Deepfakes and manipulated images are increasingly threatening democratic elections – and can hardly be seen with the naked eye. The German Research Center for Artificial Intelligence (DFKI) has therefore developed an AI detector that checks suspicious images directly via Instagram in just two clicks.
The increasing spread of artificial intelligence is also changing the rules of the game for democratic elections and also poses considerable risks. Today, AI can generate deceptively real content and, for example, use deepfakes of politicians to specifically influence opinions.
The big problem is that this content is often difficult to distinguish from real content. They are also spreading rapidly via social networks.
In recent years it has already become clear that so-called disinformation campaigns can put electoral processes under pressure and undermine trust in democratic institutions. The German Research Center for Artificial Intelligence (DFKI) has now developed an AI detector to expose such content, with which users can check suspicious images in just a few clicks.
What can the DFKI’s new AI detector do?
For the “Check First. Vote Smart” project, the DFKI worked with the DFKI spin-off Gretchen AI and the Rhineland-Palatinate State Center for Civic Education. The result is an AI detector that can check suspicious images, for example from Instagram, in just a few clicks.
“AI-controlled bots simulate social majorities, deepfakes distort statements, foreign actors deliberately intervene in election campaigns,” explains Dr. Tobias Wirth, subject area leader in the Smart Data & Knowledge Services research area at DFKI Kaiserslautern. “We are developing AI that detects these manipulations – before they get caught.”
However, the developed tool does not limit its inspection to pixel artifacts, as is the case with conventional approaches. Rather, forensic image and audio analysis are combined with a kind of “logical counter-reading”.
While the image is being checked, AI agents can simultaneously research parallels on the Internet and incorporate verified fact banks. Here, among other things, close cooperation with the dpa fact-checking team comes into play.
How to check suspicious images in two clicks
If you have discovered a suspicious post on Instagram, you can send it to the Checker account “gretchen_ai_berlin” using the share function. The AI detector will then give you an assessment of the likelihood of a fake and a brief assessment of influencing factors.
In the beta phase, the detector should already be able to detect more than 80 different types of manipulation. According to the DFKI, this ranges from classic deepfake videos to seemingly harmless screenshots.
For example, interventions such as face swaps or lip sync manipulations are detected. But the detector should also be able to detect image splicing, in which real photos are put together to create a misleading scene.
The analysis should remain particularly low-threshold without an additional app or registration. “We only use functions that Instagram already provides – direct message, sharing, forwarding,” emphasizes the team that is making the implementation GDPR-compliant.
However, this only applies to image analysis. If you want to access video analysis and analysis for previous contexts and original material, you must access Gretchen AI’s dashboard solution.
Also interesting:
