When ‘Her’ was released in 2013, it depicted a vision of the future where it could be possible to have a relationship with a machine.
Eleven years on, the film in which Joaquin Phoenix’s character Theodore falls in love with his AI assistant is getting old – but the story it depicts is eerily current.
You might even think his virtual girlfriend had appeared in real life, after OpenAI unveiled the latest iteration of its large language model ChatGPT, which offers the ability to chat via speech, not text.
A video demo of the new technology had many mistaking the ‘flirtatious’ voice for that of Scarlett Johansson, who voiced AI interface Samantha in the film.
The actress herself was unhappy, saying she had declined for ‘personal reasons’ when asked to be the voice of Sky.
She said the voice was so similar to hers that even family were confused, and contacted OpenAI via lawyers to ask them how they obtained it.
The company denied they had used her voice, saying it was trained on a different professional actor, but amid the controversy they removed Sky from the five voices used in the app.
It highlights the potential problems with AI being able to mimic someone’s voice in real life, an issue we will have to grapple with more and more.
It is now possible to fake someone’s voice entirely from just a few seconds of audio of them speaking, such as from someone’s answerphone message, AI ethics expert Nell Watson told Metro.co.uk.
She said that the issue is becoming urgent as legislation has not caught up with the technology, particularly in the UK which lags behind other countries such as France and Canada.
There is currently no law in this country specifically allowing people the rights to their own voice.
So you own the copyright to a quick selfie you took on holiday and could take legal action to stop others using it, but would have to rely on secondary laws like harassment or GDPR if you wanted to stop someone using your actual voice.
A law on deepfakes is working its way through Parliament, but this only looks at pornographic fakes of real people, not the creation of synthetic media of their likenesses in general.
In the past, it wasn’t necessary to copyright personal characteristics as people had very limited ways to use them, and certainly could not manipulate the recording into making it seem you had said something you had not.
But now, the issue is a big one. Actors in particular are concerned, as they could lose out on work if companies were able to simply pay them once, and then use those initial recordings to make them say anything they wanted to without it costing them anything further.
Scammers are also already using ‘voice phishing’ technology, with a finance worker in Hong Kong recently tricked into paying £20 million of his company’s money to fraudsters after joining a video call where his boss and colleagues were all deepfakes.
We’ve all had WhatsApp text messages from scammers claiming to be family members, but manipulated audio where a loved one could sound distressed is much more difficult to dismiss, such as when scammers convinced a mother they had kidnapped her daughter using faked audio.
Nell, whose book Taming the Machine looks at responsible use of AI, said: ‘The UK has an opportunity to learn from other nations in how how it engages in publicity rights, now more than ever when it becomes so trivial for people to create these fakes.
‘It may be very difficult to track down who has actually done something, and so it’s important that there are some investigatory powers which typically civic law doesn’t provide.’
There are some positives to being able to create such natural-sounding likenesses, for example allowing people to play lifelike games where they can remix or expand on content.
But if not regulated, it creates the risk of people losing control of their own identities.
There are now ‘off the shelf technologies’ that can be bought or rented for as little as £20 that give people the capability to make such convincing fakes, Nell said.
Referring to Scarlett Johansson’s spat with OpenAI, Dominic Lees, a deepfake expert at the University of Reading, said: ‘AI developers need to be careful.
‘High-profile cases like this demonstrate the many problems that can be caused through the misuse of deepfake technology, and highlights calls for new regulations to protect individuals from unauthorised digital replication.
‘Ethical AI development should prioritise consent, transparency, and respect for personal rights to prevent exploitation and preserve public trust.’
