A
Milestone Reached
Chief Scientific Officer, Microsoft
Published on January 31, 2022
New digital content provenance
systems certify the origin and history of changes made to media and can alert
viewers to tampering.
|
S |
trong democracies are built on an informed and
compassionate citizenry. Disinformation campaigns, designed to mislead,
persuade, and sow division, can disrupt our understandings, and fan the flames
of discontent. We just hit a significant milestone in technologies that can
bring more trust to digital content. The members of the Coalition for Content
Provenance and Authenticity (C2PA), including Adobe, Arm, BBC, Intel,
Microsoft, Truepic, and Twitter, announced the release of a Content Provenance
Specification and offered a preview of the new technology at a recent event. I
was invited to kick off the summit with some framing thoughts about why I
believe media provenance is so important in our quest to protect people and
society from deceptive photos, videos, and audio engineered by malevolent
actors.
Over
the last five years, advances in machine learning and graphics have led to the
general availability of surprisingly powerful tools for modifying and synthesizing
digital content that’s so realistic, it is difficult to distinguish fact from
fiction. The proliferation of techniques for generating deepfakes—and also to make lighter-weight edits and
manipulations—threaten to cast doubt on the veracity of all digital media.
A
year ago, we helped
to found C2PA with other partners as a standards body pursuing methods
that could bolster trust and transparency in digital content. While we had
our eye on deepfakes, we also sought to address the challenges coming with
other disinformation tactics, such as the use of photos or snippets of video
from long ago as breaking news—or the calling out photos of real events as
fabricated.
The
specification, the world’s
first standard for certifying digital content provenance, will foster the
creation of interoperable tools for media provenance, enabling content creators
to cryptographically certify the source and history of changes made to digital
assets, like images and videos, and to confirm that content has not been
tampered with. The methods work by immutably embedding information about the
media and its source and history as metadata that travels along with the
digital content.
Methods
for certifying content provenance will fill a gap in our ability to detect
deepfakes. Some have hoped that we might thwart the flow of synthesized or
manipulated digital content by using AI methods to detect artifacts or
irregularities. The problem is that our best AI-based techniques for creating
deepfakes actually use our best detectors of fakes in
the very process of generating the realistic content. Better detectors
will only mean that the fabrications will get even more realistic. In short, we
end up in an AI versus AI scenario, where AI cannot reliably win. Thus, we
realized the need to develop a different kind of approach—and this led to
innovation with the authentication of content provenance.
The
new content provenance methods will not solve the deepfake and broader
disinformation threats on their own. It’s going to take a multi-pronged
approach, including education aimed at media literacy, awareness, and
vigilance, investments in quality journalism--with trusted reporters on the
ground, locally, nationally, and internationally, and new kinds of regulations
that make it unlawful to generate or manipulate digital content with an aim to
deceive. However, I see today’s announcement, and the technologies that will
flow from it, as a significant step forward in injecting a new layer of
resilience and trust into the digital content that we see.
None
of this would be possible without the cooperation and productivity of the
multi-party C2PA standards body, which initially brought together top talent
from two complementary projects, Project
Origin and the Content
Authenticity Initiative (CAI)—and then expanded to include additional
stakeholders, including hardware producers, broadcasters, news agencies, and
providers of software and services.
We
recognize that restoring trust in digital content is an ambitious goal that
will require diverse perspectives and participation. We are encouraged to see
industry, non-profit organizations, research institutions, and governments
embracing the content provenance approach.
Progress
is already being made in the United States and in other areas of the world. For
example, Senators Rob Portman (R-Ohio) and Gary Peters (D-Mich) co-sponsored
the bipartisan Deepfake
Task Force Act earlier this year. The National
Security Commission on AI (NSCAI), a panel on which I served, examined
and reported on ways that our nation could leverage digital media provenance to
mitigate the rising challenge of synthetic media. And in Europe, several
national and European Union initiatives have been launched around introducing
digital content provenance into legislation.
I
believe that content provenance will have an important role to play in
fostering transparency and fortifying trust in what we see and hear
online. I encourage all to review of the presentations at the C2PA virtual summit, which features
experts in government, academia, media, and technology on this approach, its
benefits, and its limitations. Technology previews are provided, showing
how the methods work and how their use can be scaled across the
internet. I particularly urge those involved in the business of capturing,
creating, and transmitting news and information to take a careful look at this
exciting development.
___________________________________
eric@horvitz.org