Is it enough? Experts warn that AI-generated deepfakes could affect the outcome of crucial upcoming elections. But measures are finally being taken to combat them.
Social media giants to label AI-made content
Is it enough? Experts warn that AI-generated deepfakes could affect the outcome of crucial upcoming elections. But measures are finally being taken to combat them.
Fakery snakery
Jolene looks at the video in amazement. There is Donald Trump handing out food to hungry children; giving a bed in his own house to a homeless man; addressing a rally in support of women's rights; receiving a medal for courage in Vietnam; punching President Putin on the nose! Maybe she should vote for him in the presidential election.
But then she notices a small label in the corner of her screen: "Made with AIA computer programme that has been designed to think. ."
MetaThe new name of the company which owns Facebook and Instagram. announced last week that, starting in May, labels like this would appear on Instagram, Facebook and Threads. They will be applied to AI-generated videos, pictures and audio clips.
More prominent labels will be attached to fakes which pose a "particularly high risk of materially deceiving the public on a matter of importance" - whether created by AI or not.
The power of AI to make mischief is already clear. There have been fake audio clips of Keir Starmer and Sadiq Khan, and fake phone calls from President Biden. There has even been a whole fake interview with a Ukrainian official appearing to claim credit for the recent terrorist attack in Moscow.
Many people are concerned about the influence such posts could have on upcoming elections. The question is whether effective countermeasures can be implemented in time. America's presidential election will take place in November, and Britain's general election by the end of January.
In February, tech companies including TikTok, Microsoft and X signed an agreement promising to combat political fakes. New HampshireA state in the north-east US with a population of nearly 1.4 million. 's attorney general announced legal action against the company which faked the Biden phone calls, while the US Federal Communications Commission banned any such calls using AI.
That same month the US House of RepresentativesThe lower chamber of the United States congress. There are 435 representatives, with a certain number allocated to each state based on the state's population. set up a taskforce to see how AI might be regulated. In March, leading researchers signed an open letter calling for AI companies to be made legally responsible for harmful fakes using their technology.
Both Meta and OpenAI have started putting watermarksA logo, text or pattern that is deliberately put over an image to make it more difficult to use without permission. on AI-generated images. Maryland University is creating an encryptionA way of scrambling data so that only certain people can understand it. system to authenticate live recordings. A new charity, TrueMedia.org, has released free tools to help anyone detect fakes.
But the charity's head, Dr Oren Etzioni, warns that no tools are entirely effective. Some have been fooled by obviously fake images, such as two robots kissing. And watermarks can be removed by slightly altering the pictures they are attached to.
When it comes to elections, Etzioni says, "I'm terrified. There is a very good chance we are going to see a tsunamiA Japanese word describing a succession of waves caused when an earthquake or volcano displaces a large body of water. of misinformation."1 To deal with it, there needs to be co-operation between governments, AI companies and tech giants - and he does not believe the chances of that happening before November are high.
According to Professor Rory Cormac of Nottingham University, the greatest worry is the effect of fakes on "the perceived legitimacy of those elections and the trust people have in the outcome... The threat is chronic".2
Is it enough?
Yes: People are on the lookout for deepfakes anyway, and this will make them far easier to spot. Tech giants like Microsoft have so much expertise at their disposal that little will get past them.
No: The technology used to create deepfakes is amazing and is going to get more and more sophisticated as time goes on. It will always be one step ahead of those who are trying to combat it.
Or... It is unclear whether deepfakes can swing an election or not. China's attempts to undermine Taiwan's recent elections failed, but Russia's efforts in Slovakia seem to have been successful.
Keywords
AI - A computer programme that has been designed to think.
Meta - The new name of the company which owns Facebook and Instagram.
New Hampshire - A state in the north-east US with a population of nearly 1.4 million.
House of Representatives - The lower chamber of the United States congress. There are 435 representatives, with a certain number allocated to each state based on the state's population.
Watermarks - A logo, text or pattern that is deliberately put over an image to make it more difficult to use without permission.
Encryption - A way of scrambling data so that only certain people can understand it.
Tsunami - A Japanese word describing a succession of waves caused when an earthquake or volcano displaces a large body of water.
Social media giants to label AI-made content
Glossary
AI - A computer programme that has been designed to think.
Meta - The new name of the company which owns Facebook and Instagram.
New Hampshire - A state in the north-east US with a population of nearly 1.4 million.
House of Representatives - The lower chamber of the United States congress. There are 435 representatives, with a certain number allocated to each state based on the state's population.
Watermarks - A logo, text or pattern that is deliberately put over an image to make it more difficult to use without permission.
Encryption - A way of scrambling data so that only certain people can understand it.
Tsunami - A Japanese word describing a succession of waves caused when an earthquake or volcano displaces a large body of water.