Deepfakes & Reality

Deepfakes are synthetic media — video, audio, or images — generated by artificial intelligence, typically deep learning neural networks, that plausibly depict real people doing or saying things they never did or said. The technology has existed in primitive form since the 1990s, but the combination of generative adversarial networks (GANs), diffusion models, and the massive increase in available computational power has made deepfake generation accessible to anyone with a consumer-grade computer and a few hours of training data.

Current deepfake technology can produce video of public figures saying anything the creator chooses, with lip-sync and facial expression matching that is often indistinguishable from genuine footage. Voice cloning can replicate any person's voice from a few minutes of audio. Image synthesis can produce photorealistic images of people, events, and places that never existed. Text generation — which AI was originally designed for — can produce convincing written content attributed to any author in any style.

The progression is important: what required a professional effects studio and months of work in 2015 required a consumer GPU and hours of work in 2020, and requires a smartphone app and minutes of work in 2025. The barrier to fabrication is approaching zero. The implications of zero-cost fabrication for the concept of evidence — and for the shared reality that democratic society depends on — are not yet fully understood.

Epistemology — the philosophy of knowledge — asks: how do we know what we know? Before deepfakes, visual and auditory evidence was one of the most reliable categories of knowledge available: I can trust my senses, and recordings extend the reliability of sensory experience to times and places I was not present. This assumed reliability underwrote journalism, legal proceedings, historical documentation, and the ordinary social process of believing what we see and hear.

Deepfakes undermine this foundation not by replacing reliable evidence with unreliable evidence — which would be serious but manageable — but by making the reliability of all evidence uncertain. The epistemological problem is not "this specific video is false." It is "any video might be false, and I cannot reliably determine which ones are." This generalised uncertainty is more corrosive to shared reality than any specific fabrication, because it cannot be addressed by fact-checking specific content. It requires a fundamental restructuring of how we establish what is true.

The "liar's dividend" — a term coined by legal scholars Robert Chesney and Danielle Citron — describes the benefit that bad actors derive not from creating deepfakes but from the existence of deepfake technology as such. In a world where deepfakes are known to exist and to be convincing, a real video of a genuine event can be dismissed as a deepfake. A genuine audio recording of a genuine conversation can be denied. A real photograph of a real moment can be declared synthetic.

The liar's dividend may be the most significant long-term consequence of deepfake technology — more significant than any specific fabrication. It is the political weaponisation of epistemological uncertainty: the ability to deny genuine evidence by invoking the possibility that it might be fake. This is not a hypothetical future risk. It is already being used: public figures facing genuine video evidence of misconduct have dismissed it as AI-generated. The technology creates a universal alibi for those who produce genuine content and wish it had not been recorded.

The deepest consequence of living in a world of pervasive synthetic media is not the specific false things people will believe but the generalised erosion of epistemic trust — the willingness to believe that what others report about reality is accurate. Democratic society, journalism, science, and ordinary human cooperation all depend on a baseline of shared epistemic trust: the assumption that most people, most of the time, are reporting what they genuinely believe to be true about a world that genuinely exists.

When this baseline is seriously eroded, the social consequences are severe: withdrawal into epistemically closed communities that only trust information from within their own group; the collapse of shared factual ground necessary for democratic deliberation; the flourishing of conspiratorial thinking as a rational response to genuinely unreliable evidence; and the retreat to personal relationship and direct experience as the only reliable sources of truth — which, paradoxically, may be the deepfake era's most important spiritual contribution.

The philosophical and spiritual traditions have long distinguished between two kinds of knowing: knowledge derived from external evidence and testimony (what the Vedanta tradition calls pramana, the Gnostic tradition calls pistis, and ordinary epistemology calls justified true belief) and knowledge derived from direct inner perception (gnosis, direct knowing, contemplative insight). The first kind is mediated — it depends on the reliability of external sources. The second is immediate — it is the direct perception of what is true by the consciousness that is aware of it.

The deepfake crisis attacks the first kind of knowing comprehensively. It has nothing to say about the second. The direct perception of one's own experience — the felt sense of what is true in one's own life, the direct knowing that comes from genuine contemplative development, the relational truth that builds through years of genuine encounter — none of this is threatened by AI-generated video. It operates on a different track entirely: not the track of evidence and testimony but the track of direct experience.

This is not a call to abandon concern about mediated reality — the political, journalistic, and legal consequences of deepfake technology are real and serious. It is a recognition that the spiritual traditions' emphasis on developing inner faculties of discernment — on building the capacity to know from the inside rather than depending entirely on what can be seen from the outside — has never been more practically relevant than it is in the deepfake era.

Navigating the deepfake era requires both external practices (media literacy, provenance verification, source evaluation) and internal ones (developing genuine discernment, building trust through direct relationship, maintaining epistemic humility about mediated claims). Neither is sufficient alone. The external practices address the specific problem of synthetic media. The internal practices address the deeper problem of what to do when external evidence is insufficient — which is the condition the deepfake era is making permanent.