The age of deception

0
26
the age of deception

When AI can alter reality

Since 2020, artificial intelligence has increasingly made its way into our lives. We began to notice this when the first deepfakes appeared: a technique that uses artificial intelligence to replace a subject’s face in a video or photo with another one in an almost perfect way. Although their official birth predates 2020, their use has gradually spread thanks to the development of tools that have increasingly simplified their creation.

Deepfakes immediately highlighted one of the main problems with artificial intelligence: the ability to modify and make plausible photographs or videos of events that never happened.

While replacing famous actors’ faces with other subjects to see them as movie protagonists immediately appeared revolutionary and fun, seeing the same technology applied to pornography quickly generated outcry and fear.

Many famous women have unknowingly found themselves featured in pornographic videos and photos, and the worst part was having to deny involvement, despite the obvious fraud. Nevertheless, many will continue to believe that many of these photos or videos are real since debunking false information is always more difficult than creating it.

However, deepfakes haven’t only made inroads in pornography but also in politics, thus being able to easily ruin the victim’s image and consequently influence public opinion.

But this was just the beginning. We became more concerned when Google Duplex was introduced, an AI that (although limited in its tasks) demonstrated how such technology could easily communicate on the phone to make appointments without the interlocutor noticing, using pauses, discourse markers (listen, well, so, …), interjections (mmm, …), to make the conversation more realistic.

>>>  Organoid intelligence

However, the real revolution came with OpenAI’s GPT (Generative Pretrained Transformer), which in its second version had already demonstrated the ability to write newspaper articles, showing writing capabilities equal to those of a human being. But the greatest amazement came especially with ChatGPT, the first chatbot equipped with this technology that allowed us to communicate as if we were really talking to a human and ask it practically anything…

Nevertheless, many must remember another chatbot that preceded ChatGPT and had already demonstrated the potential of AI applied to chatbots: Replika. Replika was born as the first AI-based chatbot. The idea came from an unfortunate episode of its creator, who, having lost a friend in an accident, decided to create a chatbot trained to talk like the deceased pal through their messages. An episode of Black Mirror references this event.

However, the fascination with AIs like ChatGPT lies more in their predictive capability than in their reasoning. Where responses seem to be the result of reasoning, they are instead the result of probabilistic calculation.

But writing wasn’t the only revolution in the AI field, especially when DALL-E and then Midjourney came out because AI began to become capable of producing art from a simple description, managing to replicate styles and techniques of famous artists on completely new image ideas.

True creativity is still an illusion because, despite the exceptional results, everything is the product of training an algorithm on existing works and techniques.

And if that wasn’t enough, there were also applications in the field of voices. Old voice synthesis generators have evolved significantly thanks to AI, producing very natural results. Many of the most recent applications have options to modify emphasis and tone, but the most striking revolution in this field has certainly been the ability to clone human voices and use them as voice synthesizers, and manage to make the clone voice say anything. An early attempt at this was made by Lyrebird, later incorporated into Descript.

>>>  China’s terrifying robot factory

The trend then spread to the music field; we started hearing many covers of famous songs reinterpreted by equally famous singers thanks to AI, raising new fears about the possibility of easily replacing singers and being able to produce songs with someone else’s voice without permission.

However, the most concerning developments came later, when many of these fields of application began to converge into a single tool, such as Heygen, which quickly spread due to its ability to produce audio translations from videos, not only maintaining the original voice tone but also accordingly modifying the subject’s lip movements to match the speech. This created the impression that the subject was really speaking that language. This caused quite a stir, especially regarding the world of dubbing.

The most extreme case of this tool’s application, however, was used to modify what a person can normally say. If we can maintain the tone of voice and modify lip movements, we can create an ad hoc video of a person saying anything they never said. This questions any video and audio evidence.

That’s why we have officially entered the age of deception. From now on, everything we see or hear from a photo, video, or audio could have been manipulated. Anyone will be able to make you say and do things very easily. The truth will become increasingly buried.

What will be the next step, though?

If AI evolves exponentially as it is happening, it’s difficult to imagine its limits, but we will surely begin to see the consequences of multimodal AI capabilities, which can use every source: text, images, video, and sounds to interact with us and provide increasingly complex responses, like ChatGPT 4, Google’s Gemini, and subsequent developments.

>>>  Is Google's A.I. LaMDA sentient?

Subsequently, general AI (AGI) will arrive when AI becomes able to match human capabilities. And Super AI when it’s able to surpass these capabilities.

Who knows how society will have changed by that time and what consequences there will be?