Unfake the Face: Tackling AI Disinformation
Did you know that estimates suggest that 90% of online content will either be synthetically generated or altered?
AI Disinformation has shifted in the recent past from easily distinguishable images to deceptive deepfakes that could fool even the keenest eye. The fourth edition of the DSA AI Webinar series tackled deep fakes and demystified this new scourge.
Previously, spreading misinformation required a small level of technical skill. However, the introduction of artificial intelligence and generative AI tools has democratised the process, making it quicker and easier for bad actors to spread hyperrealistic content.
Anxieties and fears around AI are real – during the session, participants noted that their confidence in spotting deepfakes is low overall. This points to a deep need for strong practical detection tools, not just the theory of how to do it. Participants also cited disinformation as the greatest risk that AI poses to them currently, but privacy and data breaches were also key items of concern.
AI disinformation is no longer a distant risk; it is an issue that is permeating our news rooms, CSOs, offices and lives. Because bad actors have the same access to tools as those using them legitimately, it is important that online users remain vigilant.
When it comes to identifying AI generated content, we can no longer rely on simply using our eyes. However, there are some strategies that you could use to assist you.
- Pay Attention to Small Details – AI generated photos tend to have small inconsistencies, particularly in the background of photos/videos, or small areas such as necklaces or odd wording. Paying attention to context can also help you decide if something is likely to be AI or not. Unfortunately, we can no longer say that this feels spiritually AI
- Ask if the Logic of the Image Makes Sense – AI generated content can go overboard with the details. There may be too many legs or arms, or just illogical details in the content. Shadows and lighting are also good tells, AI still gets physics wrong.
- Ask if the Mouth Movements and Voice Match Up – typically, AI generated videos have inconsistent audio and video. In the case of deepfakes, the voice of the speaker may not be a 1 to 1 match of the mouth movements. Deepfakes also miss the natural facial expressions of regular human speech. It may also use the standard AI monotone.
- Use Online Tools Cautiously – AI detection tools can help with identifying AI generated content, however, some tools are actually used to secretly train more data. Ensure that you effectively read the terms and conditions of the tool.