Misleading ai-generated videos tend to top the list when critics and researchers try to guess what kind of manipulation might threaten the 2018 and 2020 elections, Wired reported. While the technology is still on the rise, the likelihood of its misuse is so staggering that technology companies and academic laboratories prioritize research and fund testing methods. Social platforms have special policies for posts that contain “synthetic and manipulative media” in the hope of strikeing the right balance between protecting free expression and preventing widespread lies.
But now, with about three months to go until November 3rd, a wave of deepfake content seems to have never been broken. Instead, another form of ai-artificial intelligence-generated media is making headlines, making it harder to spot but more likely to become an all-encompassing force on the Internet: deepfake text.
According to Wired, GPT-3, launched last month, is the next frontier of generative writing: an AI that produces shocking sentences. As its output becomes increasingly difficult to distinguish from human-generated texts, it is conceivable that the vast majority of written content seen on the Internet in the future is produced by machines. If that happens, how will it change the way we react to what we do around?
This won’t be the first such media inflection point. Thirty years ago, when Photoshop, After Effects, and other image editing and CGI tools began to emerge, their potential for change in artistic creation — and their impact on our perception of the world — was immediately recognized. “Adobe Photoshop could easily be the life-changing program in publishing history,” Macworld announced photoshop 6.0 in a 2000 article. “Today, artists decorate their work through Photoshop, and producers of pornography don’t do photoshop with each of their images, and they’ll get nothing but reality.”
People began to accept the nature of the technology and developed a healthy scepticism. Today, few people believe that magazine covers show what a model really looks like. (In fact, it’s often un-PS content that attracts public attention.) )
The resulting media, such as the deepfake video or GPT-3 output, is different. If used maliciously, there are no unmodified originals, no original materials that can be produced as a basis for comparison, and no evidence of fact verification. In the early 2000s, it was easy to dissect the before and after photos of celebrities and discuss whether the latter created unrealistic ideals of perfection. In 2020, people will be confronted with increasingly plausible celebrity changes on, and clips of world leaders saying things they have never said. People will have to adjust and adapt to new unrealistic levels. Even social media platforms recognize this distinction; their deepfake policy of fake restraint distinguishes between synthetic media content and simply “modifying” media content.
Wired argues that the insinuated generation of text has the potential to distort our social communication ecosystem. However, to manage deepfake content, you must know that it exists. In all forms that exist today, video can become the easiest to detect. “Soft biometrics”, such as a person’s facial movements are not correct; Many of these can be overcome by software tuning. In the 2018 Deepfake video, for example, subjects often blinked wrong, but the problem was fixed shortly after the discovery was made public. The resulting audio may be more subtle — there are no visual effects, so there are fewer chances of making mistakes — but promising research is also under way to understand these issues. The war between the counterfeiters and the experts will last forever.
But synthetic texts — especially the ones that are being produced now — represent a more challenging frontier. It is easy to generate in large quantities and has fewer detections. Deepfake text, unlike synthetic video or audio, can be used at sensitive times. As anyone who has followed the intense Twitter hashtag can attest, activists and marketers recognize the value of dominating the so-called “voice share.” Seeing many people express ingons at the same time or place can convince the observer that everyone has a feeling, whether or not the person who speaks is truly representative or even true. As the time and effort required to produce reviews declines, it will be possible to produce a lot of AI-generated content on any conceivable topic. In fact, people may soon have algorithms to read the web, form “opinions” and then post their own responses. These endless databases of new content and reviews, mainly made by machines, can then be processed by other machines, leading to a feedback loop that will dramatically change our information ecosystem.
Now, you can detect duplicate or recycled comments that use the same text snippets to flood the comment snippets, game Twitter tags, or convince viewers through Facebook posts. This strategy has been observed in a series of maneuvers in the past, including a campaign to solicit public opinion on issues such as payday loans and the Federal Communications Commission’s net neutrality policy. The Wall Street Journal analyzed some of these cases and found hundreds of thousands of questionable comments that were deemed suspicious because they contained long, repetitive sentences and were unlikely to be written spontaneously by different people. If these comments are generated independently by artificial intelligence, for example, these manipulations are difficult to extract.
In the future, deepfake video and audio is likely to be used to create unique, sensational moments to occupy the news cycle or divert attention from other, more organic scandals. But the imperceptible deepfake text – disguised as regular chats on sites such as Twitter, Facebook and Reddit – is likely to be more subtle, general, and sinister. The ability to create a majority opinion, or to create an arms race of false participants — with little chance of detection — would make complex and widespread impact activities possible. The perforated generation of text can distort the social communication ecosystem: algorithmic-generated content receives algorithmic-generated responses that are fed back into algorithm-mediated systems that float information based on engagement.
With the popularity of all types of synthetic media — text, video, photos and audio — and the increasing difficulty of detection, people will find it increasingly difficult to trust what they see, Wired says.