Skip to main content

How creative AI techniques became a threat for national security

 In 2014, a 28-year-old American researcher called Ian Goodfellow published a paper called Generative Adversarial Networks (GANs) that presented a powerful technique to generate brand new images (i.e., human faces) with Artificial Intelligence.

Goodfellow's idea got a lot of attention from the academic and industry communities and new research papers followed improving the state of the art and, as it can be seen in the images below, the perceived quality of the human faces that GANs generated.

Needless to say that none of these six people actually exists. However, for the last three ones, only a skilled computer forensic would be able to detect the deception.

GANs are making a tremendous positive impact in different domains, such as healthcare or in creative industries. Unfortunately, malicious applications of this apparently harmless technique are emerging and posing a thread to our personal and even national security.

Scams. Deep Learning has become a gold mine for criminal organizations that combine hyperrealistic GAN-generated faces with voice synthesizers to perform scams online. The so-called romance scams, for example, cost Americans $143M in 2018.

Revenge porn. One of the most shameful applications of this technology is to ruin people's reputation by creating footages of sexual acts with the target person's face. According to this research, 100% of the victims of these videos were women.

Fake news. Deepfake videos are a threat to national security, according to the Pentagon. The defense and intelligence services fear scenarios targeting democratic processes (manipulate elections) and the incitement of crowd mobilization under false pretenses. This deepfake video of former US President Barack Obama is an example of the dangers this technique poses.

Improving the deception detection techniques, together with more strict laws controlling unethical uses of these technologies, is essential in order to limit the negative impacts on our society.

Nevertheless, it is not only up to the research community and the authorities to tackle this issue. We citizens must strengthen our awareness of these threats and be more suspicious when consuming digital content from untrusted sources.

Popular posts from this blog

How to disable cookies on Google Analytics so that you don't need a consent banner

The integration of Google Analytics into a website or blog is not GDPR-compliant by default . You must first obtain explicit consent of the end-users to store cookies, describing in your privacy policy how you intend to use collected personal data. This is the reason why most websites nowadays display an annoying (but necessary) consent banner. If you fail to do so or if you only ask for implicit consent, you are at risk of being fined. However, it is possible to disable cookies on Google Analytics (GA) respecting end-users privacy, so that you don't need to ask for consent. The downside is that you will not be able to distinguish the type of user (unique vs new vs returning) and you will miss some session insights. If these details are not relevant for you, here is how you do it. Disable Google Analytics cookies on a custom website If you have a custom website with full access to the source code, you can simply insert the script below between the <head>  and </head>

How to jump to time offsets in HTML5 video

Let's say that you have a 30-minute WEBM video file, from which you just want to play the following video segments , jumping from one to the other automatically  without interruptions : [00:01:25.00 - 00:02:25.00] -> from second 85 to 145 [00:11:40.00 - 00:11:55.00] -> from second 700 to 715 [00:20:26.00 - 00:21:07.00] -> from second 1226 to 1267 [00:26:11.00 - 00:28:01.00] -> from second 1571 to 1681 To increase the complexity, let's think that you have these video segments in a PHP variable $arrayVideoSegments  (normally the case if they were retrieved from the database).   $arrayVideoSegments[0]->startTime = 85   $arrayVideoSegments[0]->endTime = 145   $arrayVideoSegments[1]->startTime = 700   $arrayVideoSegments[1]->endTime = 715   $arrayVideoSegments[2]->startTime = 1226   $arrayVideoSegments[2]->endTime = 1267   $arrayVideoSegments[3]->startTime = 1571   $arrayVideoSegments[3]->endTime = 1681 The fo

Temario del curso "Startup Innovation Lab"

En 2017 publiqué, en el marco del curso Startup Innovation Lab , este vídeo explicando el Lean Canvas en español usando Spotify como ejemplo ilustrativo.  Me ha sorprendido muy gratamente comprobar que, cinco a ñ os después, el vídeo se acerca a las 100.000 reproducciones. Algunas personas me han escrito preguntando dónde pueden acceder al resto de materiales del curso. El curso ya no se imparte, pero he decidido publicar el temario completo en PDF por si fuese de utilidad. En total son 10 módulos  que cubren distintas temáticas sobre la puesta en marcha de ideas de negocio innovadoras. 1. Introducción a Lean Startup 2. Lean Canvas y Business Model Canvas 3. Métricas de Negocio 4. Estudio de Mercado 5. Producto Mínimo Viable 6. Constitución de la Empresa 7. Fuentes de Financiación 8. Modelo de Ingresos y Plan Financiero 9. Plan de Negocio 10. Pitch de Presentación del Negocio Tengan en cuenta que este contenido no ha sido actualizado desde su creación en 2017 y se ofrece si