In an age of Photoshop, filters and social media, many people are used to seeing manipulated pictures — subjects become slimmer and smoother or, in the case of Snapchat, transformed into puppies.
However, there is a new breed of video and audio manipulation tools, made possible by advances in artificial intelligence and computer graphics, that would allow for the creation of realistic-looking footage of public figures appearing to say, well, anything.
US President Donald Trump declaring his proclivity for water sports. Former US secretary of state Hillary Rodham Clinton describing the stolen children she keeps locked in her wine cellar. Actor Tom Cruise finally admitting what we suspected all along — that he is a brony (a My Little Pony fan.)
This is the future of fake news. People have long been told not to believe everything they read, but soon they will have to question everything they see and hear as well.
For now, there are several research teams working on capturing and synthesizing different visual and audio elements of human behavior.
Software developed at Stanford University is able to manipulate video footage of public figures to allow a second person to put words in their mouth — in real time.
Face2Face captures the second person’s facial expressions as they talk into a webcam and then morphs those movements directly onto the face of the person in the original video.
The research team demonstrated their technology by puppeteering videos of former US president George W. Bush, Russian President Vladimir Putin and Trump.
On its own, Face2Face is a fun plaything for creating memes and entertaining late-night talk show hosts.
However, with the addition of a synthesized voice, it becomes more convincing — not only does the digital puppet look like the politician, but it can also sound like the politician.
A research team at the University of Alabama at Birmingham has been working on voice impersonation.
With three to five minutes of audio of a victim’s voice — taken live or from YouTube videos or radio shows — an attacker can create a synthesized voice that can fool both humans and voice biometric security systems used by some banks and smartphones.
The attacker can then talk into a microphone and the software will convert it so that the words sound like they are being spoken by the victim — whether that is over the telephone or on a radio show.
Canadian start-up Lyrebird has developed similar capabilities, which it says can be used to turn text into on-the-spot audiobooks “read” by famous voices or for characters in video games.
Although their intentions might be well-meaning, voice-morphing technology could be combined with face-morphing technology to create convincing fake statements by public figures.
You only have to look at the University of Washington’s Synthesizing Obama project, where they took the audio from one of former US president Barack Obama’s speeches and used it to animate his face in an entirely different video with incredible accuracy — thanks to training a recurrent neural network with hours of footage — to get a sense of how insidious these adulterations can be.
Beyond fake news, there are many other implications, said Nitesh Saxena, associate professor and research director of the University of Alabama at Birmingham’s department of computer science.
“You could leave fake voice messages posing as someone’s mom. Or defame someone and post the audio samples online,” Saxena said.
These morphing technologies are not yet perfect. The facial expressions in the videos can seem a little distorted or unnatural and the voices can sound a little robotic.
However, given time, they will be able to faithfully recreate the sound or appearance of a person — to the point where it might be very difficult for humans to detect the fraud.
Given the erosion of trust in the media and the rampant spread of hoaxes via social media, it will become even more important for news organizations to scrutinize content that looks and sounds like the real deal.
Telltale signs will be where the video or audio was created, who else was at the event and whether the weather conditions match the records of that day.
People should also be looking at the lighting and shadows in the video, whether all of the elements featured in the frame are the right size and whether the audio is synced perfectly, said Mandy Jenkins, from social news company Storyful, which specializes in verifying news content.
Doctored content might not pass the scrutiny of a rigorous newsroom, but if posted as a grainy video to social media, it could spread virally and trigger a public relations, political or diplomatic disaster. Imagine Trump declaring war on North Korea, for example.
“If someone looks like Trump and speaks like Trump, they will think it’s Trump,” Saxena said.
“We already see it doesn’t even take doctored audio or video to make people believe something that isn’t true,” Jenkins added. “This has the potential to make it worse.”
When 17,000 troops from the US, the Philippines, Australia, Japan, Canada, France and New Zealand spread across the Philippine archipelago for the Balikatan military exercise, running from tomorrow through May 8, the official language would be about interoperability, readiness and regional peace. However, the strategic subtext is becoming harder to ignore: The exercises are increasingly about the military geography around Taiwan. Balikatan has always carried political weight. This year, however, the exercise looks different in ways that matter not only to Manila and Washington, but also to Taipei. What began in 2023 as a shift toward a more serious deterrence posture
Reports about Elon Musk planning his own semiconductor fab have sparked anxiety, with some warning that Taiwan Semiconductor Manufacturing Co (TSMC) could lose key customers to vertical integration. A closer reading suggests a more measured conclusion: Musk is advancing a strategic vision of in-house chip manufacturing, but remains far from replacing the existing foundry ecosystem. For TSMC, the short-term impact is limited; the medium-term challenge lies in supply diversification and pricing pressure, only in the long term could it evolve into a structural threat. The clearest signal is Musk’s announcement that Tesla and SpaceX plan to develop a fab project dubbed “Terafab”
China’s AI ecosystem has one defining difference from Silicon Valley: It is embrace of open source. While the US’ biggest companies race to build ever more powerful systems and insist only they can control them, Chinese labs have been giving the technology away for free. Open source — making a model available for anyone to use, download and build on — once seemed a niche, nerdy topic that no one besides developers cared about. However, when a new technology is driving trillions of dollars of investments and leading to immense concentrations of power, it offered an antidote. That is part of
In late January, Taiwan’s first indigenous submarine, the Hai Kun (海鯤, or Narwhal), completed its first submerged dive, reaching a depth of roughly 50m during trials in the waters off Kaohsiung. By March, it had managed a fifth dive, still well short of the deep-water and endurance tests required before the navy could accept the vessel. The original delivery deadline of November last year passed months ago. CSBC Corp, Taiwan, the lead contractor, now targets June and the Ministry of National Defense is levying daily penalties for every day the submarine remains unfinished. The Hai Kun was supposed to be
RSS