The problem of detecting deepfakes and manipulated media is not as straightforward as one might think. While there have been advancements in AI technology to detect such content, there are still major hurdles to overcome. One of the primary challenges is that these detection models were initially trained on high-quality media, which is not representative of the content produced in many parts of the world, especially in countries in the Global South. For instance, cheap Chinese smartphone brands that dominate the market in Africa produce photos and videos of much lower quality, making it difficult for detection models to accurately identify manipulated content.
The sensitivity of these detection models is also a cause for concern. Background noise in audio or compressing a video for social media can lead to false positives or negatives, which are common in real-world scenarios. This poses a significant problem for journalists, fact-checkers, and civil society members who rely on free, public-facing tools for deepfake detection. These tools are not only inaccurate but also fail to address the inequity of representation in training data and the challenges posed by lower quality media.
Apart from generative AI, cheapfakes or manipulated media that involve simple alterations like adding misleading labels or editing audio and video are prevalent in the Global South. However, these cheapfakes can be mistakenly flagged as AI-manipulated by faulty models or untrained researchers. This misclassification could have serious repercussions on a policy level, potentially leading legislators to take action based on inaccurate information. Inflating the numbers of AI-generated content could result in unnecessary crackdowns on content creators.
Developing new tools for detecting deepfakes is not a simple task. Access to energy and data centers is essential for building, testing, and running detection models, which are not readily available in many parts of the world. Without local alternatives, researchers in countries like Ghana are left with limited options: pay for expensive off-the-shelf tools, use inaccurate free tools, or rely on academic institutions for access. The lack of computational resources hinders the development of localized solutions for deepfake detection.
Moreover, the process of verifying and detecting deepfakes can be time-consuming, especially when relying on external resources. The lag time between submitting content for verification and receiving results can be significant, resulting in delays in addressing false information. By the time a definitive conclusion is reached, the damage caused by manipulated media may have already spread, making it challenging to contain the impact.
While there is a growing emphasis on developing advanced detection models for deepfakes, some experts argue that this focus might divert attention and resources away from other critical aspects of promoting information integrity. Directing funding towards news outlets and civil society organizations that foster public trust can be more effective in creating a resilient information ecosystem. Yet, there is a concern that the funding allocation does not prioritize these essential entities that play a crucial role in combatting misinformation.
The challenges of deepfake detection and manipulated media in the Global South highlight the complexities involved in combating digital misinformation. Addressing issues such as the quality of training data, sensitivity of detection models, access to computational resources, and the timeliness of verification are critical for developing effective solutions. By acknowledging these challenges and working towards collaborative efforts between researchers, journalists, and policymakers, we can strive towards a more transparent and accountable digital information landscape.
Leave a Reply