Have you seen that viral video of Tom Cruise flawlessly doing coin magic tricks? Or Jon Snow apologizing for the Game of Thrones finale? If so, you‘ve witnessed deep fakes – false depictions of events, quotes or actions generated by advanced AI. While some deep fakes provide entertainment, the technology also fuels rising epidemics of fraud, psychological abuse and institutional distrust.
What Exactly Are Deep Fakes?
Simply put, deep fakes leverage deep learning algorithms to create fabricated images, audio or video. The AI studies vast datasets of a person‘s face, voice or behaviors – breaking these down to encoded mathematical representations capturing subtle similarities. It then uses these learned encoded patterns to render new synthetic media mimicking how someone looks, sounds and moves with high precision.
So if you captured enough home videos and photos of your grandma over the years, a deep fake algorithm could produce new footage of her saying or doing things she never actually did in flawless imitation. Now scale this up to generating forged videos of celebrities, politicians and your local bank branch manager!
While basic computerized cut-and-paste media manipulation predates social media, deep fakes represent exponential leaps in sophistication. Their realism stems from recent breakthroughs in AI fields like computer vision and graphics processing. In particular, deep fakes rely on generative adversarial networks (GANs) – algorithms pitted against each other to refine outputs until convincingly human.
GAN frameworks optimize synthetic media by pits generator vs discriminator algorithms against each other
So in a nutshell, deep fakes leverage AI to falsify video/audio/images of people with alarming realism. Their risks stem from enabling large-scale impersonation to deceive or coerce targets at unprecedented scale.
The Origin Story of Deep Fakes
While photographic doctoring predates Instagram face filters by over a century, many highlight a 2017 Reddit user named "DeepFakes" as catalyzing today‘s boom. Leveraging open source AI tools, they began anonymously producing hardcore porn videos overlaying celebrity faces on adult film star bodies without consent.
Despite public outrage and lawsuits, the visually striking clips continued spreading across social media. Their remnants still widely circulated today normalized the nonconsensual use of deep fake technology troublingly.
Yet the computer vision techniques powering deep fakes originate from longstanding academic research into neural networks and graphics. Significant milestones include:
- 1990s: Researchers publish foundational innovations in deep learning for image analysis and synthesis
- 2014-16: AI papers establish key GAN algorithms; Source code circulates online
- 2016-17: DeepFakes sparks ethical controversies applying AI techniques to pornography
- 2018: Zao mobile app launches allowing Chinese users to easily face-swap themselves into films using minimal photos
- 2020s: Deep fake detection lags behind generation capabilities as apps further democratize sophisticated forgeries
So while "DeepFakes" notoriously sparked today‘s boom, deep fake tech builds on decades of pioneers publishing open academic research later weaponized by malicious actors.
Deep Fakes By the Numbers
- 95,000+: Google searches per month estimated for "revenge porn" frequently involving deep fakes
- 96%: Of deep fake videos analyzed by cybersecurity firm Deeptrace had nonconsensual pornography content
- 3,000%+ Growth in deep fakes from late 2018 to late 2019 alone as algorithms and apps improved exponentially
- 13,000: Fake nude images of women shared online without consent…per day
- $250K: Maximum damages possible under Virginia‘s new deep fake laws
- 63%: People trusting video evidence less due to concerns over deep fakes in 2021 CIGI-Ipsos survey
Over 60% of people trust images and video less due to deep fake technology
Real-Life Examples of Deep Fakes in Action
Deep fakes have enables a sweeping range of fraud, coercion propaganda and injustice globally:
Extortion Scams
- Scammers used a deep faked CEO video to convince a UK energy firm employee to transfer €200,000 to a Hungarian supplier‘s account.
State Propaganda
- During Russia‘s invasion of Ukraine, a crude Zelensky surrender speech deep fake circulated on social media. Within just a day, the misinformation reached over 200,000 views propagating dangerous lies.
Nonconsensual Pornography
- Apps like DeepNude facilitate producing fake nude photos from images of clothed women without consent at scale. Over 100,000 deep fakes already spread via Telegram channels.
Anonymous Activism
- The anonymized documentary "Welcome to Chechnya" leveraged deep fake technology to protects identities of vulnerable sources. The film exposed anti-LGBTQ persecution bringing crucial human rights attention.
From sexual exploitation to multinational fraud to political sabotage, these examples reveal deep fakes‘ rising real-world harms. Their threats compound as algorithms grow more accessible through consumer apps while many public figures and organizations remain unaware of associated risks.
Deep Fake Detection Tips
Fortunately, visual artifacts in synthetic media generated to date provide clues allowing savvy viewers to spot potential deep fakes:
- Lighting/textures don‘t match across facial regions
- Strange skin tones or follicles around boundaries
- Asymmetrical detailed missed (odd earrings, wrinkles etc)
- Unnatural reactions and movements give CGI feel
However amateurish quality on platforms like Reddit or TikTok promote dangerous assumptions more polished deep fakes couldn‘t plausibly fool experts or achieve virality. In reality, convincing forgeries have repeatedly circulated on mass media through Facebook, Twitter and broadcast news before post-hoc debunking.
This illustrates risks even quite flawed synthetic media poses in flooding the information ecosystem with misinformation faster than fact checkers can purge them. And incentives remain for actors to produce slick forgeries indistinguishable from reality to even trained eyes.
Deep Fakes and the Infocalypse
In closing, deep fakes have emerged as an ascendant technology capturing grave threats algorithmic disinformation poses digital societies. Individually, each incremental AI and graphics breakthrough proved harmless enough curiosity. Yet combined then weaponized by malicious actors, deep fakes now scale gaslighting, fraud and coercion once limited by human capacity.
Left unchecked by meaningful countermeasures and accountability, synthetic media seems poised to flood the digital world in coming years further eroding factuality, privacy and consent. Like climate change, deep fakes represent a slowly mounting "infocalypse" whose most severe potential consequences require urgent action today before descending into dystopia tomorrow.
