Tag: Safety

A lobby group backed by Elon Musk and associated with a controversial ideology popular among tech billionaires is fighting to prevent killer robots from terminating humanity, and it’s taken hold of Europe’s Artificial Intelligence Act to do so.

The Future of Life Institute (FLI) has over the past year made itself a force of influence on some of the AI Act’s most contentious elements. Despite the group’s links to Silicon Valley, Big Tech giants like Google and Microsoft have found themselves on the losing side of FLI’s arguments.

In the EU bubble, the arrival of a group whose actions are colored by fear of AI-triggered catastrophe rather than run-of-the-mill consumer protection concerns was received like a spaceship alighting in the Schuman roundabout. Some worry that the institute embodies a techbro-ish anxiety about low-probability threats that could divert attention from more immediate problems. But most agree that during its time in Brussels, the FLI has been effective. 

“They’re rather pragmatic and they have legal and technical expertise,” said Kai Zenner, a digital policy adviser to center-right MEP Axel Voss, who works on the AI Act. “They’re sometimes a bit too worried about technology, but they raise a lot of good points.” 

Launched in 2014 by MIT academic Max Tegmark and backed by tech grandees including Musk, Skype’s Jaan Tallinn, and crypto wunderkind Vitalik Buterin, FLI is a nonprofit devoted to grappling with “existential risks” — events able to wipe out or doom humankind. It counts other hot shots like actors Morgan Freeman and Alan Alda and renowned scientists Martin (Lord) Rees and Nick Bostrom among its external advisers.

Chief among those menaces — and FLI’s priorities — is artificial intelligence running amok.

“We’ve seen plane crashes because an autopilot couldn’t be overruled. We’ve seen a storming of the U.S. Capitol because an algorithm was trained to maximize engagement. These are AI safety failures today — as these systems become more powerful, harms might become worse,” Mark Brakel, FLI director of European policy, said in an interview.

But the lobby group faces two PR problems. First, Musk, its most famous backer, is at the center of a storm since he started mass firings at Twitter as its new owner, catching the eye of regulators, too. Musk’s controversies could cause lawmakers to get skittish about talking to FLI. Second, the group’s connections to a set of beliefs known as effective altruism are raising eyebrows: The ideology faces a reckoning and is most recently being blamed as a driving force behind the scandal around cryptocurrency exchange FTX, which has unleashed financial carnage. 

How FLI pierced the bubble

The arrival of a lobby group fighting off extinction, misaligned artificial intelligence and killer robots was bound to be refreshing to otherwise snoozy Brussels policymaking.

FLI’s Brussels office opened in mid-2021, as discussions about the European Commission’s AI Act proposal were kicking off.

“We would prefer AI to be developed in Europe, where there will be regulations in place,” Brakel said. “The hope is that people take inspiration from the EU.”

A former diplomat, the Dutch-born Brakel joined the institute in May 2021. He chose to work in AI policy as a field that was both impactful and underserved. Policy researcher Risto Uuk joined him two months later. A skilled digital operator — he publishes his analyses and newsletter from the domain artificialintelligenceact.eu — Uuk had previously done AI research for the Commission and the World Economic Forum. He joined FLI out of philosophical affinity: like Tegmark, Uuk subscribes to the tenets of effective altruism, a value system prescribing the use of hard evidence to decide how to benefit the largest number of people.

Since starting in Brussels, the institute’s three-person team (with help from Tegmark and others, including law firm Dentons) has deftly spearheaded lobbying efforts on little-known AI issues.

Exhibit A: general-purpose AI — software like speech-recognition or image-generating tools used in a vast array of contexts and sometimes affected by biases and dangerous inaccuracies (for instance, in medical settings). General-purpose AI was not mentioned in the Commission’s proposal, but wended its way into the EU Council’s final text and is guaranteed to feature in Parliament’s position.

“We came out and said, ‘There’s this new class of AI — general-purpose AI systems — and the AI Act doesn’t consider them whatsoever. You should worry about this,’” Brakel said. “This was not on anyone’s radar. Now it is.”

The group is also playing on European fears of technological domination by the U.S. and China. “General-purpose AI systems are built mainly in the U.S. and China, and that could harm innovation in Europe, if you don’t ensure they abide by some requirements,” Brakel said, adding this argument resonated with center-right lawmakers with whom he recently met. 

Another of FLI’s hobbyhorses is outlawing AI able to manipulate people’s behavior. The original proposal bans manipulative AI, but that is limited to “subliminal” techniques — which Brakel thinks would create loopholes. 

But the AI Act’s co-rapporteur, Romanian Renew lawmaker Dragoș Tudorache, is now pushing to make the ban more comprehensive. “If that amendment goes through, we would be a lot happier than we are with the current text,” Brakel said.

So smart it made crypto crash

While the group’s input on key provisions in the AI bill was welcomed, many in Brussels’ establishment look askance at its worldview.

Tegmark and other FLI backers adhere to what’s referred to as effective altruism (or EA). A strand of utilitarianism codified by philosopher William MacAskill — whose work Musk called “a close match for my philosophy” — EA dictates that one should better the lives of as many people as possible, using a rationalist fact-based approach. At a basic level, that means donating big chunks of one’s income to competent charities. A more radical, long-termist strand of effective altruism demands that one strive to minimize risks able to kill off a lot of people — and especially future people, who will greatly outnumber existing ones. That means that preventing the potential rise of an AI whose values clash with humankind’s well-being should be at the top of one’s list of concerns.

A critical take on FLI is that it is furthering this interpretation of the so-called effective altruism agenda, one supposedly uninterested in the world’s current ills — such as racism, sexism and hunger — and focused on sci-fi threats to yet-to-be-born folks. Timnit Gebru, an AI researcher whose acrimonious exit from Google made headlines in 2020, has lambasted FLI on Twitter, voicing “huge concerns” about it.

“They are backed by billionaires including Elon Musk — that already should make people suspicious,” Gebru said in an interview. “The entire field around AI safety is made up of so many ‘institutes’ and companies billionaires pump money into. But their concept of AI safety has nothing to do with current harms towards marginalized groups — they want to reorient the entire conversation into preventing this AI apocalypse.”

Effective altruism’s reputation has taken a hit in recent weeks after the fall of FTX, a bankrupt exchange that lost at least $1 billion in customers’ cryptocurrency assets. Its disgraced CEO Sam Bankman-Fried used to be one of EA’s darlings, talking in interviews about his plan to make bazillions and give them to charity. As FTX crumbled, commentators argued that Effective Altruism ideology led Bankman-Fried to cut corners and rationalize his recklessness. 

Both MacAskill and FLI donor Buterin defended EA on Twitter, saying that Bankman-Fried’s actions contrasted with the philosophy’s tenets. “Automatically downgrading every single thing SBF believed in is an error,” wrote Buterin, who invented the Ethereum blockchain, and bankrolls FLI’s scholarship for AI existential risk research.

Brakel said that the FLI and EA were two distinct things, and FLI’s advocacy was focused on present problems, from biased software to autonomous weapons, e.g. at the United Nations level. “Do we spend a lot of time thinking about what the world would look like in 400 years? No,” he said. (Neither Brakel nor the FLI’s EU representative, Claudia Prettner, call themselves effective altruists.)

Californian ideology

Another critique of FLI’s efforts to stave off evil AI argues that they obscure a techno-utopian drive to develop benevolent human-level AI. At a 2017 conference, FLI advisers — including Musk, Tegmark and Skype’s Tallinn — debated the likelihood and the desirability of smarter-than-human AI. Most panelists deemed “superintelligence” bound to happen; half of them deemed it desirable. The conference’s output was a series of (fairly moderate) guidelines on developing beneficial AI, which Brakel cited as one of FLI’s foundational documents.

That techno-optimism led Emile P. Torres, a Ph.D. candidate in philosophy who used to collaborate with FLI, to ultimately turn against the organization. “None of them seem to consider that maybe we should explore some kind of moratorium,” Torres said. Raising such points with an FLI staffer, Torres said, led to a sort of excommunication. (Torres’s articles have been taken down from FLI’s website.)

Within Brussels, the worry is that going ahead, FLI might change course from its current down-to-earth incarnation and steer the AI debate toward far-flung scenarios. “When discussing AI at the EU level, we wanted to draw a clear distinction between boring and concrete AI systems and sci-fi questions,” said Daniel Leufer, a lobbyist with digital rights NGO Access Now. “When earlier EU discussions on AI regulation happened, there were no organizations in Brussels placing focus on topics like superintelligence — it’s good that the debate didn’t go in that direction.”

Those who regard the FLI as the spawn of Californian futurism point to its board and its wallet. Besides Musk, Tallinn and Tegmark, donors and advisers include researchers from Google and OpenAI, Meta co-founder Dustin Moskovitz’s Open Philanthropy, the Berkeley Existential Risk Initiative (which in turn has received funding from FTX) and actor Morgan Freeman. 

In 2020 most of FLI’s global funding ($276,000 out of $482,479) came from the Silicon Valley Community Foundation, a charity favored by tech bigwigs like Mark Zuckerberg; 2021 accounts haven’t been released yet. 

Brakel denied that the FLI is cozy with Silicon Valley, saying that the organization’s work on general-purpose AI made life harder for tech companies. Brakel said he had never spoken to Musk. Tegmark, meanwhile, is in regular touch with the members of the scientific advisory board, which includes Musk. 

In Brakel’s opinion, what the FLI is doing is akin to early-day climate activism. “We currently see the warmest October ever. We worry about it today, but we also worry about the impact in 80 years’ time,” he said last month. “[There] are AI safety failures today — and as these systems become more powerful, the harms might become worse.”