Embedded Files
FOR JOURNALISTS & BROADCASTERS
FOR JOURNALISTS & BROADCASTERS
THE BBC, AUTHENTICATION & TRUST
THE BBC, AUTHENTICATION & TRUST
WHAT THE BBC WANTS TO SOLVE
To build trust, there needs to be a way to prove where content has come from... or if it has been manipulated.
The BBC is working on this problem, asking itself a series of questions.
For all of the video, audio and text that it receives;
- Can material prove where it has come from?
- Can a signal be added, that will show if the content has been manipulated?
- Can the nature of any manipulation be worked out?
- Can 'bonafide' manipulation be allowed to pass through systems and between sources?
- Can technology or standards be created, that are open source, and that can be used across media and social media ecosystems?
WHAT THE BBC IS NOT DOING
- Applying any new process to ALL outgoing media immediately... though it could be applied to all online media.
- Content without 'trust' signals will still exist (from us and others), but there would be encouragement to signal all content eventually
- Not providing any guarantee of integrity of published stories – this is not a fact checking exercise
- It will not be able to spot all cases of mis-contextualisation (genuine material, but with misleading context) though it maybe able to flag such content in some instances
WHAT ARE THE BBC'S ASSUMPTIONS?
- That 'the padlock' can be assured if delivering directly via BBC websites or via established links to social media. (It is once material is republished that the lock breaks)
- The signals are to be readable by human (the public/journalists) and machine (primarily the platforms)
- That a combination of technologies are needed to address this problem
WHAT KINDS OF TECHNOLOGY EXIST IN THIS SPACE?
WHAT KINDS OF TECHNOLOGY EXIST IN THIS SPACE?
Hashing
Hashing or cryptographic hashing is providing a unique identifier for a piece of content via a character stream. In cryptographic hashing, the hash is intended to ensure the integrity of a piece of content. If the hash has gone, the content has been tampered with. It is specifically designed to match only content that is completely identical on every level (perceptual and technical).
Fingerprinting
Fingerprinting is an attempt to use the characteristics of audio and video themselves to signal its identity. An example is YouTube’s Content ID system that is used in rights management. Fingerprinting is also useful because it can offer a percentage of similarity – which might be useful in determining the integrity of our media.
Locality-sensitive hashing (and developments like Microsoft’s PhotoDNA) roughly fall under the same “similarity” detection approach as fingerprinting.
Watermarking
Watermarking traditionally has meant embedding an identifying mark in media to show ownership or copyright – for example a photographer’s name stamped across an image. However the term is also used to denote an imperceptible piece of data added to a file. The watermarking we’d want to apply would be specifically designed to avoid “fragile” removal – it should be robust to manipulation like transcoding and manipulation, potentially allowing us to find where our content is being manipulated.
Might the Blockchain help?
The concept of a verifiable ledger that records genuine articles, and allows public auditing is almost certainly part of our work here – whether it’s a blockchain and whether its private is still something we’re working on.
Using a permissioned blockchain (a private blockchain which is quicker, less resource intensive and more data compliant than the public permissionless Blockchain used by Bitcoin, for eg), might help us keep a ledger of ‘genuine’ articles but there may be other solutions.
VERIFYING SOURCE MATERIAL: TOOLS
VERIFYING SOURCE MATERIAL: TOOLS
Alongside human verification, we also have technology based verification methods such as watermarks and finger-pointing.
Some of the tools used by BBC Monitoring include;
- Google images reverse search: https://images.google.com/
- Exiff metadata viewer: http://exif.regex.info/exif.cgi
- Exiff purge: http://www.exifpurge.com/
- FotoForensics: http://fotoforensics.com/
- Image forensics: https://29a.ch/photo-forensics/#forensic-magnifier
- InVid verification plugin: https://www.invid-project.eu/tools-and-services/invid-verification-plugin/
- TinEye: https://www.tineye.com/
- Reveal image verification assistant: http://reveal-mklab.iti.gr/reveal/index.html
- Serelay - 'Trusted Media Capture': https://www.serelay.com/
Video demonstrating a plugin from InVid that helps identify parts of 'flipped' images stolen from other sources
FOR THE PUBLIC
FOR THE PUBLIC
SOURCES & LITERACY
SOURCES & LITERACY
Originally, most of our content would only appear on our own platforms. In effect, a CLOSED SYSTEM which the audience understood and TRUSTED. Therefore 'fake news' was very easy to spot if it appeared anywhere else that was not an original platform.
Now, many media organisations legitimately publish to MULTIPLE PLATFORMS, but this is where confusion grows, particularly in social media.
How can it be easily proved that a 'BBC story' on a non-BBC platform actually originated from the BBC, and not from someone else?
In a world of AI, these methods will have to go further, exploring cryptographic sign-in, or perhaps some form of blockchain.
However this space is ill-defined, with few standards or approaches that are consistent - and that still require MEDIA LITERACY to augment any technology solutions.
Google Sites
Report abuse