What is Synthetic Media?

What Is Synthetic Media?

Synthetic media is content produced or altered by artificial intelligence rather than recorded or created by humans in the traditional way. The category includes AI-generated images and art, cloned voices, AI-written text, virtual influencers, and deepfakes — realistic videos of people doing or saying things they never did. What makes synthetic media a distinct category is not that it is fake — humans have always made fakes — but that AI can produce it at scale, at low cost, and at a level of realism that defeats ordinary human detection.

The term gained currency in the late 2010s as generative models improved. Earlier, faking media required expertise and equipment; today, free tools can generate a convincing fake video of almost anyone within minutes. This is the "democratization" of forgery: the ability to fabricate evidence — or to fabricate celebrities, politicians, and loved ones — is no longer confined to state intelligence agencies or Hollywood studios.

Synthetic media matters philosophically because of what it does to evidence. Video and audio have functioned as the gold standard of proof: a recording shows what happened. If recordings can no longer be trusted, then the epistemic foundations of journalism, law, and memory are shaken. The problem is not that every video is fake, but that any video could be — and the doubt spreads to all of them.

Historical Background

The manipulation of images is as old as photography. Stalin's propaganda machine erased disgraced officials from photographs; darkroom technicians retouched images for decades; Photoshop made manipulation routine by the 1990s. But each of these techniques required skill and left detectable traces, and the public retained a baseline of trust in photographs and recordings.

The modern era began with the invention of generative adversarial networks (GANs) in 2014. GANs pit two neural networks against each other — one generates images, the other tries to detect them — until the generator produces images the detector cannot distinguish from real ones. The first viral deepfakes appeared in 2017, almost all non-consensual pornographic videos of celebrities. The term "deepfake" — a portmanteau of "deep learning" and "fake" — was born on an anonymous forum.

The 2020s brought the diffusion revolution. Diffusion models, which learn to generate images and video from text descriptions, made synthetic media not just possible but easy, and voice cloning became a consumer feature. Audio deepfakes of executives' voices were used in real-world frauds; political deepfakes began appearing in elections; and synthetic media moved from a novelty to a mainstream risk that platforms, companies, and governments all take seriously.

Key Concepts

Generation and manipulation are the two forms of synthetic media. Generation creates new content — an AI painting, a fake conversation, a synthetic voice reading a script. Manipulation alters existing content — swapping a face, changing the words, reenacting a scene. Both undermine the link between media and reality, but manipulation is the more dangerous for evidence, since it exploits recordings we have reason to trust.

The deepfake is the paradigm case: a synthetic video of a real person. Deepfakes range from harmless jokes to devastating frauds. The defining danger is identity theft of the most intimate kind — a person's face and voice placed in contexts they never chose.

The epistemic problem is the deepest harm. Philosophers of information describe the "liar's dividend": when fakes are plausible, liars benefit even without being believed, because genuine evidence can be dismissed as fake. The politician who can dismiss real footage as a deepfake, and the criminal who can deny real recordings, both exploit the uncertainty synthetic media creates.

Detection is the technical response, and it is an arms race. Detection models identify artifacts — unnatural blinking, inconsistent lighting, audio-visual mismatches — but generators improve to defeat them. The more robust response is provenance: embedding watermarks and metadata at creation so that media carries its own history, combined with authenticated capture for high-stakes contexts.

Consent is the ethical core for victims. Much harmful synthetic media — especially non-consensual intimate imagery — is created without the subject's consent and distributed against their will. The ethical and legal response treats the harm as a violation of the person, not merely a problem of misinformation, and many jurisdictions have criminalized the worst forms.

Contemporary Relevance

Synthetic media is now an everyday force. It is used legitimately in film and advertising, in accessibility tools that give people their own voice when they have lost it, in education and simulation, and in research. It is also used to scam, defame, manipulate elections, and harass. The regulatory response is evolving: some countries criminalize deepfake pornography and election interference; the EU's AI Act imposes transparency obligations on synthetic content; and platforms are rolling out labeling systems.

The deeper question is cultural. Societies are learning to live with a world in which "seeing is no longer believing." The adjustment is not simply to distrust everything — that way lies paranoia and the liar's dividend — but to develop new epistemic habits: checking provenance, relying on authenticated sources, and treating viral media with the same skepticism that earlier generations applied to anonymous rumor.

For individuals, synthetic media is both a tool and a risk. The tools can enrich creativity; the risks require vigilance about what we share, what we believe, and what we are willing to label as evidence. The philosopher's reminder is that trust was never based on the mere existence of a recording; it was based on the chain of custody of that recording. Synthetic media does not destroy trust — it relocates it, from the image itself to the processes that authenticate, label, and verify it.

Sources

• Chesney, Robert, and Danielle Keats Citron. Deep Fakes: A Looming Challenge for Privacy, Democracy, and National Security. California Law Review, 2019. https://doi.org/10.15779/Z38RV0D15J
• Stanford Encyclopedia of Philosophy. Ethics of Artificial Intelligence and Robotics. https://plato.stanford.edu/entries/ethics-ai/

Related Topics

• What Are the Ethics of Deepfakes?
• What Is AI Misinformation?
• How to Detect AI-Generated Content
• Epistemology in the Age of AI
• Information Ethics
• Philosophy of Artificial Intelligence