Inside the shadowy global battle to tame the
world's most dangerous technology
Can anyone control AI?
By Mark
Scott,
Gian
Volpicelli,
Mohar
Chatterjee,
Vincent
Manancourt,
Clothilde
Goujard
and Brendan
Bordelon
MARCH 26,
2024 4:00 AM CET
https://www.politico.eu/article/ai-control-kamala-harris-nick-clegg-meta-big-tech-social-media/
On a wet
November afternoon, U.S. Vice President Kamala Harris and Meta’s Nick Clegg
trudged into a large tent in the grounds of a 19th century English country
house north of London, took their seats at a circular table, and, among other
things, set about trying to save humanity.
Amid the
gloomy weather at Bletchley Park, which had been used as the Allied World War
II code-breaking headquarters, Clegg and Harris joined an elite gathering of
global leaders, academics and tech executives to tackle what pessimists fear is
a devastating new threat facing the world: runaway artificial intelligence.
The
politicians and tech executives agreed on a joint declaration of good
intentions after two days of talks, but no unified answer on what should be
done. Instead, they made rival pitches on how to manage a technology that will
dominate much of the next decade — and will likely upend everything from
business and health care to democracy itself.
Clegg, a
former British deputy prime minister, argued that policing AI was akin to
building a plane already in flight — inherently risky and difficult work.
Harris trumpeted Washington’s efforts to address the dangers of AI through
voluntary business agreements as the world’s gold standard. Ursula von der
Leyen, the European Commission president, who was also in attendance, urged
others to follow Brussels’ new, legally binding rulebook to crack down on the
tech.
The debate
represented a snapshot of a bigger truth. For the past year, a political fight
has been raging around the world, mostly in the shadows, over how — and whether
— to control AI. This new digital Great Game is a long way from over. Whoever
wins will cement their dominance over Western rules for an era-defining
technology. Once these rules are set, they will be almost impossible to
rewrite.
For those
watching the conversation firsthand, the haggling in the British rain was akin
to 19th-century European powers carving up the world.
“It felt
like an alternate reality,” said Amba Kak, head of the AI Now Institute, a
nonprofit organization, who participated in the discussions. At the end of the
meeting, 29 countries, including China, European Union members, and the United
States, signed a voluntary agreement to reduce risks that have climbed the
political agenda thanks to the arrival of OpenAI’s ChatGPT.
In the year
ahead, the cut-throat battle to control the technology will create winners and
losers. By the end of 2024, policymakers expect many new AI standards to have
been finalized.
For this
article, POLITICO spoke to more than three dozen politicians, policymakers,
tech executives and others, many of whom were granted anonymity to discuss
sensitive matters, to understand the dynamics at play as the world grapples
with this new disruptive technology.
The
question they face is whether the U.S., the EU or the United Kingdom — or
anyone else — will be able to devise a plan that Western democracies can agree
on. If liberal industrialized economies fail to reach a common regime among
themselves, China may step in to set the global rulebook for a technology that
— in a doomsday scenario — some fear has the potential to wipe humanity off the
face of the Earth.
As Brussels
and Washington pitch their conflicting plans for regulating AI, the chances of
a deal look far from promising.
Jousting in Japan
A month
before the conference in the English rain, policymakers had been frantically
trying to make progress on the other side of the world. It was October, and
Věra Jourová stepped off her 16-hour flight from Brussels to Japan exhausted.
The Czech politician was only a few weeks into her new role as the EU’s top
tech envoy, and her first international assignment would not be easy.
Jourová’s
mission was to sell Europe’s AI rulebook at a G7 meeting where Western leaders
had gathered to try to design new global standards for the most advanced form
of this technology, known as “generative AI,” the powerful development behind
ChatGPT and rival tools.
Brussels’
approach is enshrined in the EU’s Artificial Intelligence Act, the world’s
first attempt at binding legislation on the issue. Unlike the stance favored by
the U.S., the EU’s vision includes bans on the most invasive forms of the
technology and strict rules requiring companies like Google and Microsoft to be
more open about how they design AI-based products.
“Generative
AI invaded our lives so quickly and we need something fast,” Jourová told
POLITICO after taking the two-hour bullet train from Tokyo to Kyoto for the
summit.
At the
three-day meeting in Japan, Nathaniel Fick had a rival pitch.
As Joe
Biden’s top digital diplomat, Fick, a former tech executive, proposed no bans
or strict requirements. Instead, he pushed for a lighter-touch regime based
mostly on voluntary commitments from industry and existing domestic laws.
“People can
expect the United States to weave in AI policy issues in everything we do,”
Fick told POLITICO after the Kyoto summit concluded. “The frameworks, the
codes, the principles we develop will become the basis for action.”
In dueling
meetings with G7 policymakers, tech company executives and other influential
figures, Jourová and Fick made their cases.
For the
EU’s Jourová, the pitch was simple. Brussels had already marked itself out as
the West’s digital police officer, with a flurry of regulations on everything
from protecting consumers’ privacy to taming social media.
The summit
timetable was packed, with little downtime beyond some snatched cigarette
breaks and rushed lunches in the cafeteria. Jourová argued that only Europe
could deliver the necessary rigor. The EU, she said, could hit companies with
blockbuster fines and ban the most invasive forms of AI — such as social
scoring, which are complex algorithms tracking people’s movements, infamously
used in China.
“She came
with a plan, and that was to convince us Europe’s rules were the only game in
town,” said one of the people who met Jourová. Another official from a G7
country said Europe’s digital chief dismissed Washington’s alternative proposal
for its lack of binding legislation.
Fick’s
counteroffensive relied on America’s undisputed position as the world’s
powerhouse of AI development.
A political
stalemate on Capitol Hill means no comprehensive legislation from Washington is
likely to come anytime soon. But the White House has already made a flurry of
announcements. Last July, the Biden administration secured voluntary
commitments from leading tech giants like Amazon to make AI safer. Then Biden
issued an executive order, announced on Oct. 30, which empowered federal
agencies to act.
Fick’s
pitch included criticizing Brussels’ legislation for imposing too many
regulatory burdens on businesses, compared to Washington’s willingness to allow
companies to innovate, according to two individuals who met Fick in Japan.
“The
message was clear,” said another Western diplomat who attended the Kyoto G7
summit. “Washington wasn’t going to let Brussels get its way.”
Fine dining before the apocalypse
Back in
Europe, efforts to reach a consensus were continuing — in some unlikely
surroundings.
A lavish
six-course dinner at the Elysée Palace, the French president’s 18th-century
official residence in Paris, doesn’t exactly scream high-tech.
But over a
five-hour meal of gourmet cuisine and fine wine in November, Emmanuel Macron
held court with 40 industry experts, including OpenAI President Greg Brockman
and Meta’s chief AI guru, Yann LeCun.
The dinner
conversation quickly turned to rulemaking.
Macron is a
global powerbroker. France has a large domestic AI industry and the president
is personally eager to lead international efforts to set global rules for the
technology. Unlike stereotypes of regulation-loving French politicians, Macron
remains wary of Brussels’ AI Act, fearing it will hamstring firms like Mistral,
an AI startup co-founded by Cedric O, France’s former digital minister.
As Macron
listened carefully, his dinner guests laid out their rival proposals. The
discussion exposed another key facet of the global AI debate: the clash between
those who believe the gravest risks are still many years away, and those who
think they are already here. This is the battle at the heart of the entire
global debate. It pits those who take a more optimistic approach to AI’s
potential, and want to allow as much experimentation as possible, against
others who fear it is already too late to protect humanity from serious harm.
OpenAI’s
Brockman, one of those who is relaxed about the immediate risks and thinks the
focus should be on addressing longer-term threats, told the French president
that AI was overwhelmingly a force for good, according to three people who
attended the dinner. Any regulation — particularly rules that could hamper the
company’s meteoric growth — should focus on long-term threats like AI
eventually overriding human control, he added.
Others like
Meredith Whittaker, an American technologist also present at the Paris dinner
in November, argued that the real-world harms of existing AI — including faulty
datasets that entrenched racial and gender biases — required politicians to act
now.
“Macron
took it all in,” said one of the attendees who, like others present, was
granted anonymity to discuss the private meeting. “He wanted people to know
France was behind greater regulation, but also that France was also open for
business.”
Similar
arguments — pitting the need for rules to focus on longer-term risks against an
immediate, emergency crackdown — are now playing out across Western capitals.
Scary deepfakes
While
national leaders tried to wrap their heads around the problem, tech executives
were criss-crossing the planet urging politicians not to over-regulate.
Eric
Schmidt, Google’s former chief executive, and LinkedIn founder Reid Hoffman
flew between Washington, London and Paris to offer their own take on how to
handle AI, according to five individuals familiar with the discussions. An
overzealous crackdown, the tech titans argued, would hand authoritarian China a
decisive advantage in AI. Many like Schmidt and Hoffman have invested heavily
in AI startups.
To
reinforce their arguments, tech bosses who want lighter-touch regulation got
personal. AI lobbyists showed at least two Western leaders life-like deepfake
videos — of the leaders themselves. These AI-generated forgeries are still at
the technology’s cutting edge and are often clunky and easy to spot. But the
lobbyists used the deepfakes to show the leaders how the tech had the potential
to evolve to pose serious risks in the future, according to three individuals
briefed on those meetings.
Opponents
claim such tactics allow companies like OpenAI and Alphabet to underplay how
the technology is harming people right now — for example, by rejecting
government benefits claims from minorities and underprivileged people in
society.
“Most
Americans don’t realize AI is already out there,” said Cory Booker, a
Democratic U.S. senator. “I want to make sure that our current laws against
discrimination … our current laws affirming protections, are actually being
enforced.”
Heading
into 2024, those who want a lighter touch appear to be winning, despite EU’s
new binding rules on AI.
After
Brussels reached a political deal in December on its AI Act, Macron — fresh
from his personal AI dinner — sided with German Chancellor Olaf Scholz and
Italian Prime Minister Giorgia Meloni to urge fewer controls on the latest
forms of the technology. The French president worried such regulation would
hamper local champions — including Mistral, the French AI company backed, in
part, by Schmidt, Google’s ex-boss.
France
remained a holdout until finally giving in to political pressure in early
February to support the rules, albeit with major reservations. “We will
regulate things that we will no longer produce or invent,” Macron told an
audience in Toulouse after securing some last-minute carve-outs for European
firms. “This is never a good idea.”
Bioweapons and Big Tech
Tristan
Harris thought he had landed a knockout punch.
In a
private hearing between U.S. lawmakers and tech experts in September, Harris, a
co-founder of the Center for Humane Technology, a nonprofit, described how his
engineers had coerced Meta’s latest AI product into committing a terrifying
act: the construction of a bioweapon.
This showed
that Meta’s approach to artificial intelligence was far too lax and potentially
dangerous, Harris argued, according to two officials who attended the meeting.
Mark
Zuckerberg’s tech giant favored so-called open-source technology — AI easily
accessible to all — with few safeguards against abuse. Such openness, Harris
added, would lead to real-world harm, including the spread of AI-generated
weapons of mass destruction.
Unlike other Silicon Valley giants, Mark Zuckerberg’s
Meta has sided with those who want to keep the technology open to all. |
His triumph
didn’t last long. Zuckerberg, who was also present at the Capitol Hill hearing,
quickly whipped out his phone and found the same bioweapon information via a
simple Google search. Zuckerberg’s counterpunch led to a smattering of laughter
from the room. It blunted Harris’ accusation that Meta’s open-source AI
approach was a threat to humanity.
“It wasn’t
the slam dunk Harris thought it would be,” said one of the officials granted
anonymity to discuss the meeting.
The
Harris-Zuckerberg showdown captures another key question for policymakers:
Should AI be limited to a handful of trusted companies or opened up to all
comers to exploit?
According
to those who favor licensing advanced AI to a select few companies, the
immediate risks are too stark to let loose on the wider world. A PowerPoint
presentation warning that open-sourced AI models would unleash uncontrollable
bioweapons accessible to terrorists has been presented in private to government
officials in London, Paris and Washington, according to six policymakers
involved in those meetings. POLITICO could not determine which organization was
behind these dire predictions.
Microsoft
and OpenAI are among the companies that favor restricting the technology to a
small number of firms so regulators can build ties with AI innovators.
“A
licensing-based approach is the right way forward,” Natasha Crampton,
Microsoft’s chief responsible AI officer, told POLITICO. “It allows a close
interaction between the developer of the tech and the regulator to really
assess the risks.”
Critics of
Silicon Valley’s dominance argue Big Tech has a vested interest in shutting out
competition. According to open-source advocate Mark Surman, executive director
of the Mozilla Foundation, the world’s largest tech companies want to capture
the fledgling AI industry for themselves and cut off rival upstarts.
During the
fall, Surman toured Western capitals to urge senior U.S., EU and U.K. officials
not to limit who could develop next-generation AI models. Unlike other Silicon
Valley giants, Zuckerberg’s Meta has sided with people like Surman who want to
keep the technology open to all, a point he reinforced with his showdown with
Harris over bioweapons in September.
The open
approach is winning friends in powerful parts of Washington, although U.S.
officials remain divided. In Brussels, policymakers agreed to free almost all
open-source AI from the toughest oversight — as long as the technology is not
used in outlawed practices, such as wholesale facial recognition. In the U.K.,
however, officials lean toward restricting the most advanced AI to a handful of
countries.
“If we
leave AI in the hands of the few, there aren’t enough eyes out there to
scrutinize what they do with it,” said Surman.
The future is here
If 2023 was
the year AI lobbying burst into the political mainstream, 2024 will decide who
wins.
Key
decision dates are fast approaching. By the summer, parts of Europe’s AI Act
will come into force. The White House’s rival AI executive order is already
changing how federal agencies handle the technology. More reforms in Washington
are expected by the year’s end.
Other
countries like the U.K., Japan and Canada — as well as China — are making their
own plans, which include greater oversight of the most advanced AI and efforts
to get other countries to adopt their rules.
What’s
clear, according to the dozens of AI experts who spoke to POLITICO, is that in
2024, the public’s desire for political action is unlikely to disappear. AI’s
impact, especially in a year of mass elections from the U.S. and EU to the
Philippines to India — is anything but certain.
“This
business about AI’s loss of control is suddenly being taken seriously,” said
Max Tegmark, president of the Future of Life Institute, a nonprofit that
advocates for greater checks on AI. “People are realizing this isn’t a
long-term thing anymore. It’s happening.”
Mark Scott
reported from London, Brussels, Paris, Berlin and Washington; Clothilde Goujard
from Kyoto, Japan, Vincent Manancourt from London, Gian Volpicelli from
Brussels, Mohar Chatterjee and Brendan Bordelon from Washington.
CORRECTION:
A previous version of this article misstated which firms favor restricting the
technology to a small number of firms. They are Microsoft
and OpenAI.
Sem comentários:
Enviar um comentário