menu ☰
menu ˟

Thanks to AI, the Future of 'Fake News' May Be Easily-Faked Video

08 Feb 2018

Julian Sanchez

From the printing press and home VCRs to Snapchat and virtual
reality, the pervasive desire to look at attractive naked people
has been a great unsung driver of technological progress. If you
want to know where technology is going, in other words, a good rule
of thumb is: Look to porn.

When it comes to the future of news, however, that advice may
leave you feeling unsettled — and for reasons having nothing
to do with prudery.

As the technology news site Motherboard reported late last month, the
latest merger of high tech and low urges is a phenomenon dubbed
“deepfakes.” Using free, readily available software,
the everyday horndog can now swap the faces of celebrities —
or anyone else — into pornographic videos. While once such
fakery would have required advanced video editing skills, the
FakeApp, designed for the convenience of deepfake aficionados,
makes use of machine learning algorithms to produce what is, in
effect, a video editing Artificial Intelligence.

The upshot is that shoehorning an onscreen — or real life
— crush into an ersatz but highly convincing porn no longer
requires a serious technical background.

That ought to be disturbing enough: Most of us would rather not
contemplate the prospect of discovering we’ve been
unwillingly cast in an obscene video that’s gone viral
online, even if it’s known to be a fake. (Some major porn
sites are now barring the phony videos, though plenty remain in
circulation.)

But perhaps even more unsettling should be the inevitable
application of this free-to-download tech to politics and
journalism. Combined with software like Adobe
Voco
, which can create a pitch-perfect virtual simulation of
anyone’s voice based on a short audio sample, you’ve
got a recipe for realistic viral “fake news” fodder
that the average prankster can manufacture in an afternoon.

Technology has made it
easier to fake; the economics of the internet make it increasingly
likely that the fakes become news. And the inevitable blunders will
confirm diminishing public trust in professional news media —
the effect of which to date, ironically, has been to drive many
viewers and readers into the arms of outlets with even fewer
journalistic scruples.

Just imagine the October Surprise potential: The candidate
caught cavorting with prostitutes, spewing racial epithets,
outlining a plan to round up Lutherans for secret medical
experiments! Even the most brazen political campaign might fear the
damage of such a forgery being traced back to its own doorstep
— but when the software to pull it off is available to anyone
with a broadband connection, they likely won’t have to.

In an ecosystem flooded with forged amateur videos, of course,
many viewers will naturally become more skeptical about the idea
that “seeing is believing.” But that, too, has a cost:
Recall Donald Trump’s strange, belated efforts to raise doubts among his associates about the
veracity of the infamous Access Hollywood “grab ‘em by
the pussy” tape.

In a world of fake video, such a denial might well have seemed
plausible, at least to those who wished to believe. A sufficiently
shameless politician might deny even actions caught on tape, with
supporters given license to trust their preconceptions over their
eyes.

Democratized digital fakery is nothing new, of course: Photoshopped images of political figures have
long been a staple of those chain e-mails your uncle forwards along
periodically. But they’ve typically remain confined to the
fringes of political discourse for a few important reasons. One is
that amateur Photoshop jobs are usually not too hard for even
untrained eyes to detect: Zoom in close enough, and the pixelated
hallmarks of a sloppy edit are apparent.

Just as importantly, however, is the fact that it’s harder than
you might initially think to construct a still image that’s
unambiguously scandalous without being so comically heavy-handed as
to raise instinctive suspicion in the minimally savvy viewer. (For
instance, most of us understand that, when politicians take bribes,
they rarely come in the form of giant sacks of cash emblazoned with
dollar signs.)

Audio alone, by contrast, offers more opportunity for creating
plausibly scandalous content — it’s much easier to concoct
damaging things that a politician might unwisely say out loud in an
unguarded moment — but we’re all accustomed enough to hearing
uncanny impersonations of famous people that an audio recording
alone lacks persuasive power without a relatively ironclad
provenance.

All of that, taken together, make mainstream media outlets less
likely to be taken in by and amplify such forgeries. Thus, what
harm they do stays confined to chain e-mails.

Video combined with audio, however, is another matter,
especially as algorithmic assistants get better at concealing the
more obvious digital artifacts of editing. Even in an era of
sophisticated CGI, we are all still inclined to believe what we can
both see and hear.

Maybe more importantly, something that’s caught on video
makes for good television. And the technology is arriving precisely
as the incentives that media outlets face make them less able to
resist paying attention to something that’s gone viral.

Recall the path taken by the now-infamous “Steele
Dossier,” the research compiled by a former British
Intelligence officer purporting to document collusion between the
Trump campaign and the Russian government. Many media outlets had
obtained copies of the dossier, but because — however much of
it ultimately proves accurate — they could not verify its
claims, it remained unpublished.

Until, that is, the online news site Buzzfeed decided that the
dossier was sufficiently newsworthy to publish with the caveat that
its allegations could not be confirmed. Instantly, the fact that
one news organization had run with the story was itself a
newsworthy development that others could justify covering.

In the Internet Era, there are no more regional media
oligopolies. Every news outlet is, essentially, in competition with
hundreds, if not thousands, of others. That makes the traditional
benign paternalism exercised by news organizations much harder to
sustain economically: If you don’t run with that explosive
video, your competitors may — and when there are thousands of
competitors, it becomes a near certainty that at least one
will.

A site with dubious journalistic standards deciding that a fake
clip is “newsworthy” merely on the grounds that it has
gone viral on social media can plausibly kick off a chain reaction,
as more credible outlets rush to cover the coverage, lest they be
left last in the increasingly pitiless competition for
eyeballs.

Technology has made it easier to fake; the economics of the
internet make it increasingly likely that the fakes become news.
And the inevitable blunders will confirm diminishing public trust
in professional news media — the effect of which to date,
ironically, has been to drive many viewers and readers into the
arms of outlets with even fewer journalistic scruples.

Eventually, of course, both news producers and news consumers
will adapt to the new reality, with some combination of
professional protocols and personal skepticism. But the chaotic
period of fumbling toward a new equilibrium promises to be a wild
ride.

Julian
Sanchez
is a senior fellow at the Cato Institute studying
issues at the busy intersection of technology, privacy, and civil
liberties.

Click here to view the full article which appeared in CATO Journal