In the context of the upcoming US presidential elections, the issue of cybersecurity, online fraud, fake news and so-called deepfakes is on the rise. Concerns about manipulation are not surprising, because the news that appears in the public space is not positive. More recently, a study by the AI Democracy Project revealed that artificial intelligence is wrong on every second election question. In a world dominated by this technology, the risk increases significantly.
Continued reports of multiple cyberattacks are also a cause for concern. We survived in Poland the case of a disinformation telegram allegedly sent by the PAPand in the United States, in just a few months, corporations such as Roku, Ticketmaster and recently even Disney have been attacked. Well-known brands, celebrities and people in public positions often fall victim to fake news and deepfakes. Some of them, like Rafał Brzoska, are fighting against this practice and want to assert their rights in court. However, there are also other methods to deal with this type of fraud.
Read also: Poland on the infamous list. Deepfake used by politicians
The best deepfake detection tools
Deepfake creation programs are not perfect, and sometimes you can still judge for yourself whether, for example, a catchy headline about a famous person offered to you on Facebook is true. However, the development of this technology is inevitable, so the quality of fake videos and photos will only get worse. Soon, the best way to find out what’s true on the internet will be to clearly define what’s not true..
The absurdity of this situation is that both the tools for creating deepfakes and detecting them are often based on generative artificial intelligence. The latter is software or systems designed to identify and detect fake videos or images. It typically uses various methods to analyze digital content and determine whether it has been manipulated or generated by AI.
There is no shortage of programs on the Internet created for this purpose. How do they work, how do they differ from each other, and how reliable can they be? We recommend our list.
Sentinel, Sensity and WeVerify
Among the tools cited as the most effective in identifying fake videos and photos, it is often cited Sentinel technology. It is no coincidence that the program is used by security agencies, non-governmental organizations and media outlets across Europe. The advanced AI-based detection system allows you to upload a digital image or video via an app or website, which is then analyzed on several levels. The material is first checked for forgery using artificial intelligence and then it is determined whether it is a deepfake. Finally, the user maintains a graph visualizing all the manipulations.
Another artificial intelligence-based solution is Sensitivity. The program can be used to detect face swapping by capturing manipulated audio and AI-generated images. Sensity is recommended as a quick and effective tool for verifying the authenticity of digital interactions with another person, and also used as part of the KYC, or Know Your Customer, procedure.
Recommended programs also include: We checkedBut in this case, we’re talking about challenges that require advanced content verification methods. The project focuses on analyzing and contextualizing social media and website content within the broader online ecosystem in order to expose fabricated content. The system detects deepfakes using a blockchain database of known fakes, powered by open-source algorithms and machine learning. WeVerify Deepfake Detector can be used for both photos and videos.
Tools from Google, Intel and Microsoft
Some of the deepfake detection tools available on the market are owned by small IT companies that focus on this specific part of internet culture. However, big players also offer their solutions. Big Tech is not blind to the problem of fabricated photos and videos, even if (as the Meta example shows) in some cases it likes to ignore it. What do companies like Google and Microsoft have to offer?
Read also: Former Google and OpenAI employees have revealed the truth. They were afraid of retaliation for opposing artificial intelligence
Google DeepMind, an AI research lab founded by an Alphabet company, launched last year SynthID tool to mark a water footprint invisible to the human eye in photos and audio recordings created with artificial intelligence. Software testing has been ongoing since last year, and in May Google announced that it will now also be possible to watermark texts created by artificial intelligence. Importantly, SynthID will allow not only to watermark your own work, but also to identify photos, recordings and texts processed in this way. The tool therefore aims to promote the responsible and open use of AI.
The Redmont giant is also contributing to the detection of artificially processed content. Microsoft Video Authenticator serves to combat misinformation by carefully analyzing photos or videos for changes made by AI that are invisible to the naked eye. The program analyzes the image presented and then presents the user with a percentage chance that manipulation has occurred.. In the case of videos, the material is analyzed in real time, and the mentioned result applies to every subsequent second of its duration. Microsoft also boasts that Video Authenticator was built on publicly available data and was trained on the DeepFake Detection Challenge dataset.
Intel has also created its own tool for detecting fake videos. FakeCatcher combines AI-based facial and landmark detection algorithms with a suite of real-time image and video analysis tools. The entire system is powered by the 3rd generation Intel Xeon Scalable processor, which can simultaneously manage up to 72 separate detection streams. Unlike other programs of its kind, Intel FakeCatcher does not look for signs of inauthenticity by analyzing raw data. Instead, it focuses on defining authentic human tags in real videos. This means, for example, it identifies changes in the pixels of the video that indicate blood flow in the veins of the face, and its algorithm transforms these changes into spatiotemporal maps. Then, using deep learning methods, FakeCatcher recognizes whether the video is real or fake.
The best deepfake detection programs:
- Sentinel,
- Intel FakeCatcher
- Sensitivity
- Google Synthetic ID
- Microsoft Video Authenticator Tool
- We checked
The biggest weaknesses of AI-based programs
It’s easy to see that the vast majority of tools available for detecting deepfakes are based on solutions supported by artificial intelligence. So we have a situation where we use AI to fight AI. However, it is worth remembering that generative artificial intelligence may have problems detecting manipulations that were not created with this type of software..
In the past, AI has repeatedly made mistakes in analyzing deepfakes when dealing with cropped image fragments or compressed files. This is because AI training often does not cover all methods of metadata removal. The same goes for low-quality fake audio files, which deepfake detection tools often flag as real. Even a method as simple as taking a screenshot of an AI-generated photo deprives the image of the metadata needed to identify it as fake. It is therefore not surprising that many of the aforementioned programs are still in the testing and continuous improvement phase.