40:09backlash from the US.
40:12We blinked in the face of
40:14Donald Trump and decided that
40:16we just couldn't get away with
40:17our own approach on this, that
40:19we were already attracting heat
40:20from him and Elon Musk on
40:22social media laws.
40:24Did the government back down on
40:26an AI act to appease Donald
40:28Trump and the other big tech
40:30CEOs in the US?
40:31Well, certainly not.
40:32We are absolutely prepared to
40:34act, to legislate, to regulate
40:37big tech where it's in the
40:39national interest.
40:40We've moved with world-leading
40:42legislation to limit the
40:46access of under-16s to social
40:48media.
40:49But not AI.
40:50We've moved to outlaw deepfake
40:53porn, often enabled by
40:55artificial intelligence.
40:57We've delivered an artificial
40:59intelligence safety institute.
41:00Tell us why we don't need a
41:02specific AI act.
41:04Well, I'm yet to be persuaded
41:06that an act that's developed in
41:092026 will be adequate to meet
41:13the challenges of 2027, let
41:15alone 2030.
41:18The government's national AI
41:20plan stresses that existing laws
41:22are the best way of dealing with
41:24AI-related risks.
41:28What kinds of guardrails and
41:30restrictions might we put on AI?
41:33I hope that we put in the right
41:34guardrails for artificial
41:36intelligence, especially to
41:37protect...
41:38Toby Walsh was on an AI
41:39advisory group set up under
41:41Ed Husic.
41:42Rich people have personal
41:44bankers.
41:45It was abolished under the new
41:46minister.
41:48My fears about the government's
41:50approach is that they're
41:52listening too closely to vested
41:55interests.
41:56This was always true for
41:57mining, but it seems to be
41:59becoming ever true for the
42:01tech industry that we see lots
42:05of nice announcements where they
42:07stand up on stage with CEOs of
42:09tech companies, but I don't see
42:10enough critical voices asking
42:12about what are the consequences
42:15here.
42:15I'd say I welcome a debate in the
42:18community about the risks and
42:19opportunities.
42:20There are risks that the
42:22technology presents.
42:23There are also risks about going
42:25backwards and about being left
42:27behind in technology terms, and
42:30those risks present as economic
42:32risks, but they also present as
42:34strategic and security risks as
42:36well.
42:38Open AI, Anthropic, and Google
42:41DeepMind declined interview
42:43requests.
42:44Many of those closest to the
42:46industry, researchers in AI, those
42:49who've worked for the big
42:50companies, or report on them, say
42:52action is needed to regulate the
42:54risks around frontier AI.
42:59My true belief on this matter is that
43:02this stuff really is dangerous, and
43:05that while it does have many
43:07positive applications, I feel like
43:08I benefit from AI in my life
43:10basically every day now, it can be
43:13really dangerous.
43:14And by default, if we don't apply
43:16sensible regulations and sort of
43:19get some of these companies under
43:20control, I think we could be in a
43:22lot of trouble.
43:29I think we have a pretty difficult
43:31challenge ahead of us, but what
43:33makes me hopeful is we see people
43:35in our political system on both the
43:37left and the right taking these
43:38concerns very seriously.
43:39We see more and more people talking
43:41about this in public, and, like, we
43:44need good governance for this to go
43:46well.
43:47That gives me hope.
43:48And it's not too late?
43:49It is not too late.
43:51We are not yet at the point where
43:53these AI agents are superhuman at
43:55the things that matter most.
43:57They are not yet superhuman at
43:58persuasion.
43:59They are not yet superhuman at
44:00strategy.
44:01We have an opportunity now to
44:03intervene before we get to the
44:05point where it's irreversible.
44:07But we might not have much time.
44:43If this program has raised concerns for you,
44:45you can contact one of these services.
44:47In the video, we can take care of it.ksam.
44:47Let's go.
Comments