OpenAI introduces a new program: Voice Engine can be used to imitate voices. In a super election year, this increases fears of manipulation.

You can also impersonate Donald Trump on the phone, thanks to the voice engine Photo: Toru Hanai/Reuters

In March 2022, President Zelensky left Ukraine and surrendered to Russia. In November 2023, Chancellor Olaf Scholz declared that he would ban the AfD. And in January of this year, US President Joe Biden himself called his voters and asked them not to go to the primaries. Of course, none of this ever happened. And yet, it is part of our political reality. “Deepfakes”, the creation of fake videos using artificial intelligence, have become a phenomenon in recent years.

Since 2019 at the latest, reports and warnings about the dangers of deception have increased. Even in a super election year like this, deepfakes are perceived as a danger. They are a simple, low-threshold means of influencing European elections and especially American presidential elections. Now, more than ever, one of the biggest artificial intelligence companies is introducing new software that makes imitating voices easier than ever.

With the ChatGPT chatbot, the American company OpenAI has established itself as a market leader in artificial intelligence. The chatbot can hold conversations with users, answer questions, and write complete essays using a simple task. On Friday, OpenAI debuted the voice engine. The software should allow you to perfectly imitate a voice, even in several languages. A 15-second example is enough and the program will be able to imitate the voice in all positions and with all the characteristics.

Public figures can hardly protect themselves from the misuse of their voice. Even for individuals, a voice message from messaging services is enough for your voice to be misused. OpenAI knows the dangers that come with its own program. This is also why the company is still holding back the software and wants to be careful about releasing it more widely.

Joe Biden's wrong call

The United Nations also sees the need to act in the field of artificial intelligence. On March 21, the organization passed a resolution to better control AI. The goal is “safe and reliable artificial intelligence systems” and the sustainable development of the technology, without abuses or violations of data protection. How this will be implemented remains to be seen in the legally non-binding resolution.

Deepfakes, whether of the face, voice or both, represent a massive violation of personal rights: they not only create the risk of blackmail and harassment, but also massive psychological stress. In recent years, deepfake porn has exploded, with both well-known stars and private faces.

It is now easy to get people to say and do things they have never said or done. The technology still looks a little bumpy in places, but its rapid development will smooth out these issues. Some hoaxes, such as election interference in the US primaries, don't even require a face, just a few sentences. In the US state of New Hampshire, Democratic voters were summoned by a fake presidential vote. The supposed Joe Biden asked not to attend the primaries.

It is still unclear who is behind the manipulation. But a case like this is enough to fuel uncertainty. Although OpenAI's speech engine opens a new and even faster chapter in speech imitation, the problem has long been a part of everyday media life. Much more media literacy is needed, especially in schools, to be able to recognize and classify deepfakes.

Trump and Obama on chip rates

For some time now, on social media, especially YouTube and Tiktok, there has also been a trend called “AI Presidents”: Joe Biden, Donald Trump and Barack Obama talk to each other about various topics in fake videos of different lengths. Sometimes they make racist jokes, rate different types of chips or video games, insult each other, or play UNO against each other.

Sometimes you can also hear the voices of George W. Bush or Hillary Clinton in the videos. Although presidents usually appear only in cropped images and in slight movements, their voices are deceptively real.

Humorous videos may seem entertaining, but they show how easy it is now to fake voices and entire conversations. Anyone can create deepfakes and make presidents and other celebrities say whatever they want, from extremist statements to declarations of surrender and love. While it used to take several minutes of the original voice, thanks to Voice Engine it is now just a few seconds.