SWITCHED ON
The daily technology series nobody asked for but everyone needed
Nobody Here But Us Algorithms: The Age of Synthetic Reality
The moment you can't tell if the person you're watching, reading, or talking to is human at all has already arrived. We just haven't fully admitted it yet.
The technology doesn't just threaten the authenticity of synthetic content. It retroactively undermines the credibility of real content.
In Episode 1 we established that AI is quietly eating the career ladder from the bottom up while economists argue about whether it's technically doing that. In Episode 2 we discovered that we handed an entire generation of children a dopamine slot machine, watched what happened, and are now in the middle of a heated political debate about whether to feel bad about it. If you haven't read those yet, go back. They'll still be there. Unlike, increasingly, your ability to tell what's real online.
Because today we need to talk about something that has been creeping up on us for the better part of five years. Something that the tech industry discusses in hushed, slightly embarrassed tones. Something that most people haven't fully reckoned with yet, despite it already reshaping the information environment they live in every single day.
Today we're talking about synthetic reality. AI-generated content. Deepfakes. Digital humans. The rapidly approaching — and in some corners already arrived — moment when you genuinely cannot tell whether the person you are watching speak, the article you are reading, the voice you are hearing on a phone call, or the photograph you are looking at depicts something that actually happened, or something that a machine decided should appear to have happened.
This is not science fiction. This is Tuesday.
01 — The Technology, Briefly, Because You Need the Context
In 2021, AI-generated images were impressive but obviously synthetic. The hands were wrong. The teeth were wrong. The eyes had that specific uncanny valley quality that made your skin crawl in a way you couldn't quite articulate but absolutely felt. You could spot them, mostly, if you knew what to look for.
In 2026, you frequently cannot. The hands are fine. The teeth are fine. The lighting is coherent. The background makes sense. The facial expressions track. Video synthesis has reached a point where, without forensic analysis tools that most people don't have access to, the output is indistinguishable from real footage to the human eye.
Voice cloning technology can now produce a convincing replica of a specific person's voice from a sample of less than thirty seconds of source audio. That is a single voicemail. A short video clip. A fragment of a podcast appearance. Enough.
And text — well, you're reading text generated by a large language model right now, which makes this particular point somewhat self-referential, but the capability to produce fluent, convincing, authoritative-sounding written content at industrial scale has been with us for several years. The question of whether what you're reading was written by a human or a machine is now, genuinely, non-trivial for a significant proportion of content online.
02 — The Liar's Dividend
Here's the thing about synthetic media that gets lost in the more breathless coverage: the threat isn't primarily the individual sophisticated deepfake. It's not mainly about some nation-state producing a flawless fake video of a world leader declaring war. Those scenarios are real concerns, but they're the dramatic, obvious end of the spectrum.
The more insidious problem is what researchers call the liar's dividend. Even if a piece of media is completely real — genuine footage of a real person doing or saying something that actually happened — the existence of deepfake technology now gives any sufficiently motivated person a plausible basis to claim it's fake.
A politician caught on camera saying something embarrassing can now point at the footage and say "that's AI generated." A genuine photograph of a newsworthy event can be dismissed as synthetic. A real audio recording of a genuine conversation can be waved away as a voice clone. The technology poisons the well not just for fabrications but for reality itself.
If we cannot agree on what is real, we cannot agree on what happened, which means we cannot hold anyone accountable for anything, which means power operates without meaningful check or consequence.
That is not a small thing. That is, potentially, a very large thing indeed.
03 — The Platforms Are Not Ready
The platforms have made various announcements about AI content detection and labelling policies. Meta announced that AI-generated content must be disclosed. TikTok implemented labels for AI-generated media. YouTube introduced policies requiring creators to disclose synthetic content. These policies share a common feature: they rely primarily on voluntary disclosure by the person posting the content. The person most likely to post a misleading deepfake — someone with an intent to deceive — is being asked to politely label their deception as such. The deterrent effect of this is, charitably speaking, limited.
Automated detection is the obvious alternative — use AI to detect AI — and it is being pursued actively. The problem is fundamental: detection models and generation models are locked in an arms race in which the generation side has structural advantages. Every time a detection model gets good at identifying a particular type of synthetic content, the generation models update and the artifacts disappear. Detection is always, by definition, chasing generation. It cannot get ahead.
There are watermarking approaches that show genuine promise. Google's SynthID embeds imperceptible watermarks in AI-generated images and audio. The EU's AI Act includes provisions requiring watermarking of certain AI-generated content. But it requires cooperation from the companies producing the generation tools, consistent enforcement, and global coordination across jurisdictions that have very different interests. None of which is happening at anything close to the required speed.
04 — The Human Dimension Nobody Talks About Enough
The conversation about synthetic media focuses almost exclusively on political disinformation — elections, world leaders, geopolitical manipulation. It is legitimate and important. It is also incomplete. Because the most prevalent use of non-consensual synthetic media right now is not political. It is sexual. It is overwhelmingly targeted at women. It is devastating in its personal impact. And it is, in most jurisdictions around the world, operating in a legal grey zone that has been far too slow to resolve.
Deepfake pornography has existed as a technology for several years and as a mass phenomenon for roughly three. The targets are celebrities, public figures, and increasingly, private individuals — ex-partners, colleagues, classmates. The harm is real, documented, and severe. The UK passed a law in 2024 criminalising the creation of non-consensual intimate deepfakes. Several US states have enacted similar protections. But enforcement across borders, against anonymous creators, remains extraordinarily difficult. The technology has dramatically outpaced both the law and the social norms that laws eventually codify.
05 — So Where Does That Leave Trust?
The honest answer is that we are in a transitional period that is going to be genuinely uncomfortable for an indeterminate amount of time. The tools to navigate a synthetic media environment — AI literacy, verification skills, forensic awareness, institutional frameworks for authentication — are being built more slowly than the problem is growing.
It means developing a healthier scepticism about unverified media without tipping into the kind of blanket cynicism that dismisses everything as potentially fake and therefore treats nothing as real. That balance — between appropriate scepticism and functional trust — is going to be one of the defining epistemic challenges of the next decade. And it means supporting enforceable law, technical standards, and international coordination. None of that happens without sustained public pressure.
Tomorrow we're wading into something both simpler and more personally uncomfortable than synthetic reality: your data. Specifically who has it, what they're doing with it, and the growing movement to establish a legal right to make companies forget you ever existed. The right to be forgotten. And why the tech industry really, really doesn't want you to have it. See you then.
Switched On is a daily technology series covering AI, social media, data privacy, and the digital forces reshaping modern life — with no corporate spin, no false comfort, and absolutely no mercy for buzzwords.
