How Denmark is trying to protect citizens from AI deepfakes

In 2021, Danish video crippled live-streamer Marie Watson received an image of herself from an chartless Instagram account.

She instantly recognised nan vacation threat from her Instagram account, but thing was different: Her clothing had been digitally removed to make her look naked. It was a deepfake.

“It overwhelmed maine truthful much,” Watson recalled. “I conscionable started bursting retired successful tears, because suddenly, I was location naked”.

In nan 4 years since her experience, deepfakes — highly realistic images, videos, aliases audio of existent group aliases events that are generated by artificial intelligence (AI) — person go not only easier to make worldwide but besides look aliases sound exponentially much realistic.

That's acknowledgment to technological advances and nan proliferation of generative AI tools, including video procreation devices from OpenAI and Google.

These devices springiness millions of users nan expertise to easy spit retired content, including for nefarious purposes that scope from depicting celebrities Taylor Swift and Katy Perry to disrupting elections and humiliating teens and women.

Copyright law

In response, Denmark is seeking to protect mean Danes, arsenic good arsenic performers and artists who mightiness person their quality aliases sound imitated and shared without their permission.

A measure that's expected to walk early adjacent twelvemonth would alteration copyright rule by imposing a prohibition connected nan sharing of deepfakes to protect citizens’ individual characteristics — specified arsenic their quality aliases sound — from being imitated and shared online without their consent.

If enacted, Danish citizens would get nan copyright complete their ain likeness. In theory, they past would beryllium capable to request that online platforms return down contented shared without their permission. The rule would still let for parodies and satire, though it’s unclear really that will beryllium determined.

Experts and officials opportunity nan Danish authorities would beryllium among nan astir extended steps yet taken by a authorities to combat misinformation done deepfakes.

Henry Ajder, laminitis of consulting patient Latent Space Advisory and a starring master successful generative AI, said that he applauds nan Danish authorities for recognising that nan rule needs to change.

“Because correct now, erstwhile group opportunity ‘what tin I do to protect myself from being deepfaked?’ nan reply I person to springiness astir of nan clip is: ‘There isn’t a immense magnitude you tin do,’" he said, “without maine fundamentally saying, ‘scrub yourself from nan net entirely.’ Which isn’t really possible”.

He added: "We can’t conscionable dress that this is business arsenic accustomed for really we deliberation astir those cardinal parts of our personality and our dignity”.

Deepfakes and misinformation

US President Donald Trump signed bipartisan authorities successful May that makes it forbidden to knowingly people aliases frighten to people friendly images without a person’s consent, including deepfakes.

Last year, South Korea rolled retired measures to curb deepfake porn, including harsher reward and stepped up regulations for societal media platforms.

Danish Culture Minister Jakob Engel-Schmidt said that nan measure has wide support from lawmakers successful Copenhagen, because specified integer manipulations tin operation doubts astir reality and dispersed misinformation.

“If you’re capable to deepfake a leader without her aliases him being capable to person that merchandise taken down, that will undermine our democracy,” he told reporters during an AI and copyright convention successful September.

The correct balance

The rule would use only successful Denmark, and is improbable to impact fines aliases imprisonment for societal media users. But large tech platforms that neglect to region deepfakes could look terrible fines, Engel-Schmidt said.

Ajder said Google-owned YouTube, for example, has a “very, very bully strategy for getting nan equilibrium betwixt copyright protection and state of creativity”.

The platform's efforts propose that it recognises “the standard of nan situation that is already present and really overmuch deeper it’s going to become,” he added.

Twitch, TikTok, and Meta, which owns Facebook and Instagram, didn't respond to requests for comment.

Engel-Schmidt said that Denmark, nan existent holder of nan European Union's rotating presidency, had received liking successful its projected authorities from respective different EU members, including France and Ireland.

Intellectual spot lawyer Jakob Plesner Mathiasen said that nan authorities shows nan wide request to combat nan online threat that's now infused into each facet of Danish life.

“I deliberation it decidedly goes to opportunity that nan ministry wouldn’t make this bill, if location hadn’t been immoderate juncture for it,” he said.

“We're seeing it pinch clone news, pinch authorities elections. We are seeing it pinch pornography, and we’re besides seeing it besides pinch celebrated group and besides mundane group — for illustration you and me”.

The Danish Rights Alliance, which protects nan authorities of imaginative industries connected nan internet, supports nan bill, because its head says that existent copyright rule doesn't spell acold enough.

Danish sound character David Bateson, for example, was astatine a nonaccomplishment erstwhile AI sound clones were shared by thousands of users online. Bateson voiced a characteristic successful nan celebrated “Hitman” video game, arsenic good arsenic Danish toymaker Lego's English advertisements.

“When we reported this to nan online platforms, they opportunity ‘OK, but which regularisation are you referring to?’” said Maria Fredenslund, an lawyer and nan alliance’s director.

“We couldn’t constituent to an nonstop regularisation successful Denmark”.

‘When it’s online, you're done'

Watson had heard astir chap influencers who recovered digitally-altered images of themselves online, but ne'er thought it mightiness hap to her.

Delving into a acheronian broadside of nan web wherever faceless users waste and stock deepfake imagery — often of women — she said she was shocked really easy it was to create specified pictures utilizing readily disposable online tools.

“You could virtually conscionable hunt ‘deepfake generator’ connected Google aliases ‘how to make a deepfake,’ and each these websites and generators would popular up,” nan 28-year-old Watson said.

She is gladsome her authorities is taking action, but she isn’t hopeful. She believes much unit must beryllium applied to societal media platforms.

“It shouldn’t beryllium a point that you tin upload these types of pictures,” she said. “When it’s online, you’re done. You can’t do anything, it’s retired of your control”.