Índice
Deepfakes are a rapidly growing concern in the digital world, as they have the potential to cause significant harm to individuals and society as a whole. As artificial intelligence and machine learning technologies become more and more complex, creating realistic and convincing deepfakes is becoming easier day by day. Deepfakes are manipulated videos or images that can be used for a variety of purposes, including spreading misinformation and identity theft. In fact, they have the ability to spread propaganda and misinformation that could undermine people’s trust in institutions, incite chaos and confusion, and upend the very fabric of our society. In this article, we’ll explore the dangers of deepfakes and provide practical tips on how to detect deepfake videos.
O que é um Deepfake?
We can define deepfake as a type of synthetic media that has been manipulated or generated using artificial intelligence (AI) algorithms. In other words, deepfakes use machine learning to create or alter audio, video, or images to depict events that did not occur or to portray individuals saying or doing things that they did not actually do.
Let’s create AI Images Using Midjourney and Dall-E 2
If you have seen videos of Mark Zuckerberg boasting about his total control of billions of people’s stolen data, Barack Obama using offensive words to describe Donald Trump, or Kit Harrington delivering a heartfelt apology for the disappointing finale of Game of Thrones, then you already have an idea about deepfakes. Deepfakes are the modern-day equivalent of photoshopping. These altered and manipulated digital media like videos, audio, and images can cause harm to both individuals and organizations, just like computer viruses. Identifying good deepfake videos can be exceptionally challenging; however, in contrast to computer viruses, anyone can produce good deep fakes.
Did you know that neural networks can create realistic people who have never existed before? By visiting the website essa pessoa não existe or Gerado.Fotos, you can see a vast array of faces generated by AI. In fact, none of the people you see on this website is real. You can find more of such AI wizardry in este x não existe website.
Os especialistas identificaram os três tipos ou métodos de deepfake mais perigosos. O primeiro são vozes deepfake ou deepfake de áudio. Esta é uma voz gerada por IA que imita de perto a voz de uma pessoa real. O áudio deepfake é atualmente a técnica mais lucrativa empregada por golpistas para a pesca de voz falsa profunda. Por exemplo, em 2019, criminosos usaram tecnologia de voz clonada para extorquir US$ 243.000 do CEO de uma empresa britânica de energia.
O segundo método são os vídeos deepfake, que podem ser usados para manipular a opinião pública e fazer lavagem cerebral nas pessoas, apresentando informações falsas de forma convincente. Os infames vídeos com Barack Obama e o presidente Trump são exemplos de tais vídeos. Eles também podem ser usados para criar vídeos falsos de pessoas comuns, especialmente para criar pornografia de vingança.
This AI technology can also produce deep fake text that convincingly mimics the writings of real people as well as genuine social media posts. These fake accounts may look authentic and even attract followers over time. A group of these accounts can be easily used to slander a company or a product. The only way to differentiate such AI-generated text is to identify patterns in the language used, such as word choice and sentence structure, and detect inconsistencies.
Perigos da tecnologia Deepfake
No mundo atual de notícias falsas e desinformação, as pessoas usam a tecnologia deep fake para espalhar desinformação, criar narrativas falsas, criar uma imagem pública específica, censurar ou zombar de pessoas e criar conteúdo pornográfico.
Deep fakes often target celebrities. Their fame and availability in media make them perfect for memes, and AI technology can easily create convincing deep fakes of them. A few years ago, TikTok was flooded with videos featuring actor Tom Cruise engaging in activities that were unusual for him, such as fooling around in an expensive men’s fashion store, exhibiting a coin trick, and playfully growling while singing a brief version of “Crash Into Me” by the Dave Matthews Band. Another example is the video of Yong Mei, a famous Chinese celebrity that went viral a few years ago. In this video, she was inserted into a 1983 Hong Kong TV series. The video got 240 million views before it was eventually removed by Chinese authorities. However, the creator, who was a fan of Young Mei, apologized publicly on Weibo and said he made the video to raise awareness about deepfake technology.
In politics, deepfakes can be used to create fake videos or images of political figures saying or doing things that they did not actually say or do. This can be done to spread misinformation or damage someone’s reputation. In 2019, a video of Nancy Pelosi was edited to make it seem like she was speaking oddly and looking drunk. It was shared globally, but Facebook did not remove it after being proven fake. Videos on Donald Trump’s story about a reindeer and Barack Obama’s public service announcement are other examples of this type of fake video.
Did you know that 96% of deep fakes are pornographic? A significant majority (99%) of the faces used in these manipulated videos belong to real women, including non-celebrities. The use of deep fakes can allow a person to effortlessly superimpose the image of an ex-girlfriend onto that of a porn actor and circulate it within their social circle. As a result, the victim may face the negative effects of being falsely accused of a taboo act, even if it never happened.
Deepfakes are also increasingly being used in financial scams. These scams typically involve scammers creating deepfakes of important people like CEOs or politicians and then using them to trick people into transferring money or sensitive information. For example, a deepfake video of a CEO might be used to persuade an employee to transfer money to a fraudulent account. In 2019, a group of scammers used the voice of a CEO of a UK-based energy company to impersonate him and trick an employee into transferring $243,000 to a fraudulent account.
Outro aspecto perigoso das falsificações profundas é que existem muitas ferramentas simples disponíveis online para criar falsificações profundas. De acordo com Ben Coleman, do Reality Defender, qualquer estudante do ensino médio pode criar uma falsificação profunda usando um iPhone de cinco anos.
Algumas aplicações éticas de falsificações profundas
It is essential to remember that deep fake technology can have ethical applications in today’s world. One such example is enhancing the image quality of old or low-resolution videos. Instead of using the traditional upscale method, this technology redraws the image, allowing for changes in image quality. Additionally, deep fake technology can synchronize speech with lip movements, known as lip syncing. Perfecting lip movements to match any linguistic articulation would be a significant advancement in the field of film and TV dubbing. A practical example of this is the PSA featuring David Beckham aimed at combating malaria. As Beckham speaks nine languages, deep fake technology was used to adjust his lip movements to each language.
O potencial do deep fake também se estende à revivência de artistas falecidos, como Salvador Dali, no Museu Salvador Dali, na Flórida. Além disso, uma aplicação única dessa tecnologia é animar arte. Por exemplo, o laboratório de pesquisa de IA da Samsung permitiu que a Mona Lisa exibisse movimentos em sua cabeça, olhos e boca. No futuro, essa tecnologia pode economizar tempo e dinheiro para a indústria cinematográfica.
Como identificar deepfakes
Procure inconsistências no vídeo
Deepfakes geralmente têm inconsistências sutis que os denunciam. Por exemplo, a iluminação, as sombras e os reflexos podem não corresponder ao ambiente ou os movimentos da pessoa podem parecer um pouco artificiais.
Um sinal de alerta comum de um vídeo ou imagem deepfake são os olhos não naturais, especialmente se os olhos de uma pessoa parecerem estranhos ou se não piscarem. Piscar e movimentos naturais dos olhos são difíceis de falsificar, então se alguém parece estar olhando sem piscar ou seus movimentos oculares parecem estranhos, pode ser um sinal de que o vídeo foi manipulado. Além disso, quando as pessoas falam umas com as outras, seus olhos geralmente se movem naturalmente e é difícil imitar isso com precisão em um vídeo falso.
An unpropotional face and body is another sign of a deepfake. This is because the AI algorithms that create deepfakes can struggle to replicate the proportions of a person’s face and body accurately, which results in a distorted or exaggerated appearance. If a person looks strange or weird when they turn their head to the side, or if their movements seem disconnected or unnatural from one moment to the next, it might be a sign that a video has been changed. Mismatched skin tones and oddly placed shadows are two other important clues.
Verifique a qualidade do áudio
Os deepfakes podem ter áudio que não corresponde ao movimento dos lábios da pessoa ou ao tom de sua voz. Ouça com atenção para ver se o áudio soa natural e corresponde ao contexto do vídeo. Ao fazer deepfakes, os criadores geralmente prestam mais atenção ao visual do que ao áudio. Os sinais de áudio falso podem incluir movimentos labiais que não correspondem às palavras faladas, vozes robóticas, pronúncia estranha de palavras, ruído de fundo digital ou até mesmo uma completa ausência de áudio.
Use a tecnologia para identificar falsificações profundas
The advancements in deep fake technology have led to the development of protection tools. The University of Buffalo has created a tool that automatically identifies deep fake photos. This tool examines the light reflection in the eyes to distinguish between real and fake photos. Generally, when we look at something, the image we look at is reflected in our eyes, so in genuine photos, both eyes show the same shape and color of reflection. However, computers cannot generate identical reflections in the eyes since they combine multiple photos to generate one.
Moreover, Intel has introduced an AI tool named FakeCatcher that can identify whether a video has been altered using deepfake technology in real-time. This technology claims to achieve a 96% accuracy rate and detect deepfakes within milliseconds. Unlike other deepfake detectors that rely on deep learning to analyze a video’s raw data for manipulation signs, FakeCatcher looks for human-like indicators, such as subtle changes in blood flow in the pixels of a video.
Existem também outras ferramentas para analisar o ruído digital e fotos. A estrutura do ruído natural e gerado por computador pode variar consideravelmente, ajudando-nos a diferenciar entre fotos genuínas e falsas.
A Coalition for Content Provenance and Authenticity, liderada pela Adobe, Microsoft, Intel e BBC, também desenvolveu um padrão para diferenciar entre conteúdo gerado por computador e conteúdo real. Até que seja amplamente adotado, o potencial de uso malicioso dos deepfakes é significativo.
Como se proteger de deepfakes
You should always exercise caution when consuming information online, especially when it comes to sensitive topics that can cause conflict or provoke intense emotions. Be alert to possible manipulations or distortions of facts that can be used to push a particular agenda or influence public opinion. Always seek out multiple independent sources of information to verify the accuracy of online content. Don’t rely only on video, photographs, or audio on someone’s profile, as these may be deep fakes or altered. It’s important to do your research and cross-check information from different sources to ensure its authenticity. Finally, stay away from synthetic social networks and fake social media accounts.
Para evitar ser vítima de deepfakes, certifique-se de não postar informações pessoais sobre você online, pois elas podem ser usadas para roubo de identidade ou outros fins maliciosos. Seja cauteloso com o tipo de informação que você compartilha e com quem você as compartilha e sempre tome medidas para proteger sua privacidade e segurança online.
É crucial educar os indivíduos sobre os perigos das falsificações profundas e desenvolver novas tecnologias para detectar e combater seu uso. A criação e distribuição de deep fakes deve ser regulamentada e deve haver consequências legais para aqueles que usam deep fakes de forma maliciosa.
Conclusão
Deepfakes are a technology that can have severe impacts on people all over the globe, including damage to reputation, financial loss, and identity theft. The only way to stay safe from deepfakes is by understanding the technology behind deepfakes and using deepfake detection tools; we must always remain vigilant, verify information sources, and stay up-to-date about the latest developments in technology. Ultimately, the key to prevent the spread of deepfakes is through education and awareness, so let’s all stay informed and stay safe in this digital age.
If you enjoyed this article then you may also like to read nosso guia completo para ChatGPT.
